Top Banner
Missing in Action: Teacher and Health Worker Absence in Developing Countries Nazmul Chaudhury, Jeffrey Hammer, Michael Kremer, Karthik Muralidharan and F. Halsey Rogers I n this paper, we report results from surveys in which enumerators made unannounced visits to primary schools and health clinics in Bangladesh, Ecuador, India, Indonesia, Peru and Uganda and recorded whether they found teachers and health workers in the facilities. 1 Averaging across the countries, about 19 percent of teachers and 35 percent of health workers were absent, as shown in Table 1. The survey focused on whether providers were present in their facilities, but since many providers who were at their facilities were not working, even these figures may present too favorable a picture. For example, in India, one-quarter of government primary school teachers were absent from school, but only about one-half of the teachers were actually teaching when enumerators arrived at the schools. We find that absence rates are generally higher in poorer regions. Absence is typically fairly widespread, rather than being concentrated on a small number of 1 A number of researchers have examined the problem of absence among education and health providers in recent years (Alca ´zar and Andrade, 2001; Banerjee, Deaton and Duflo, 2004; Begum and Sen, 1997; Chaudhury and Hammer, 2003; Das, Dercon, Habyarimana and Krishnan, 2005; Glewwe, Kremer and Moulin, 1999; King, Orazem and Paterno, 1999; Kingdon and Muzammil, 2001; Pandey, 2005; Pratichi Education Team, 2002; PROBE Team, 1999; Sen, 1997; World Bank 2003; 2004). This paper measures teacher and health worker absence in nearly nationally representative samples in several countries using a common methodology based on direct observations during unannounced visits. y Nazmul Chaudhury is Economist, South Asia Human Development, World Bank, Wash- ington, D.C. Jeffrey Hammer is Lead Economist, South Asia Social Development, World Bank, New Delhi, India. Michael Kremer is Gates Professor of Developing Societies, Harvard University, Cambridge, Massachusetts. Karthik Muralidharan is a graduate student in economics, Harvard University, Cambridge, Massachusetts. F. Halsey Rogers is Senior Economist, Development Research Group, World Bank, Washington, D.C. Journal of Economic Perspectives—Volume 20, Number 1—Winter 2006 —Pages 91–116
30

Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

May 28, 2018

Download

Documents

dodat
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Page 1: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

Missing in Action Teacher and HealthWorker Absence in DevelopingCountries

Nazmul Chaudhury Jeffrey HammerMichael Kremer Karthik Muralidharan andF Halsey Rogers

I n this paper we report results from surveys in which enumerators madeunannounced visits to primary schools and health clinics in BangladeshEcuador India Indonesia Peru and Uganda and recorded whether they

found teachers and health workers in the facilities1 Averaging across the countriesabout 19 percent of teachers and 35 percent of health workers were absent asshown in Table 1 The survey focused on whether providers were present in theirfacilities but since many providers who were at their facilities were not workingeven these figures may present too favorable a picture For example in Indiaone-quarter of government primary school teachers were absent from school butonly about one-half of the teachers were actually teaching when enumeratorsarrived at the schools

We find that absence rates are generally higher in poorer regions Absence istypically fairly widespread rather than being concentrated on a small number of

1 A number of researchers have examined the problem of absence among education and healthproviders in recent years (Alcazar and Andrade 2001 Banerjee Deaton and Duflo 2004 Begum andSen 1997 Chaudhury and Hammer 2003 Das Dercon Habyarimana and Krishnan 2005 GlewweKremer and Moulin 1999 King Orazem and Paterno 1999 Kingdon and Muzammil 2001 Pandey2005 Pratichi Education Team 2002 PROBE Team 1999 Sen 1997 World Bank 2003 2004) Thispaper measures teacher and health worker absence in nearly nationally representative samples in severalcountries using a common methodology based on direct observations during unannounced visits

y Nazmul Chaudhury is Economist South Asia Human Development World Bank Wash-ington DC Jeffrey Hammer is Lead Economist South Asia Social Development World BankNew Delhi India Michael Kremer is Gates Professor of Developing Societies HarvardUniversity Cambridge Massachusetts Karthik Muralidharan is a graduate student ineconomics Harvard University Cambridge Massachusetts F Halsey Rogers is SeniorEconomist Development Research Group World Bank Washington DC

Journal of Economic PerspectivesmdashVolume 20 Number 1mdashWinter 2006mdashPages 91ndash116

ldquoghostrdquo workers Higher-ranking and more powerful providers such as headmastersand doctors are absent more often than lower-ranking ones for example averag-ing across countries 39 percent of doctors were absent while only 31 percent ofother health workers were absent Men are absent more often than womenTeachers from the local area are absent less often There is little evidence that paystrongly affects absence (at least in the range of pay where we have data) bycontrast we do find evidence suggesting a role for the quality of infrastructure at thefacility This finding is consistent with the idea that teachers and health workers areextremely unlikely to be fired for absence but that their decisions about whether to goto work are influenced by the working conditions they face Contract teachers who arenot subject to civil service protection and earn a fraction of what civil service teachersearn do not have lower absence rates In India we also examine absence rates amongteachers in rural private schools and in locally managed nonformal education centersAbsence rates are high among these teachers as well although private school teachershave lower absence than public teachers in the same village

Much recent discussion of economic development revolves around the role ofinstitutions Much of this discussion focuses on property rights institutions but italso seems possible that weak institutions for supplying public goodsmdasheducationand health in particularmdashare a significant barrier to economic development inmany countries

Background on Education and Health Care Systems in DevelopingCountries

In many developing countries including those in our survey education andhealth systems are bifurcated with a highly centralized and formalized government

Table 1Provider Absence Rates by Country and Sector

Absence rates () in

Primary schools Primary health centers

Bangladesh 16 35Ecuador 14 mdashIndia 25 40Indonesia 19 40Peru 11 25Uganda 27 37Unweighted average 19 35

Notes Providers were counted as absent if they could not be found in the facility for any reason at thetime of a random unannounced spot check (see text for further detail) In Uganda the sampled districtswere divided into subcounties and schools in subcounties with level III health centers comprise theschool sampling frame This sampling strategy may have had the effect of understated slightly thenational absence rate there given that schools in more rural areas appear to have higher absence rates

92 Journal of Economic Perspectives

system coexisting with a range of less formal arrangements Hiring and financingdecisions in the formal systems are made by the national government or in Indiaby state governments responsible to millions of people Teachers and healthworkers are typically unionized and their unions are strong and politically influ-ential Teachers in low-income countries earn about four times GDP per capitawhile their counterparts in rich countries earn only about two times per capita GDP(Bruns Mingat and Rakotomalala 2005) This is in part because teachers are moreeducated relative to the typical member of the labor force in poor countries butthe long queues of qualified people waiting to be hired as teachers in manydeveloping countries suggest teachers also receive greater premia over marketwages The vast bulk of education budgets and a large share of health budgets goto pay salaries and expenditure on nonsalary inputs is widely seen as inefficientlylow (Pritchett and Filmer 1999)

Hiring salaries and promotion are determined largely by educational qualifi-cations and seniority with less scope for performance-based pay than in developedcountries General practitioners for example are typically paid a straight salary indeveloping countriesmdashunlike in developed ones like the United Kingdom wheregeneral practitioners in the National Health Service are typically paid based on thenumber of patients who sign up for their practice Whereas many teachers indeveloped countries could aspire to become head teachers and education admin-istrators these promotion opportunities are cut off for many developing countryteachers because they lack the necessary educational qualifications

Wages under national civil-service systems are typically not fully responsive tolocal labor market conditions nor to individual characteristics and are often com-pressed relative to those in the private sector Many teachers receive substantialrents in the form of wages that are higher than their outside options (as evidencedby the long queues of applicants for government teaching positions) However itis likely that skilled medical personnelmdashdoctors in particularmdashearn much smallerrents and it is possible that if they were present in their clinics as frequently asstipulated as their official contracts they would be much worse off than underalternative market opportunities and would quit the public system entirely

While official rules provide for the possibility of punitive action in the case ofrepeated absence disciplinary action for absences are rare Teachers and healthworkers are almost never fired Despite Indiarsquos 25 percent teacher-absence rateonly one head teacher in our sample of nearly 3000 Indian government-runschools reported a case in which a teacher was fired for repeated absence The mainform of sanctions for teachers would be a transfer to an undesirable location butless than 1 percent of head teachers (18 out of nearly 3000) report having gottenteachers transferred for repeated absence

Given the rarity of disciplinary action for repeated absence the mystery foreconomists may not be why absence from work is so high but why anyone shows upat all For many providers the answer must be that important intrinsic andnonpecuniary motivationsmdashsuch as professional pride or concern for the regard ofpeersmdashaffect attendance decisions In Peru for example an average of 89 percent

Nazmul Chaudhury et al 93

of teachers show up each day despite an apparent lack of significant rewards orpunishments related to teacher performance (Alcazar et al 2005)

Against the background of these highly formalized and bureaucratized sys-tems a plethora of informal systems have grown up virtually outside the ambit ofregulation These include private schools and clinics that are not recognized by thegovernment publicly supported community-managed schools such as nonformaleducation centers in India and systems for hiring contract teachers at publicschools outside of normal civil service rules Teachers in these informal systemsoften have lower educational qualifications than their civil service counterpartsearn much less (often only a third as much or lower) and have little or no jobsecurity Hiring and salary decisions are subject to more discretion with lessemphasis on formal educational qualifications There are also a range of healthproviders outside of formal government systems including many nonlicensedproviders without medical education as well as government providers operatingprivate practices on the side

We conducted a survey focused on the presence of teachers and health workersat public primary schools and primary health centers to assess what would seem toconstitute a minimal prima facie condition for efficacy of these systems Surveyswere typically close to nationally representative but excluded some areas from thesampling frame for security or logistical reasons2 In rural India enumerators alsocollected data from private schools and nonformal education centers located in thesame village as public schools and in Indonesia they also collected data fromprivate schools As we discuss below absence rates are high in the informal sectoras well as the formal sector

Our absence data are based on direct physical verification of the providerrsquospresence rather than attendance logbooks or interviews with the facility head InBangladesh Ecuador Indonesia Peru and Uganda enumerators made two visitsmdashtypically several months apartmdashto each of about ten randomly chosen health carecenters and ten randomly chosen public schools in each of ten randomly chosendistricts On average we visited 100 schools and 100 health care centers in eachcountry With around eight providers in the average facility and two observationson each of these providers we had an average of over 1500 observations on teacherattendance in each country and an average of over 1350 observations for healthworker attendance in each country In India the survey was designed to berepresentative in each of 20 states which together account for 98 percent of Indiarsquospopulation Three unannounced visits were made to each of about 3000 publicschools over a span of three to four months Since the average school in our samplehas around four teachers we have nearly 35000 observations on teacher atten-dance Similarly enumerators made three unannounced visits to over 1350 publicclinics and since these had an average of eight or nine health workers each wehave approximately 32500 observations on health worker presence The majority

2 In Indonesia the excluded provinces account for only about 8 percent of the countryrsquos population inother countries even less

94 Journal of Economic Perspectives

of the field work in all countries was carried out between October 2002 and April2003

A worker was counted as absent if at the time of a random visit during facilityhours he or she was not in the school or health center The enumerators for thesurvey took several measures to ensure that the rate of absence would not beoverestimated The list of employees used for checking attendance was created atthe facility itself based on staff lists and schedule information provided by thefacility director or other principal respondent Enumerators then checked theattendance only of those who were ordinarily supposed to be on duty at the time ofthe visit3 We omitted from the absence calculations all employees who werereported by the director as being on another shift whether or not this could beverified Only full-time employees were included in our analysis to minimize therisk that shift workers would be counted as absent when they were not supposed tobe on duty Measured absences in education were slightly lower in later surveyrounds consistent with the hypothesis that awareness of the first round of thesurvey created a bit of a ldquowarning effectrdquo regarding the presence of the surveyteams Adjusting for survey round and time-of-day effects would increase theestimated teacher absence by 1ndash2 percentage points (Kremer et al 2004) Nosimilar effect was found in health

We do not think that the absence rate is overstated because health workerswere working outside the facility At the beginning of the facility interview theenumerator asked to see the schedule of all health workers Only those assigned towork at the clinic on the day of the interview (as opposed for example to beingassigned to a subclinic for that day) were included in the sample Moreover we didnot find that health workers whose schedules include outreach or field work areabsent more than those who are always supposed to be in the clinic such aspharmacists A recent detailed study in Rajasthan which found absence ratessimilar to those we report made efforts to track down nurses who were absent fromhealth subcenters and found that only in 12 percent of cases of absence was thenurse in one of the villages served by her subcenter (Banerjee Deaton and Duflo2004)

High Absence Rates

At 19 percent and 35 percent respectively absence rates among teachers andhealth care workers in developing countries are high relative to those of both theircounterparts in developed countries and other workers in developing countriesStrictly comparable numbers are not available for the United States but adminis-trative data from a large sample of school districts in New York state in themid-1980s revealed a mean absence rate of 5 percent (Ehrenberg Rees and

3 This included employees who might have been on authorized leave that day although as we arguebelow reports of leave were often not credible

Missing in Action Teacher and Health Worker Absence in Developing Countries 95

Ehrenberg 1991) Even among Indian factory workers who enjoy a high degree ofjob security due to rigid labor laws reported absence rates are only around 105percent (Ministry of Labor Industry Survey 2000ndash2001) much lower than the 25and 40 percent rates of absence among Indian teachers and medical personnelrespectively

The welfare consequence of teacher and health worker absence may be evengreater in the countries that we surveyed than they would be in developed coun-tries In low-income countries substitutes rarely replace absent teachers and sostudents simply mill around go home or join another class often of a differentgrade Small schools and clinics are common in rural areas of developing countriesand these may be closed entirely as a result of provider absence In nearly12 percent of the visits enumerators in India encountered schools that were closedbecause no teacher was present An estimate of the effect of teacher absence onstudent outcomes is provided by Duflo and Hanna (2005) who show that arandomized intervention that reduced teacher absence from 36 to 18 percent ledto a 017 standard deviation improvement in student test scores

As noted in the introduction many teachers and health workers who are intheir facilities are not working Across Indian government-run schools we find thatonly 45 percent of teachers assigned to a school are engaged in teaching activity atany given point in timemdasheven though teaching activity was defined very broadly toinclude even cases where the teacher was simply keeping class in order and noactual teaching was taking place According to the official schedules teachersshould be teaching most of the time when school is in session Fewer than30 percent of schools in the sample had more teachers than classes and the schoolschedule is therefore typically designed so that teachers and students have breaksat the same time rather than with teachers having certain periods off to prepareas in most schools in developed countries Assuming that the number of teacherswho should officially be teaching is equal to the minimum of the number of classesand the number of teachers4 only 50 percent of teachers in Indian public schoolswho should be teaching at a given point are in fact doing so

In assessing these activity numbers itrsquos worth bearing in mind that they couldpotentially have been affected by the presence of the surveyor On the one handenumerators report that teachers sometimes started teaching when the surveyorarrived On the other hand although the enumerators were instructed to look fora respondent who was not teaching to ask questions regarding the school (andtypically they found the headmaster or other teacher in the office) the survey itselfmay have diverted teachers from teaching in some cases But even if we excludethose teachers from the calculation whose activity was recorded as ldquotalking to theenumeratorrdquo only 55 percent of those teachers who should have been teachingwere doing so

4 So if a school had four classes and three teachers we would expect three teachers to be teachingwhereas if it had five teachers and four classes we would only expect four teachers to be teaching

96 Journal of Economic Perspectives

Absence Across Sectors and Countries

Two clear generalizations emerge from the cross-country cross-sector data onabsence and from the variation across Indian states First health care providers aremuch more likely to be absent than teachers As Table 1 shows averaging acrosscountries for which we have data on absence for both types of providers health careworkers are 15 percentage points more likely to be absent than are teachers Thisdifference may arise because health care workers have more opportunities tomoonlight at other jobs or because health care workers receive smaller rentsrelative to what they would earn in the private sector or because health careworkers are harder to monitor If a teacher does not show up regularly a class fullof pupils and potentially their parents will know about it On the other hand it ismuch harder for patients who presumably come to health care centers irregularlyto know if a particular health care worker is absent frequently

Second higher-income areas have lower absence rates Figure 1 shows theabsence-income relationship for the sample countries other than India (repre-sented by triangles and labeled) and for the Indian states in our sample (repre-sented by circles) The left-hand panel shows the relationship among teachers theright-hand panel among health-care workers Combining the two sectors acrosscountries and Indian states an ordinary least squares regression of absence on logof per capita GDP (measured in purchasing power parity terms) and a dummy forsector (health or education) suggests that doubling of per capita income is asso-ciated with 60 percentage points lower absence The coefficient on per capitaincome is significant at the 1 percent level and the income and sector variablestogether account for more than half of the variation in sector-country and sector-state absence rates When we run two separate regressions one for the countriesand one for the Indian states we obtain very similar coefficients on log income Inthe cross-country regression doubling income is associated with a 58 percentage-point decline in absence and in the Indian cross-state regression a 48 percentage-point drop

However the relationship between a countryrsquos per capita income and absenceis stronger in education than in health Among teachers doubling income isassociated with an 80 percentage-point absence decline (significant at the01 percent level) compared with only a 38 percentage point decline in healthworker absence (falling short of significance at even the 10 percent level)5

Again a very similar pattern holds in the cross-country and the Indian cross-state regressions

One possible explanation for the correlation between income and absence isthat exogenous variation in institutional quality in service provision drives human

5 The absence-income relationship in the health sector appears to hold more strongly for doctors thanfor other medical personnel Within India regressing doctor absence on state per capita income yieldsa much larger coefficient (in absolute value) significant at the 10 percent level whereas the coefficientis small and insignificant for health workers as a group

Nazmul Chaudhury et al 97

capital acquisition and thus income Another is that the overall level of develop-ment drives the quality of education and health delivery While it is impossible todisentangle these stories completely to the extent that the overall level of devel-opment influences provider absence one might expect low income levels to lead tohigh absence rates in both education and health On the other hand if educationis particularly important for human capital acquisition and thus income whilemedical clinics have a larger consumption component then exogenous variation inquality of education systems will lead to variation in income while the quality ofhealth care systems will be less correlated with income This pattern matches whatwe see in the data

It is intriguing that the relationship between income and absence is so similaracross countries and across Indian states and that it is so tight in each case Whilesalaries typically rise with GDP (although not proportionally) teacher salariesacross Indian states are relatively flat6 Thus across the states of India salaries forteachers and health workers in poor states are considerably higher relative to thecost of living and relative to workersrsquo outside opportunities than are salaries in richstates Nonetheless absence rates are higher in poor states The similarity betweenthe absence-income regression line across countries and the comparable line acrossIndian states despite the difference in the relationship between income andsalaries in the two samples suggests a limited role for salaries in influencing

6 Ministry of Human Resource Development India

Figure 1Absence Rate versus NationalState Per Capita Income

Source Authorsrsquo calculationsNote BNG Bangladesh ECU Ecuador IDN Indonesia PER Peru UGA Uganda Indiarsquosnational averages are excluded due to the inclusion of the Indian states For Indian states incomesare the official per capita net state domestic products

98 Journal of Economic Perspectives

absence over the existing salary range Of course it is important to bear in mindthat the samples of countries and states are very small and other factors couldinfluence these slopes

Teacher and health worker absence are correlated across countries and stateseven after controlling for per capita income The residuals from the two regressionsdepicted in Figure 1 (with an additional dummy added for Indian states) are highlycorrelated with each other with a correlation coefficient of 044 (significant at the5 percent level) This correlation could potentially be due to mismeasurement ofincome but it could also reflect spillover effects in social norms across sectors oromitted variables such as the quality of governance

Concentration of Absence

To understand and potentially design policies to counter high absence ratesit is useful to know whether absences are spread out among providers or concen-trated among a small number of ldquoghost workersrdquo who are on the books but nevershow up Since our survey included only two or three observations per worker wewould observe some dispersion in absence rates even if all workers had identicalunderlying probabilities of being absent The left panel of Table 2 shows thedistribution of absence observed in the data For comparison the right panel showsthe distribution that would be observed if the probability of absence in each visitwere equal to the estimated absence rate in the specific country-sector combina-tion so all workers had the same probability of being absent For example if allteachers in Indonesia had a 019 chance of being absent (which is the averageteacher absence rate there) then on any two independent visits we would expect36 percent (019 019) to be absent both times 656 percent (081 081) to bepresent both times and the remaining 308 percent to be absent once On the otherhand if absence were completely concentrated in certain providers we wouldobserve that 19 percent of the teachers are always absent 81 percent are alwayspresent and none are absent only once

Clearly the data match neither the extreme of all workers having identicalunderlying probabilities of absence nor of all absence being due to ghost workersbut an eyeball test suggests that absence appears to be fairly widespread with theempirical distribution surprisingly close to that predicted by a model with identicalabsence probabilities Teachers in Ecuador are an exception and appear to be theleading candidates for a ldquoghost workerrdquo explanation with a very high percentage ofteachers being present in both visits and more teachers absent in both visits than inone of the two visits

The exercise above while suggestive can technically only be used to test theextreme hypotheses of complete concentration of absence and perfectly identicalabsence rates among workers Glewwe Ilias and Kremer (2004) assume providersrsquounderlying probability of absence follows a beta distribution and estimate thisdistribution in two districts of Kenya using a maximum likelihood approach They

Missing in Action Teacher and Health Worker Absence in Developing Countries 99

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 2: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

ldquoghostrdquo workers Higher-ranking and more powerful providers such as headmastersand doctors are absent more often than lower-ranking ones for example averag-ing across countries 39 percent of doctors were absent while only 31 percent ofother health workers were absent Men are absent more often than womenTeachers from the local area are absent less often There is little evidence that paystrongly affects absence (at least in the range of pay where we have data) bycontrast we do find evidence suggesting a role for the quality of infrastructure at thefacility This finding is consistent with the idea that teachers and health workers areextremely unlikely to be fired for absence but that their decisions about whether to goto work are influenced by the working conditions they face Contract teachers who arenot subject to civil service protection and earn a fraction of what civil service teachersearn do not have lower absence rates In India we also examine absence rates amongteachers in rural private schools and in locally managed nonformal education centersAbsence rates are high among these teachers as well although private school teachershave lower absence than public teachers in the same village

Much recent discussion of economic development revolves around the role ofinstitutions Much of this discussion focuses on property rights institutions but italso seems possible that weak institutions for supplying public goodsmdasheducationand health in particularmdashare a significant barrier to economic development inmany countries

Background on Education and Health Care Systems in DevelopingCountries

In many developing countries including those in our survey education andhealth systems are bifurcated with a highly centralized and formalized government

Table 1Provider Absence Rates by Country and Sector

Absence rates () in

Primary schools Primary health centers

Bangladesh 16 35Ecuador 14 mdashIndia 25 40Indonesia 19 40Peru 11 25Uganda 27 37Unweighted average 19 35

Notes Providers were counted as absent if they could not be found in the facility for any reason at thetime of a random unannounced spot check (see text for further detail) In Uganda the sampled districtswere divided into subcounties and schools in subcounties with level III health centers comprise theschool sampling frame This sampling strategy may have had the effect of understated slightly thenational absence rate there given that schools in more rural areas appear to have higher absence rates

92 Journal of Economic Perspectives

system coexisting with a range of less formal arrangements Hiring and financingdecisions in the formal systems are made by the national government or in Indiaby state governments responsible to millions of people Teachers and healthworkers are typically unionized and their unions are strong and politically influ-ential Teachers in low-income countries earn about four times GDP per capitawhile their counterparts in rich countries earn only about two times per capita GDP(Bruns Mingat and Rakotomalala 2005) This is in part because teachers are moreeducated relative to the typical member of the labor force in poor countries butthe long queues of qualified people waiting to be hired as teachers in manydeveloping countries suggest teachers also receive greater premia over marketwages The vast bulk of education budgets and a large share of health budgets goto pay salaries and expenditure on nonsalary inputs is widely seen as inefficientlylow (Pritchett and Filmer 1999)

Hiring salaries and promotion are determined largely by educational qualifi-cations and seniority with less scope for performance-based pay than in developedcountries General practitioners for example are typically paid a straight salary indeveloping countriesmdashunlike in developed ones like the United Kingdom wheregeneral practitioners in the National Health Service are typically paid based on thenumber of patients who sign up for their practice Whereas many teachers indeveloped countries could aspire to become head teachers and education admin-istrators these promotion opportunities are cut off for many developing countryteachers because they lack the necessary educational qualifications

Wages under national civil-service systems are typically not fully responsive tolocal labor market conditions nor to individual characteristics and are often com-pressed relative to those in the private sector Many teachers receive substantialrents in the form of wages that are higher than their outside options (as evidencedby the long queues of applicants for government teaching positions) However itis likely that skilled medical personnelmdashdoctors in particularmdashearn much smallerrents and it is possible that if they were present in their clinics as frequently asstipulated as their official contracts they would be much worse off than underalternative market opportunities and would quit the public system entirely

While official rules provide for the possibility of punitive action in the case ofrepeated absence disciplinary action for absences are rare Teachers and healthworkers are almost never fired Despite Indiarsquos 25 percent teacher-absence rateonly one head teacher in our sample of nearly 3000 Indian government-runschools reported a case in which a teacher was fired for repeated absence The mainform of sanctions for teachers would be a transfer to an undesirable location butless than 1 percent of head teachers (18 out of nearly 3000) report having gottenteachers transferred for repeated absence

Given the rarity of disciplinary action for repeated absence the mystery foreconomists may not be why absence from work is so high but why anyone shows upat all For many providers the answer must be that important intrinsic andnonpecuniary motivationsmdashsuch as professional pride or concern for the regard ofpeersmdashaffect attendance decisions In Peru for example an average of 89 percent

Nazmul Chaudhury et al 93

of teachers show up each day despite an apparent lack of significant rewards orpunishments related to teacher performance (Alcazar et al 2005)

Against the background of these highly formalized and bureaucratized sys-tems a plethora of informal systems have grown up virtually outside the ambit ofregulation These include private schools and clinics that are not recognized by thegovernment publicly supported community-managed schools such as nonformaleducation centers in India and systems for hiring contract teachers at publicschools outside of normal civil service rules Teachers in these informal systemsoften have lower educational qualifications than their civil service counterpartsearn much less (often only a third as much or lower) and have little or no jobsecurity Hiring and salary decisions are subject to more discretion with lessemphasis on formal educational qualifications There are also a range of healthproviders outside of formal government systems including many nonlicensedproviders without medical education as well as government providers operatingprivate practices on the side

We conducted a survey focused on the presence of teachers and health workersat public primary schools and primary health centers to assess what would seem toconstitute a minimal prima facie condition for efficacy of these systems Surveyswere typically close to nationally representative but excluded some areas from thesampling frame for security or logistical reasons2 In rural India enumerators alsocollected data from private schools and nonformal education centers located in thesame village as public schools and in Indonesia they also collected data fromprivate schools As we discuss below absence rates are high in the informal sectoras well as the formal sector

Our absence data are based on direct physical verification of the providerrsquospresence rather than attendance logbooks or interviews with the facility head InBangladesh Ecuador Indonesia Peru and Uganda enumerators made two visitsmdashtypically several months apartmdashto each of about ten randomly chosen health carecenters and ten randomly chosen public schools in each of ten randomly chosendistricts On average we visited 100 schools and 100 health care centers in eachcountry With around eight providers in the average facility and two observationson each of these providers we had an average of over 1500 observations on teacherattendance in each country and an average of over 1350 observations for healthworker attendance in each country In India the survey was designed to berepresentative in each of 20 states which together account for 98 percent of Indiarsquospopulation Three unannounced visits were made to each of about 3000 publicschools over a span of three to four months Since the average school in our samplehas around four teachers we have nearly 35000 observations on teacher atten-dance Similarly enumerators made three unannounced visits to over 1350 publicclinics and since these had an average of eight or nine health workers each wehave approximately 32500 observations on health worker presence The majority

2 In Indonesia the excluded provinces account for only about 8 percent of the countryrsquos population inother countries even less

94 Journal of Economic Perspectives

of the field work in all countries was carried out between October 2002 and April2003

A worker was counted as absent if at the time of a random visit during facilityhours he or she was not in the school or health center The enumerators for thesurvey took several measures to ensure that the rate of absence would not beoverestimated The list of employees used for checking attendance was created atthe facility itself based on staff lists and schedule information provided by thefacility director or other principal respondent Enumerators then checked theattendance only of those who were ordinarily supposed to be on duty at the time ofthe visit3 We omitted from the absence calculations all employees who werereported by the director as being on another shift whether or not this could beverified Only full-time employees were included in our analysis to minimize therisk that shift workers would be counted as absent when they were not supposed tobe on duty Measured absences in education were slightly lower in later surveyrounds consistent with the hypothesis that awareness of the first round of thesurvey created a bit of a ldquowarning effectrdquo regarding the presence of the surveyteams Adjusting for survey round and time-of-day effects would increase theestimated teacher absence by 1ndash2 percentage points (Kremer et al 2004) Nosimilar effect was found in health

We do not think that the absence rate is overstated because health workerswere working outside the facility At the beginning of the facility interview theenumerator asked to see the schedule of all health workers Only those assigned towork at the clinic on the day of the interview (as opposed for example to beingassigned to a subclinic for that day) were included in the sample Moreover we didnot find that health workers whose schedules include outreach or field work areabsent more than those who are always supposed to be in the clinic such aspharmacists A recent detailed study in Rajasthan which found absence ratessimilar to those we report made efforts to track down nurses who were absent fromhealth subcenters and found that only in 12 percent of cases of absence was thenurse in one of the villages served by her subcenter (Banerjee Deaton and Duflo2004)

High Absence Rates

At 19 percent and 35 percent respectively absence rates among teachers andhealth care workers in developing countries are high relative to those of both theircounterparts in developed countries and other workers in developing countriesStrictly comparable numbers are not available for the United States but adminis-trative data from a large sample of school districts in New York state in themid-1980s revealed a mean absence rate of 5 percent (Ehrenberg Rees and

3 This included employees who might have been on authorized leave that day although as we arguebelow reports of leave were often not credible

Missing in Action Teacher and Health Worker Absence in Developing Countries 95

Ehrenberg 1991) Even among Indian factory workers who enjoy a high degree ofjob security due to rigid labor laws reported absence rates are only around 105percent (Ministry of Labor Industry Survey 2000ndash2001) much lower than the 25and 40 percent rates of absence among Indian teachers and medical personnelrespectively

The welfare consequence of teacher and health worker absence may be evengreater in the countries that we surveyed than they would be in developed coun-tries In low-income countries substitutes rarely replace absent teachers and sostudents simply mill around go home or join another class often of a differentgrade Small schools and clinics are common in rural areas of developing countriesand these may be closed entirely as a result of provider absence In nearly12 percent of the visits enumerators in India encountered schools that were closedbecause no teacher was present An estimate of the effect of teacher absence onstudent outcomes is provided by Duflo and Hanna (2005) who show that arandomized intervention that reduced teacher absence from 36 to 18 percent ledto a 017 standard deviation improvement in student test scores

As noted in the introduction many teachers and health workers who are intheir facilities are not working Across Indian government-run schools we find thatonly 45 percent of teachers assigned to a school are engaged in teaching activity atany given point in timemdasheven though teaching activity was defined very broadly toinclude even cases where the teacher was simply keeping class in order and noactual teaching was taking place According to the official schedules teachersshould be teaching most of the time when school is in session Fewer than30 percent of schools in the sample had more teachers than classes and the schoolschedule is therefore typically designed so that teachers and students have breaksat the same time rather than with teachers having certain periods off to prepareas in most schools in developed countries Assuming that the number of teacherswho should officially be teaching is equal to the minimum of the number of classesand the number of teachers4 only 50 percent of teachers in Indian public schoolswho should be teaching at a given point are in fact doing so

In assessing these activity numbers itrsquos worth bearing in mind that they couldpotentially have been affected by the presence of the surveyor On the one handenumerators report that teachers sometimes started teaching when the surveyorarrived On the other hand although the enumerators were instructed to look fora respondent who was not teaching to ask questions regarding the school (andtypically they found the headmaster or other teacher in the office) the survey itselfmay have diverted teachers from teaching in some cases But even if we excludethose teachers from the calculation whose activity was recorded as ldquotalking to theenumeratorrdquo only 55 percent of those teachers who should have been teachingwere doing so

4 So if a school had four classes and three teachers we would expect three teachers to be teachingwhereas if it had five teachers and four classes we would only expect four teachers to be teaching

96 Journal of Economic Perspectives

Absence Across Sectors and Countries

Two clear generalizations emerge from the cross-country cross-sector data onabsence and from the variation across Indian states First health care providers aremuch more likely to be absent than teachers As Table 1 shows averaging acrosscountries for which we have data on absence for both types of providers health careworkers are 15 percentage points more likely to be absent than are teachers Thisdifference may arise because health care workers have more opportunities tomoonlight at other jobs or because health care workers receive smaller rentsrelative to what they would earn in the private sector or because health careworkers are harder to monitor If a teacher does not show up regularly a class fullof pupils and potentially their parents will know about it On the other hand it ismuch harder for patients who presumably come to health care centers irregularlyto know if a particular health care worker is absent frequently

Second higher-income areas have lower absence rates Figure 1 shows theabsence-income relationship for the sample countries other than India (repre-sented by triangles and labeled) and for the Indian states in our sample (repre-sented by circles) The left-hand panel shows the relationship among teachers theright-hand panel among health-care workers Combining the two sectors acrosscountries and Indian states an ordinary least squares regression of absence on logof per capita GDP (measured in purchasing power parity terms) and a dummy forsector (health or education) suggests that doubling of per capita income is asso-ciated with 60 percentage points lower absence The coefficient on per capitaincome is significant at the 1 percent level and the income and sector variablestogether account for more than half of the variation in sector-country and sector-state absence rates When we run two separate regressions one for the countriesand one for the Indian states we obtain very similar coefficients on log income Inthe cross-country regression doubling income is associated with a 58 percentage-point decline in absence and in the Indian cross-state regression a 48 percentage-point drop

However the relationship between a countryrsquos per capita income and absenceis stronger in education than in health Among teachers doubling income isassociated with an 80 percentage-point absence decline (significant at the01 percent level) compared with only a 38 percentage point decline in healthworker absence (falling short of significance at even the 10 percent level)5

Again a very similar pattern holds in the cross-country and the Indian cross-state regressions

One possible explanation for the correlation between income and absence isthat exogenous variation in institutional quality in service provision drives human

5 The absence-income relationship in the health sector appears to hold more strongly for doctors thanfor other medical personnel Within India regressing doctor absence on state per capita income yieldsa much larger coefficient (in absolute value) significant at the 10 percent level whereas the coefficientis small and insignificant for health workers as a group

Nazmul Chaudhury et al 97

capital acquisition and thus income Another is that the overall level of develop-ment drives the quality of education and health delivery While it is impossible todisentangle these stories completely to the extent that the overall level of devel-opment influences provider absence one might expect low income levels to lead tohigh absence rates in both education and health On the other hand if educationis particularly important for human capital acquisition and thus income whilemedical clinics have a larger consumption component then exogenous variation inquality of education systems will lead to variation in income while the quality ofhealth care systems will be less correlated with income This pattern matches whatwe see in the data

It is intriguing that the relationship between income and absence is so similaracross countries and across Indian states and that it is so tight in each case Whilesalaries typically rise with GDP (although not proportionally) teacher salariesacross Indian states are relatively flat6 Thus across the states of India salaries forteachers and health workers in poor states are considerably higher relative to thecost of living and relative to workersrsquo outside opportunities than are salaries in richstates Nonetheless absence rates are higher in poor states The similarity betweenthe absence-income regression line across countries and the comparable line acrossIndian states despite the difference in the relationship between income andsalaries in the two samples suggests a limited role for salaries in influencing

6 Ministry of Human Resource Development India

Figure 1Absence Rate versus NationalState Per Capita Income

Source Authorsrsquo calculationsNote BNG Bangladesh ECU Ecuador IDN Indonesia PER Peru UGA Uganda Indiarsquosnational averages are excluded due to the inclusion of the Indian states For Indian states incomesare the official per capita net state domestic products

98 Journal of Economic Perspectives

absence over the existing salary range Of course it is important to bear in mindthat the samples of countries and states are very small and other factors couldinfluence these slopes

Teacher and health worker absence are correlated across countries and stateseven after controlling for per capita income The residuals from the two regressionsdepicted in Figure 1 (with an additional dummy added for Indian states) are highlycorrelated with each other with a correlation coefficient of 044 (significant at the5 percent level) This correlation could potentially be due to mismeasurement ofincome but it could also reflect spillover effects in social norms across sectors oromitted variables such as the quality of governance

Concentration of Absence

To understand and potentially design policies to counter high absence ratesit is useful to know whether absences are spread out among providers or concen-trated among a small number of ldquoghost workersrdquo who are on the books but nevershow up Since our survey included only two or three observations per worker wewould observe some dispersion in absence rates even if all workers had identicalunderlying probabilities of being absent The left panel of Table 2 shows thedistribution of absence observed in the data For comparison the right panel showsthe distribution that would be observed if the probability of absence in each visitwere equal to the estimated absence rate in the specific country-sector combina-tion so all workers had the same probability of being absent For example if allteachers in Indonesia had a 019 chance of being absent (which is the averageteacher absence rate there) then on any two independent visits we would expect36 percent (019 019) to be absent both times 656 percent (081 081) to bepresent both times and the remaining 308 percent to be absent once On the otherhand if absence were completely concentrated in certain providers we wouldobserve that 19 percent of the teachers are always absent 81 percent are alwayspresent and none are absent only once

Clearly the data match neither the extreme of all workers having identicalunderlying probabilities of absence nor of all absence being due to ghost workersbut an eyeball test suggests that absence appears to be fairly widespread with theempirical distribution surprisingly close to that predicted by a model with identicalabsence probabilities Teachers in Ecuador are an exception and appear to be theleading candidates for a ldquoghost workerrdquo explanation with a very high percentage ofteachers being present in both visits and more teachers absent in both visits than inone of the two visits

The exercise above while suggestive can technically only be used to test theextreme hypotheses of complete concentration of absence and perfectly identicalabsence rates among workers Glewwe Ilias and Kremer (2004) assume providersrsquounderlying probability of absence follows a beta distribution and estimate thisdistribution in two districts of Kenya using a maximum likelihood approach They

Missing in Action Teacher and Health Worker Absence in Developing Countries 99

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 3: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

system coexisting with a range of less formal arrangements Hiring and financingdecisions in the formal systems are made by the national government or in Indiaby state governments responsible to millions of people Teachers and healthworkers are typically unionized and their unions are strong and politically influ-ential Teachers in low-income countries earn about four times GDP per capitawhile their counterparts in rich countries earn only about two times per capita GDP(Bruns Mingat and Rakotomalala 2005) This is in part because teachers are moreeducated relative to the typical member of the labor force in poor countries butthe long queues of qualified people waiting to be hired as teachers in manydeveloping countries suggest teachers also receive greater premia over marketwages The vast bulk of education budgets and a large share of health budgets goto pay salaries and expenditure on nonsalary inputs is widely seen as inefficientlylow (Pritchett and Filmer 1999)

Hiring salaries and promotion are determined largely by educational qualifi-cations and seniority with less scope for performance-based pay than in developedcountries General practitioners for example are typically paid a straight salary indeveloping countriesmdashunlike in developed ones like the United Kingdom wheregeneral practitioners in the National Health Service are typically paid based on thenumber of patients who sign up for their practice Whereas many teachers indeveloped countries could aspire to become head teachers and education admin-istrators these promotion opportunities are cut off for many developing countryteachers because they lack the necessary educational qualifications

Wages under national civil-service systems are typically not fully responsive tolocal labor market conditions nor to individual characteristics and are often com-pressed relative to those in the private sector Many teachers receive substantialrents in the form of wages that are higher than their outside options (as evidencedby the long queues of applicants for government teaching positions) However itis likely that skilled medical personnelmdashdoctors in particularmdashearn much smallerrents and it is possible that if they were present in their clinics as frequently asstipulated as their official contracts they would be much worse off than underalternative market opportunities and would quit the public system entirely

While official rules provide for the possibility of punitive action in the case ofrepeated absence disciplinary action for absences are rare Teachers and healthworkers are almost never fired Despite Indiarsquos 25 percent teacher-absence rateonly one head teacher in our sample of nearly 3000 Indian government-runschools reported a case in which a teacher was fired for repeated absence The mainform of sanctions for teachers would be a transfer to an undesirable location butless than 1 percent of head teachers (18 out of nearly 3000) report having gottenteachers transferred for repeated absence

Given the rarity of disciplinary action for repeated absence the mystery foreconomists may not be why absence from work is so high but why anyone shows upat all For many providers the answer must be that important intrinsic andnonpecuniary motivationsmdashsuch as professional pride or concern for the regard ofpeersmdashaffect attendance decisions In Peru for example an average of 89 percent

Nazmul Chaudhury et al 93

of teachers show up each day despite an apparent lack of significant rewards orpunishments related to teacher performance (Alcazar et al 2005)

Against the background of these highly formalized and bureaucratized sys-tems a plethora of informal systems have grown up virtually outside the ambit ofregulation These include private schools and clinics that are not recognized by thegovernment publicly supported community-managed schools such as nonformaleducation centers in India and systems for hiring contract teachers at publicschools outside of normal civil service rules Teachers in these informal systemsoften have lower educational qualifications than their civil service counterpartsearn much less (often only a third as much or lower) and have little or no jobsecurity Hiring and salary decisions are subject to more discretion with lessemphasis on formal educational qualifications There are also a range of healthproviders outside of formal government systems including many nonlicensedproviders without medical education as well as government providers operatingprivate practices on the side

We conducted a survey focused on the presence of teachers and health workersat public primary schools and primary health centers to assess what would seem toconstitute a minimal prima facie condition for efficacy of these systems Surveyswere typically close to nationally representative but excluded some areas from thesampling frame for security or logistical reasons2 In rural India enumerators alsocollected data from private schools and nonformal education centers located in thesame village as public schools and in Indonesia they also collected data fromprivate schools As we discuss below absence rates are high in the informal sectoras well as the formal sector

Our absence data are based on direct physical verification of the providerrsquospresence rather than attendance logbooks or interviews with the facility head InBangladesh Ecuador Indonesia Peru and Uganda enumerators made two visitsmdashtypically several months apartmdashto each of about ten randomly chosen health carecenters and ten randomly chosen public schools in each of ten randomly chosendistricts On average we visited 100 schools and 100 health care centers in eachcountry With around eight providers in the average facility and two observationson each of these providers we had an average of over 1500 observations on teacherattendance in each country and an average of over 1350 observations for healthworker attendance in each country In India the survey was designed to berepresentative in each of 20 states which together account for 98 percent of Indiarsquospopulation Three unannounced visits were made to each of about 3000 publicschools over a span of three to four months Since the average school in our samplehas around four teachers we have nearly 35000 observations on teacher atten-dance Similarly enumerators made three unannounced visits to over 1350 publicclinics and since these had an average of eight or nine health workers each wehave approximately 32500 observations on health worker presence The majority

2 In Indonesia the excluded provinces account for only about 8 percent of the countryrsquos population inother countries even less

94 Journal of Economic Perspectives

of the field work in all countries was carried out between October 2002 and April2003

A worker was counted as absent if at the time of a random visit during facilityhours he or she was not in the school or health center The enumerators for thesurvey took several measures to ensure that the rate of absence would not beoverestimated The list of employees used for checking attendance was created atthe facility itself based on staff lists and schedule information provided by thefacility director or other principal respondent Enumerators then checked theattendance only of those who were ordinarily supposed to be on duty at the time ofthe visit3 We omitted from the absence calculations all employees who werereported by the director as being on another shift whether or not this could beverified Only full-time employees were included in our analysis to minimize therisk that shift workers would be counted as absent when they were not supposed tobe on duty Measured absences in education were slightly lower in later surveyrounds consistent with the hypothesis that awareness of the first round of thesurvey created a bit of a ldquowarning effectrdquo regarding the presence of the surveyteams Adjusting for survey round and time-of-day effects would increase theestimated teacher absence by 1ndash2 percentage points (Kremer et al 2004) Nosimilar effect was found in health

We do not think that the absence rate is overstated because health workerswere working outside the facility At the beginning of the facility interview theenumerator asked to see the schedule of all health workers Only those assigned towork at the clinic on the day of the interview (as opposed for example to beingassigned to a subclinic for that day) were included in the sample Moreover we didnot find that health workers whose schedules include outreach or field work areabsent more than those who are always supposed to be in the clinic such aspharmacists A recent detailed study in Rajasthan which found absence ratessimilar to those we report made efforts to track down nurses who were absent fromhealth subcenters and found that only in 12 percent of cases of absence was thenurse in one of the villages served by her subcenter (Banerjee Deaton and Duflo2004)

High Absence Rates

At 19 percent and 35 percent respectively absence rates among teachers andhealth care workers in developing countries are high relative to those of both theircounterparts in developed countries and other workers in developing countriesStrictly comparable numbers are not available for the United States but adminis-trative data from a large sample of school districts in New York state in themid-1980s revealed a mean absence rate of 5 percent (Ehrenberg Rees and

3 This included employees who might have been on authorized leave that day although as we arguebelow reports of leave were often not credible

Missing in Action Teacher and Health Worker Absence in Developing Countries 95

Ehrenberg 1991) Even among Indian factory workers who enjoy a high degree ofjob security due to rigid labor laws reported absence rates are only around 105percent (Ministry of Labor Industry Survey 2000ndash2001) much lower than the 25and 40 percent rates of absence among Indian teachers and medical personnelrespectively

The welfare consequence of teacher and health worker absence may be evengreater in the countries that we surveyed than they would be in developed coun-tries In low-income countries substitutes rarely replace absent teachers and sostudents simply mill around go home or join another class often of a differentgrade Small schools and clinics are common in rural areas of developing countriesand these may be closed entirely as a result of provider absence In nearly12 percent of the visits enumerators in India encountered schools that were closedbecause no teacher was present An estimate of the effect of teacher absence onstudent outcomes is provided by Duflo and Hanna (2005) who show that arandomized intervention that reduced teacher absence from 36 to 18 percent ledto a 017 standard deviation improvement in student test scores

As noted in the introduction many teachers and health workers who are intheir facilities are not working Across Indian government-run schools we find thatonly 45 percent of teachers assigned to a school are engaged in teaching activity atany given point in timemdasheven though teaching activity was defined very broadly toinclude even cases where the teacher was simply keeping class in order and noactual teaching was taking place According to the official schedules teachersshould be teaching most of the time when school is in session Fewer than30 percent of schools in the sample had more teachers than classes and the schoolschedule is therefore typically designed so that teachers and students have breaksat the same time rather than with teachers having certain periods off to prepareas in most schools in developed countries Assuming that the number of teacherswho should officially be teaching is equal to the minimum of the number of classesand the number of teachers4 only 50 percent of teachers in Indian public schoolswho should be teaching at a given point are in fact doing so

In assessing these activity numbers itrsquos worth bearing in mind that they couldpotentially have been affected by the presence of the surveyor On the one handenumerators report that teachers sometimes started teaching when the surveyorarrived On the other hand although the enumerators were instructed to look fora respondent who was not teaching to ask questions regarding the school (andtypically they found the headmaster or other teacher in the office) the survey itselfmay have diverted teachers from teaching in some cases But even if we excludethose teachers from the calculation whose activity was recorded as ldquotalking to theenumeratorrdquo only 55 percent of those teachers who should have been teachingwere doing so

4 So if a school had four classes and three teachers we would expect three teachers to be teachingwhereas if it had five teachers and four classes we would only expect four teachers to be teaching

96 Journal of Economic Perspectives

Absence Across Sectors and Countries

Two clear generalizations emerge from the cross-country cross-sector data onabsence and from the variation across Indian states First health care providers aremuch more likely to be absent than teachers As Table 1 shows averaging acrosscountries for which we have data on absence for both types of providers health careworkers are 15 percentage points more likely to be absent than are teachers Thisdifference may arise because health care workers have more opportunities tomoonlight at other jobs or because health care workers receive smaller rentsrelative to what they would earn in the private sector or because health careworkers are harder to monitor If a teacher does not show up regularly a class fullof pupils and potentially their parents will know about it On the other hand it ismuch harder for patients who presumably come to health care centers irregularlyto know if a particular health care worker is absent frequently

Second higher-income areas have lower absence rates Figure 1 shows theabsence-income relationship for the sample countries other than India (repre-sented by triangles and labeled) and for the Indian states in our sample (repre-sented by circles) The left-hand panel shows the relationship among teachers theright-hand panel among health-care workers Combining the two sectors acrosscountries and Indian states an ordinary least squares regression of absence on logof per capita GDP (measured in purchasing power parity terms) and a dummy forsector (health or education) suggests that doubling of per capita income is asso-ciated with 60 percentage points lower absence The coefficient on per capitaincome is significant at the 1 percent level and the income and sector variablestogether account for more than half of the variation in sector-country and sector-state absence rates When we run two separate regressions one for the countriesand one for the Indian states we obtain very similar coefficients on log income Inthe cross-country regression doubling income is associated with a 58 percentage-point decline in absence and in the Indian cross-state regression a 48 percentage-point drop

However the relationship between a countryrsquos per capita income and absenceis stronger in education than in health Among teachers doubling income isassociated with an 80 percentage-point absence decline (significant at the01 percent level) compared with only a 38 percentage point decline in healthworker absence (falling short of significance at even the 10 percent level)5

Again a very similar pattern holds in the cross-country and the Indian cross-state regressions

One possible explanation for the correlation between income and absence isthat exogenous variation in institutional quality in service provision drives human

5 The absence-income relationship in the health sector appears to hold more strongly for doctors thanfor other medical personnel Within India regressing doctor absence on state per capita income yieldsa much larger coefficient (in absolute value) significant at the 10 percent level whereas the coefficientis small and insignificant for health workers as a group

Nazmul Chaudhury et al 97

capital acquisition and thus income Another is that the overall level of develop-ment drives the quality of education and health delivery While it is impossible todisentangle these stories completely to the extent that the overall level of devel-opment influences provider absence one might expect low income levels to lead tohigh absence rates in both education and health On the other hand if educationis particularly important for human capital acquisition and thus income whilemedical clinics have a larger consumption component then exogenous variation inquality of education systems will lead to variation in income while the quality ofhealth care systems will be less correlated with income This pattern matches whatwe see in the data

It is intriguing that the relationship between income and absence is so similaracross countries and across Indian states and that it is so tight in each case Whilesalaries typically rise with GDP (although not proportionally) teacher salariesacross Indian states are relatively flat6 Thus across the states of India salaries forteachers and health workers in poor states are considerably higher relative to thecost of living and relative to workersrsquo outside opportunities than are salaries in richstates Nonetheless absence rates are higher in poor states The similarity betweenthe absence-income regression line across countries and the comparable line acrossIndian states despite the difference in the relationship between income andsalaries in the two samples suggests a limited role for salaries in influencing

6 Ministry of Human Resource Development India

Figure 1Absence Rate versus NationalState Per Capita Income

Source Authorsrsquo calculationsNote BNG Bangladesh ECU Ecuador IDN Indonesia PER Peru UGA Uganda Indiarsquosnational averages are excluded due to the inclusion of the Indian states For Indian states incomesare the official per capita net state domestic products

98 Journal of Economic Perspectives

absence over the existing salary range Of course it is important to bear in mindthat the samples of countries and states are very small and other factors couldinfluence these slopes

Teacher and health worker absence are correlated across countries and stateseven after controlling for per capita income The residuals from the two regressionsdepicted in Figure 1 (with an additional dummy added for Indian states) are highlycorrelated with each other with a correlation coefficient of 044 (significant at the5 percent level) This correlation could potentially be due to mismeasurement ofincome but it could also reflect spillover effects in social norms across sectors oromitted variables such as the quality of governance

Concentration of Absence

To understand and potentially design policies to counter high absence ratesit is useful to know whether absences are spread out among providers or concen-trated among a small number of ldquoghost workersrdquo who are on the books but nevershow up Since our survey included only two or three observations per worker wewould observe some dispersion in absence rates even if all workers had identicalunderlying probabilities of being absent The left panel of Table 2 shows thedistribution of absence observed in the data For comparison the right panel showsthe distribution that would be observed if the probability of absence in each visitwere equal to the estimated absence rate in the specific country-sector combina-tion so all workers had the same probability of being absent For example if allteachers in Indonesia had a 019 chance of being absent (which is the averageteacher absence rate there) then on any two independent visits we would expect36 percent (019 019) to be absent both times 656 percent (081 081) to bepresent both times and the remaining 308 percent to be absent once On the otherhand if absence were completely concentrated in certain providers we wouldobserve that 19 percent of the teachers are always absent 81 percent are alwayspresent and none are absent only once

Clearly the data match neither the extreme of all workers having identicalunderlying probabilities of absence nor of all absence being due to ghost workersbut an eyeball test suggests that absence appears to be fairly widespread with theempirical distribution surprisingly close to that predicted by a model with identicalabsence probabilities Teachers in Ecuador are an exception and appear to be theleading candidates for a ldquoghost workerrdquo explanation with a very high percentage ofteachers being present in both visits and more teachers absent in both visits than inone of the two visits

The exercise above while suggestive can technically only be used to test theextreme hypotheses of complete concentration of absence and perfectly identicalabsence rates among workers Glewwe Ilias and Kremer (2004) assume providersrsquounderlying probability of absence follows a beta distribution and estimate thisdistribution in two districts of Kenya using a maximum likelihood approach They

Missing in Action Teacher and Health Worker Absence in Developing Countries 99

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 4: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

of teachers show up each day despite an apparent lack of significant rewards orpunishments related to teacher performance (Alcazar et al 2005)

Against the background of these highly formalized and bureaucratized sys-tems a plethora of informal systems have grown up virtually outside the ambit ofregulation These include private schools and clinics that are not recognized by thegovernment publicly supported community-managed schools such as nonformaleducation centers in India and systems for hiring contract teachers at publicschools outside of normal civil service rules Teachers in these informal systemsoften have lower educational qualifications than their civil service counterpartsearn much less (often only a third as much or lower) and have little or no jobsecurity Hiring and salary decisions are subject to more discretion with lessemphasis on formal educational qualifications There are also a range of healthproviders outside of formal government systems including many nonlicensedproviders without medical education as well as government providers operatingprivate practices on the side

We conducted a survey focused on the presence of teachers and health workersat public primary schools and primary health centers to assess what would seem toconstitute a minimal prima facie condition for efficacy of these systems Surveyswere typically close to nationally representative but excluded some areas from thesampling frame for security or logistical reasons2 In rural India enumerators alsocollected data from private schools and nonformal education centers located in thesame village as public schools and in Indonesia they also collected data fromprivate schools As we discuss below absence rates are high in the informal sectoras well as the formal sector

Our absence data are based on direct physical verification of the providerrsquospresence rather than attendance logbooks or interviews with the facility head InBangladesh Ecuador Indonesia Peru and Uganda enumerators made two visitsmdashtypically several months apartmdashto each of about ten randomly chosen health carecenters and ten randomly chosen public schools in each of ten randomly chosendistricts On average we visited 100 schools and 100 health care centers in eachcountry With around eight providers in the average facility and two observationson each of these providers we had an average of over 1500 observations on teacherattendance in each country and an average of over 1350 observations for healthworker attendance in each country In India the survey was designed to berepresentative in each of 20 states which together account for 98 percent of Indiarsquospopulation Three unannounced visits were made to each of about 3000 publicschools over a span of three to four months Since the average school in our samplehas around four teachers we have nearly 35000 observations on teacher atten-dance Similarly enumerators made three unannounced visits to over 1350 publicclinics and since these had an average of eight or nine health workers each wehave approximately 32500 observations on health worker presence The majority

2 In Indonesia the excluded provinces account for only about 8 percent of the countryrsquos population inother countries even less

94 Journal of Economic Perspectives

of the field work in all countries was carried out between October 2002 and April2003

A worker was counted as absent if at the time of a random visit during facilityhours he or she was not in the school or health center The enumerators for thesurvey took several measures to ensure that the rate of absence would not beoverestimated The list of employees used for checking attendance was created atthe facility itself based on staff lists and schedule information provided by thefacility director or other principal respondent Enumerators then checked theattendance only of those who were ordinarily supposed to be on duty at the time ofthe visit3 We omitted from the absence calculations all employees who werereported by the director as being on another shift whether or not this could beverified Only full-time employees were included in our analysis to minimize therisk that shift workers would be counted as absent when they were not supposed tobe on duty Measured absences in education were slightly lower in later surveyrounds consistent with the hypothesis that awareness of the first round of thesurvey created a bit of a ldquowarning effectrdquo regarding the presence of the surveyteams Adjusting for survey round and time-of-day effects would increase theestimated teacher absence by 1ndash2 percentage points (Kremer et al 2004) Nosimilar effect was found in health

We do not think that the absence rate is overstated because health workerswere working outside the facility At the beginning of the facility interview theenumerator asked to see the schedule of all health workers Only those assigned towork at the clinic on the day of the interview (as opposed for example to beingassigned to a subclinic for that day) were included in the sample Moreover we didnot find that health workers whose schedules include outreach or field work areabsent more than those who are always supposed to be in the clinic such aspharmacists A recent detailed study in Rajasthan which found absence ratessimilar to those we report made efforts to track down nurses who were absent fromhealth subcenters and found that only in 12 percent of cases of absence was thenurse in one of the villages served by her subcenter (Banerjee Deaton and Duflo2004)

High Absence Rates

At 19 percent and 35 percent respectively absence rates among teachers andhealth care workers in developing countries are high relative to those of both theircounterparts in developed countries and other workers in developing countriesStrictly comparable numbers are not available for the United States but adminis-trative data from a large sample of school districts in New York state in themid-1980s revealed a mean absence rate of 5 percent (Ehrenberg Rees and

3 This included employees who might have been on authorized leave that day although as we arguebelow reports of leave were often not credible

Missing in Action Teacher and Health Worker Absence in Developing Countries 95

Ehrenberg 1991) Even among Indian factory workers who enjoy a high degree ofjob security due to rigid labor laws reported absence rates are only around 105percent (Ministry of Labor Industry Survey 2000ndash2001) much lower than the 25and 40 percent rates of absence among Indian teachers and medical personnelrespectively

The welfare consequence of teacher and health worker absence may be evengreater in the countries that we surveyed than they would be in developed coun-tries In low-income countries substitutes rarely replace absent teachers and sostudents simply mill around go home or join another class often of a differentgrade Small schools and clinics are common in rural areas of developing countriesand these may be closed entirely as a result of provider absence In nearly12 percent of the visits enumerators in India encountered schools that were closedbecause no teacher was present An estimate of the effect of teacher absence onstudent outcomes is provided by Duflo and Hanna (2005) who show that arandomized intervention that reduced teacher absence from 36 to 18 percent ledto a 017 standard deviation improvement in student test scores

As noted in the introduction many teachers and health workers who are intheir facilities are not working Across Indian government-run schools we find thatonly 45 percent of teachers assigned to a school are engaged in teaching activity atany given point in timemdasheven though teaching activity was defined very broadly toinclude even cases where the teacher was simply keeping class in order and noactual teaching was taking place According to the official schedules teachersshould be teaching most of the time when school is in session Fewer than30 percent of schools in the sample had more teachers than classes and the schoolschedule is therefore typically designed so that teachers and students have breaksat the same time rather than with teachers having certain periods off to prepareas in most schools in developed countries Assuming that the number of teacherswho should officially be teaching is equal to the minimum of the number of classesand the number of teachers4 only 50 percent of teachers in Indian public schoolswho should be teaching at a given point are in fact doing so

In assessing these activity numbers itrsquos worth bearing in mind that they couldpotentially have been affected by the presence of the surveyor On the one handenumerators report that teachers sometimes started teaching when the surveyorarrived On the other hand although the enumerators were instructed to look fora respondent who was not teaching to ask questions regarding the school (andtypically they found the headmaster or other teacher in the office) the survey itselfmay have diverted teachers from teaching in some cases But even if we excludethose teachers from the calculation whose activity was recorded as ldquotalking to theenumeratorrdquo only 55 percent of those teachers who should have been teachingwere doing so

4 So if a school had four classes and three teachers we would expect three teachers to be teachingwhereas if it had five teachers and four classes we would only expect four teachers to be teaching

96 Journal of Economic Perspectives

Absence Across Sectors and Countries

Two clear generalizations emerge from the cross-country cross-sector data onabsence and from the variation across Indian states First health care providers aremuch more likely to be absent than teachers As Table 1 shows averaging acrosscountries for which we have data on absence for both types of providers health careworkers are 15 percentage points more likely to be absent than are teachers Thisdifference may arise because health care workers have more opportunities tomoonlight at other jobs or because health care workers receive smaller rentsrelative to what they would earn in the private sector or because health careworkers are harder to monitor If a teacher does not show up regularly a class fullof pupils and potentially their parents will know about it On the other hand it ismuch harder for patients who presumably come to health care centers irregularlyto know if a particular health care worker is absent frequently

Second higher-income areas have lower absence rates Figure 1 shows theabsence-income relationship for the sample countries other than India (repre-sented by triangles and labeled) and for the Indian states in our sample (repre-sented by circles) The left-hand panel shows the relationship among teachers theright-hand panel among health-care workers Combining the two sectors acrosscountries and Indian states an ordinary least squares regression of absence on logof per capita GDP (measured in purchasing power parity terms) and a dummy forsector (health or education) suggests that doubling of per capita income is asso-ciated with 60 percentage points lower absence The coefficient on per capitaincome is significant at the 1 percent level and the income and sector variablestogether account for more than half of the variation in sector-country and sector-state absence rates When we run two separate regressions one for the countriesand one for the Indian states we obtain very similar coefficients on log income Inthe cross-country regression doubling income is associated with a 58 percentage-point decline in absence and in the Indian cross-state regression a 48 percentage-point drop

However the relationship between a countryrsquos per capita income and absenceis stronger in education than in health Among teachers doubling income isassociated with an 80 percentage-point absence decline (significant at the01 percent level) compared with only a 38 percentage point decline in healthworker absence (falling short of significance at even the 10 percent level)5

Again a very similar pattern holds in the cross-country and the Indian cross-state regressions

One possible explanation for the correlation between income and absence isthat exogenous variation in institutional quality in service provision drives human

5 The absence-income relationship in the health sector appears to hold more strongly for doctors thanfor other medical personnel Within India regressing doctor absence on state per capita income yieldsa much larger coefficient (in absolute value) significant at the 10 percent level whereas the coefficientis small and insignificant for health workers as a group

Nazmul Chaudhury et al 97

capital acquisition and thus income Another is that the overall level of develop-ment drives the quality of education and health delivery While it is impossible todisentangle these stories completely to the extent that the overall level of devel-opment influences provider absence one might expect low income levels to lead tohigh absence rates in both education and health On the other hand if educationis particularly important for human capital acquisition and thus income whilemedical clinics have a larger consumption component then exogenous variation inquality of education systems will lead to variation in income while the quality ofhealth care systems will be less correlated with income This pattern matches whatwe see in the data

It is intriguing that the relationship between income and absence is so similaracross countries and across Indian states and that it is so tight in each case Whilesalaries typically rise with GDP (although not proportionally) teacher salariesacross Indian states are relatively flat6 Thus across the states of India salaries forteachers and health workers in poor states are considerably higher relative to thecost of living and relative to workersrsquo outside opportunities than are salaries in richstates Nonetheless absence rates are higher in poor states The similarity betweenthe absence-income regression line across countries and the comparable line acrossIndian states despite the difference in the relationship between income andsalaries in the two samples suggests a limited role for salaries in influencing

6 Ministry of Human Resource Development India

Figure 1Absence Rate versus NationalState Per Capita Income

Source Authorsrsquo calculationsNote BNG Bangladesh ECU Ecuador IDN Indonesia PER Peru UGA Uganda Indiarsquosnational averages are excluded due to the inclusion of the Indian states For Indian states incomesare the official per capita net state domestic products

98 Journal of Economic Perspectives

absence over the existing salary range Of course it is important to bear in mindthat the samples of countries and states are very small and other factors couldinfluence these slopes

Teacher and health worker absence are correlated across countries and stateseven after controlling for per capita income The residuals from the two regressionsdepicted in Figure 1 (with an additional dummy added for Indian states) are highlycorrelated with each other with a correlation coefficient of 044 (significant at the5 percent level) This correlation could potentially be due to mismeasurement ofincome but it could also reflect spillover effects in social norms across sectors oromitted variables such as the quality of governance

Concentration of Absence

To understand and potentially design policies to counter high absence ratesit is useful to know whether absences are spread out among providers or concen-trated among a small number of ldquoghost workersrdquo who are on the books but nevershow up Since our survey included only two or three observations per worker wewould observe some dispersion in absence rates even if all workers had identicalunderlying probabilities of being absent The left panel of Table 2 shows thedistribution of absence observed in the data For comparison the right panel showsthe distribution that would be observed if the probability of absence in each visitwere equal to the estimated absence rate in the specific country-sector combina-tion so all workers had the same probability of being absent For example if allteachers in Indonesia had a 019 chance of being absent (which is the averageteacher absence rate there) then on any two independent visits we would expect36 percent (019 019) to be absent both times 656 percent (081 081) to bepresent both times and the remaining 308 percent to be absent once On the otherhand if absence were completely concentrated in certain providers we wouldobserve that 19 percent of the teachers are always absent 81 percent are alwayspresent and none are absent only once

Clearly the data match neither the extreme of all workers having identicalunderlying probabilities of absence nor of all absence being due to ghost workersbut an eyeball test suggests that absence appears to be fairly widespread with theempirical distribution surprisingly close to that predicted by a model with identicalabsence probabilities Teachers in Ecuador are an exception and appear to be theleading candidates for a ldquoghost workerrdquo explanation with a very high percentage ofteachers being present in both visits and more teachers absent in both visits than inone of the two visits

The exercise above while suggestive can technically only be used to test theextreme hypotheses of complete concentration of absence and perfectly identicalabsence rates among workers Glewwe Ilias and Kremer (2004) assume providersrsquounderlying probability of absence follows a beta distribution and estimate thisdistribution in two districts of Kenya using a maximum likelihood approach They

Missing in Action Teacher and Health Worker Absence in Developing Countries 99

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 5: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

of the field work in all countries was carried out between October 2002 and April2003

A worker was counted as absent if at the time of a random visit during facilityhours he or she was not in the school or health center The enumerators for thesurvey took several measures to ensure that the rate of absence would not beoverestimated The list of employees used for checking attendance was created atthe facility itself based on staff lists and schedule information provided by thefacility director or other principal respondent Enumerators then checked theattendance only of those who were ordinarily supposed to be on duty at the time ofthe visit3 We omitted from the absence calculations all employees who werereported by the director as being on another shift whether or not this could beverified Only full-time employees were included in our analysis to minimize therisk that shift workers would be counted as absent when they were not supposed tobe on duty Measured absences in education were slightly lower in later surveyrounds consistent with the hypothesis that awareness of the first round of thesurvey created a bit of a ldquowarning effectrdquo regarding the presence of the surveyteams Adjusting for survey round and time-of-day effects would increase theestimated teacher absence by 1ndash2 percentage points (Kremer et al 2004) Nosimilar effect was found in health

We do not think that the absence rate is overstated because health workerswere working outside the facility At the beginning of the facility interview theenumerator asked to see the schedule of all health workers Only those assigned towork at the clinic on the day of the interview (as opposed for example to beingassigned to a subclinic for that day) were included in the sample Moreover we didnot find that health workers whose schedules include outreach or field work areabsent more than those who are always supposed to be in the clinic such aspharmacists A recent detailed study in Rajasthan which found absence ratessimilar to those we report made efforts to track down nurses who were absent fromhealth subcenters and found that only in 12 percent of cases of absence was thenurse in one of the villages served by her subcenter (Banerjee Deaton and Duflo2004)

High Absence Rates

At 19 percent and 35 percent respectively absence rates among teachers andhealth care workers in developing countries are high relative to those of both theircounterparts in developed countries and other workers in developing countriesStrictly comparable numbers are not available for the United States but adminis-trative data from a large sample of school districts in New York state in themid-1980s revealed a mean absence rate of 5 percent (Ehrenberg Rees and

3 This included employees who might have been on authorized leave that day although as we arguebelow reports of leave were often not credible

Missing in Action Teacher and Health Worker Absence in Developing Countries 95

Ehrenberg 1991) Even among Indian factory workers who enjoy a high degree ofjob security due to rigid labor laws reported absence rates are only around 105percent (Ministry of Labor Industry Survey 2000ndash2001) much lower than the 25and 40 percent rates of absence among Indian teachers and medical personnelrespectively

The welfare consequence of teacher and health worker absence may be evengreater in the countries that we surveyed than they would be in developed coun-tries In low-income countries substitutes rarely replace absent teachers and sostudents simply mill around go home or join another class often of a differentgrade Small schools and clinics are common in rural areas of developing countriesand these may be closed entirely as a result of provider absence In nearly12 percent of the visits enumerators in India encountered schools that were closedbecause no teacher was present An estimate of the effect of teacher absence onstudent outcomes is provided by Duflo and Hanna (2005) who show that arandomized intervention that reduced teacher absence from 36 to 18 percent ledto a 017 standard deviation improvement in student test scores

As noted in the introduction many teachers and health workers who are intheir facilities are not working Across Indian government-run schools we find thatonly 45 percent of teachers assigned to a school are engaged in teaching activity atany given point in timemdasheven though teaching activity was defined very broadly toinclude even cases where the teacher was simply keeping class in order and noactual teaching was taking place According to the official schedules teachersshould be teaching most of the time when school is in session Fewer than30 percent of schools in the sample had more teachers than classes and the schoolschedule is therefore typically designed so that teachers and students have breaksat the same time rather than with teachers having certain periods off to prepareas in most schools in developed countries Assuming that the number of teacherswho should officially be teaching is equal to the minimum of the number of classesand the number of teachers4 only 50 percent of teachers in Indian public schoolswho should be teaching at a given point are in fact doing so

In assessing these activity numbers itrsquos worth bearing in mind that they couldpotentially have been affected by the presence of the surveyor On the one handenumerators report that teachers sometimes started teaching when the surveyorarrived On the other hand although the enumerators were instructed to look fora respondent who was not teaching to ask questions regarding the school (andtypically they found the headmaster or other teacher in the office) the survey itselfmay have diverted teachers from teaching in some cases But even if we excludethose teachers from the calculation whose activity was recorded as ldquotalking to theenumeratorrdquo only 55 percent of those teachers who should have been teachingwere doing so

4 So if a school had four classes and three teachers we would expect three teachers to be teachingwhereas if it had five teachers and four classes we would only expect four teachers to be teaching

96 Journal of Economic Perspectives

Absence Across Sectors and Countries

Two clear generalizations emerge from the cross-country cross-sector data onabsence and from the variation across Indian states First health care providers aremuch more likely to be absent than teachers As Table 1 shows averaging acrosscountries for which we have data on absence for both types of providers health careworkers are 15 percentage points more likely to be absent than are teachers Thisdifference may arise because health care workers have more opportunities tomoonlight at other jobs or because health care workers receive smaller rentsrelative to what they would earn in the private sector or because health careworkers are harder to monitor If a teacher does not show up regularly a class fullof pupils and potentially their parents will know about it On the other hand it ismuch harder for patients who presumably come to health care centers irregularlyto know if a particular health care worker is absent frequently

Second higher-income areas have lower absence rates Figure 1 shows theabsence-income relationship for the sample countries other than India (repre-sented by triangles and labeled) and for the Indian states in our sample (repre-sented by circles) The left-hand panel shows the relationship among teachers theright-hand panel among health-care workers Combining the two sectors acrosscountries and Indian states an ordinary least squares regression of absence on logof per capita GDP (measured in purchasing power parity terms) and a dummy forsector (health or education) suggests that doubling of per capita income is asso-ciated with 60 percentage points lower absence The coefficient on per capitaincome is significant at the 1 percent level and the income and sector variablestogether account for more than half of the variation in sector-country and sector-state absence rates When we run two separate regressions one for the countriesand one for the Indian states we obtain very similar coefficients on log income Inthe cross-country regression doubling income is associated with a 58 percentage-point decline in absence and in the Indian cross-state regression a 48 percentage-point drop

However the relationship between a countryrsquos per capita income and absenceis stronger in education than in health Among teachers doubling income isassociated with an 80 percentage-point absence decline (significant at the01 percent level) compared with only a 38 percentage point decline in healthworker absence (falling short of significance at even the 10 percent level)5

Again a very similar pattern holds in the cross-country and the Indian cross-state regressions

One possible explanation for the correlation between income and absence isthat exogenous variation in institutional quality in service provision drives human

5 The absence-income relationship in the health sector appears to hold more strongly for doctors thanfor other medical personnel Within India regressing doctor absence on state per capita income yieldsa much larger coefficient (in absolute value) significant at the 10 percent level whereas the coefficientis small and insignificant for health workers as a group

Nazmul Chaudhury et al 97

capital acquisition and thus income Another is that the overall level of develop-ment drives the quality of education and health delivery While it is impossible todisentangle these stories completely to the extent that the overall level of devel-opment influences provider absence one might expect low income levels to lead tohigh absence rates in both education and health On the other hand if educationis particularly important for human capital acquisition and thus income whilemedical clinics have a larger consumption component then exogenous variation inquality of education systems will lead to variation in income while the quality ofhealth care systems will be less correlated with income This pattern matches whatwe see in the data

It is intriguing that the relationship between income and absence is so similaracross countries and across Indian states and that it is so tight in each case Whilesalaries typically rise with GDP (although not proportionally) teacher salariesacross Indian states are relatively flat6 Thus across the states of India salaries forteachers and health workers in poor states are considerably higher relative to thecost of living and relative to workersrsquo outside opportunities than are salaries in richstates Nonetheless absence rates are higher in poor states The similarity betweenthe absence-income regression line across countries and the comparable line acrossIndian states despite the difference in the relationship between income andsalaries in the two samples suggests a limited role for salaries in influencing

6 Ministry of Human Resource Development India

Figure 1Absence Rate versus NationalState Per Capita Income

Source Authorsrsquo calculationsNote BNG Bangladesh ECU Ecuador IDN Indonesia PER Peru UGA Uganda Indiarsquosnational averages are excluded due to the inclusion of the Indian states For Indian states incomesare the official per capita net state domestic products

98 Journal of Economic Perspectives

absence over the existing salary range Of course it is important to bear in mindthat the samples of countries and states are very small and other factors couldinfluence these slopes

Teacher and health worker absence are correlated across countries and stateseven after controlling for per capita income The residuals from the two regressionsdepicted in Figure 1 (with an additional dummy added for Indian states) are highlycorrelated with each other with a correlation coefficient of 044 (significant at the5 percent level) This correlation could potentially be due to mismeasurement ofincome but it could also reflect spillover effects in social norms across sectors oromitted variables such as the quality of governance

Concentration of Absence

To understand and potentially design policies to counter high absence ratesit is useful to know whether absences are spread out among providers or concen-trated among a small number of ldquoghost workersrdquo who are on the books but nevershow up Since our survey included only two or three observations per worker wewould observe some dispersion in absence rates even if all workers had identicalunderlying probabilities of being absent The left panel of Table 2 shows thedistribution of absence observed in the data For comparison the right panel showsthe distribution that would be observed if the probability of absence in each visitwere equal to the estimated absence rate in the specific country-sector combina-tion so all workers had the same probability of being absent For example if allteachers in Indonesia had a 019 chance of being absent (which is the averageteacher absence rate there) then on any two independent visits we would expect36 percent (019 019) to be absent both times 656 percent (081 081) to bepresent both times and the remaining 308 percent to be absent once On the otherhand if absence were completely concentrated in certain providers we wouldobserve that 19 percent of the teachers are always absent 81 percent are alwayspresent and none are absent only once

Clearly the data match neither the extreme of all workers having identicalunderlying probabilities of absence nor of all absence being due to ghost workersbut an eyeball test suggests that absence appears to be fairly widespread with theempirical distribution surprisingly close to that predicted by a model with identicalabsence probabilities Teachers in Ecuador are an exception and appear to be theleading candidates for a ldquoghost workerrdquo explanation with a very high percentage ofteachers being present in both visits and more teachers absent in both visits than inone of the two visits

The exercise above while suggestive can technically only be used to test theextreme hypotheses of complete concentration of absence and perfectly identicalabsence rates among workers Glewwe Ilias and Kremer (2004) assume providersrsquounderlying probability of absence follows a beta distribution and estimate thisdistribution in two districts of Kenya using a maximum likelihood approach They

Missing in Action Teacher and Health Worker Absence in Developing Countries 99

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 6: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

Ehrenberg 1991) Even among Indian factory workers who enjoy a high degree ofjob security due to rigid labor laws reported absence rates are only around 105percent (Ministry of Labor Industry Survey 2000ndash2001) much lower than the 25and 40 percent rates of absence among Indian teachers and medical personnelrespectively

The welfare consequence of teacher and health worker absence may be evengreater in the countries that we surveyed than they would be in developed coun-tries In low-income countries substitutes rarely replace absent teachers and sostudents simply mill around go home or join another class often of a differentgrade Small schools and clinics are common in rural areas of developing countriesand these may be closed entirely as a result of provider absence In nearly12 percent of the visits enumerators in India encountered schools that were closedbecause no teacher was present An estimate of the effect of teacher absence onstudent outcomes is provided by Duflo and Hanna (2005) who show that arandomized intervention that reduced teacher absence from 36 to 18 percent ledto a 017 standard deviation improvement in student test scores

As noted in the introduction many teachers and health workers who are intheir facilities are not working Across Indian government-run schools we find thatonly 45 percent of teachers assigned to a school are engaged in teaching activity atany given point in timemdasheven though teaching activity was defined very broadly toinclude even cases where the teacher was simply keeping class in order and noactual teaching was taking place According to the official schedules teachersshould be teaching most of the time when school is in session Fewer than30 percent of schools in the sample had more teachers than classes and the schoolschedule is therefore typically designed so that teachers and students have breaksat the same time rather than with teachers having certain periods off to prepareas in most schools in developed countries Assuming that the number of teacherswho should officially be teaching is equal to the minimum of the number of classesand the number of teachers4 only 50 percent of teachers in Indian public schoolswho should be teaching at a given point are in fact doing so

In assessing these activity numbers itrsquos worth bearing in mind that they couldpotentially have been affected by the presence of the surveyor On the one handenumerators report that teachers sometimes started teaching when the surveyorarrived On the other hand although the enumerators were instructed to look fora respondent who was not teaching to ask questions regarding the school (andtypically they found the headmaster or other teacher in the office) the survey itselfmay have diverted teachers from teaching in some cases But even if we excludethose teachers from the calculation whose activity was recorded as ldquotalking to theenumeratorrdquo only 55 percent of those teachers who should have been teachingwere doing so

4 So if a school had four classes and three teachers we would expect three teachers to be teachingwhereas if it had five teachers and four classes we would only expect four teachers to be teaching

96 Journal of Economic Perspectives

Absence Across Sectors and Countries

Two clear generalizations emerge from the cross-country cross-sector data onabsence and from the variation across Indian states First health care providers aremuch more likely to be absent than teachers As Table 1 shows averaging acrosscountries for which we have data on absence for both types of providers health careworkers are 15 percentage points more likely to be absent than are teachers Thisdifference may arise because health care workers have more opportunities tomoonlight at other jobs or because health care workers receive smaller rentsrelative to what they would earn in the private sector or because health careworkers are harder to monitor If a teacher does not show up regularly a class fullof pupils and potentially their parents will know about it On the other hand it ismuch harder for patients who presumably come to health care centers irregularlyto know if a particular health care worker is absent frequently

Second higher-income areas have lower absence rates Figure 1 shows theabsence-income relationship for the sample countries other than India (repre-sented by triangles and labeled) and for the Indian states in our sample (repre-sented by circles) The left-hand panel shows the relationship among teachers theright-hand panel among health-care workers Combining the two sectors acrosscountries and Indian states an ordinary least squares regression of absence on logof per capita GDP (measured in purchasing power parity terms) and a dummy forsector (health or education) suggests that doubling of per capita income is asso-ciated with 60 percentage points lower absence The coefficient on per capitaincome is significant at the 1 percent level and the income and sector variablestogether account for more than half of the variation in sector-country and sector-state absence rates When we run two separate regressions one for the countriesand one for the Indian states we obtain very similar coefficients on log income Inthe cross-country regression doubling income is associated with a 58 percentage-point decline in absence and in the Indian cross-state regression a 48 percentage-point drop

However the relationship between a countryrsquos per capita income and absenceis stronger in education than in health Among teachers doubling income isassociated with an 80 percentage-point absence decline (significant at the01 percent level) compared with only a 38 percentage point decline in healthworker absence (falling short of significance at even the 10 percent level)5

Again a very similar pattern holds in the cross-country and the Indian cross-state regressions

One possible explanation for the correlation between income and absence isthat exogenous variation in institutional quality in service provision drives human

5 The absence-income relationship in the health sector appears to hold more strongly for doctors thanfor other medical personnel Within India regressing doctor absence on state per capita income yieldsa much larger coefficient (in absolute value) significant at the 10 percent level whereas the coefficientis small and insignificant for health workers as a group

Nazmul Chaudhury et al 97

capital acquisition and thus income Another is that the overall level of develop-ment drives the quality of education and health delivery While it is impossible todisentangle these stories completely to the extent that the overall level of devel-opment influences provider absence one might expect low income levels to lead tohigh absence rates in both education and health On the other hand if educationis particularly important for human capital acquisition and thus income whilemedical clinics have a larger consumption component then exogenous variation inquality of education systems will lead to variation in income while the quality ofhealth care systems will be less correlated with income This pattern matches whatwe see in the data

It is intriguing that the relationship between income and absence is so similaracross countries and across Indian states and that it is so tight in each case Whilesalaries typically rise with GDP (although not proportionally) teacher salariesacross Indian states are relatively flat6 Thus across the states of India salaries forteachers and health workers in poor states are considerably higher relative to thecost of living and relative to workersrsquo outside opportunities than are salaries in richstates Nonetheless absence rates are higher in poor states The similarity betweenthe absence-income regression line across countries and the comparable line acrossIndian states despite the difference in the relationship between income andsalaries in the two samples suggests a limited role for salaries in influencing

6 Ministry of Human Resource Development India

Figure 1Absence Rate versus NationalState Per Capita Income

Source Authorsrsquo calculationsNote BNG Bangladesh ECU Ecuador IDN Indonesia PER Peru UGA Uganda Indiarsquosnational averages are excluded due to the inclusion of the Indian states For Indian states incomesare the official per capita net state domestic products

98 Journal of Economic Perspectives

absence over the existing salary range Of course it is important to bear in mindthat the samples of countries and states are very small and other factors couldinfluence these slopes

Teacher and health worker absence are correlated across countries and stateseven after controlling for per capita income The residuals from the two regressionsdepicted in Figure 1 (with an additional dummy added for Indian states) are highlycorrelated with each other with a correlation coefficient of 044 (significant at the5 percent level) This correlation could potentially be due to mismeasurement ofincome but it could also reflect spillover effects in social norms across sectors oromitted variables such as the quality of governance

Concentration of Absence

To understand and potentially design policies to counter high absence ratesit is useful to know whether absences are spread out among providers or concen-trated among a small number of ldquoghost workersrdquo who are on the books but nevershow up Since our survey included only two or three observations per worker wewould observe some dispersion in absence rates even if all workers had identicalunderlying probabilities of being absent The left panel of Table 2 shows thedistribution of absence observed in the data For comparison the right panel showsthe distribution that would be observed if the probability of absence in each visitwere equal to the estimated absence rate in the specific country-sector combina-tion so all workers had the same probability of being absent For example if allteachers in Indonesia had a 019 chance of being absent (which is the averageteacher absence rate there) then on any two independent visits we would expect36 percent (019 019) to be absent both times 656 percent (081 081) to bepresent both times and the remaining 308 percent to be absent once On the otherhand if absence were completely concentrated in certain providers we wouldobserve that 19 percent of the teachers are always absent 81 percent are alwayspresent and none are absent only once

Clearly the data match neither the extreme of all workers having identicalunderlying probabilities of absence nor of all absence being due to ghost workersbut an eyeball test suggests that absence appears to be fairly widespread with theempirical distribution surprisingly close to that predicted by a model with identicalabsence probabilities Teachers in Ecuador are an exception and appear to be theleading candidates for a ldquoghost workerrdquo explanation with a very high percentage ofteachers being present in both visits and more teachers absent in both visits than inone of the two visits

The exercise above while suggestive can technically only be used to test theextreme hypotheses of complete concentration of absence and perfectly identicalabsence rates among workers Glewwe Ilias and Kremer (2004) assume providersrsquounderlying probability of absence follows a beta distribution and estimate thisdistribution in two districts of Kenya using a maximum likelihood approach They

Missing in Action Teacher and Health Worker Absence in Developing Countries 99

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 7: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

Absence Across Sectors and Countries

Two clear generalizations emerge from the cross-country cross-sector data onabsence and from the variation across Indian states First health care providers aremuch more likely to be absent than teachers As Table 1 shows averaging acrosscountries for which we have data on absence for both types of providers health careworkers are 15 percentage points more likely to be absent than are teachers Thisdifference may arise because health care workers have more opportunities tomoonlight at other jobs or because health care workers receive smaller rentsrelative to what they would earn in the private sector or because health careworkers are harder to monitor If a teacher does not show up regularly a class fullof pupils and potentially their parents will know about it On the other hand it ismuch harder for patients who presumably come to health care centers irregularlyto know if a particular health care worker is absent frequently

Second higher-income areas have lower absence rates Figure 1 shows theabsence-income relationship for the sample countries other than India (repre-sented by triangles and labeled) and for the Indian states in our sample (repre-sented by circles) The left-hand panel shows the relationship among teachers theright-hand panel among health-care workers Combining the two sectors acrosscountries and Indian states an ordinary least squares regression of absence on logof per capita GDP (measured in purchasing power parity terms) and a dummy forsector (health or education) suggests that doubling of per capita income is asso-ciated with 60 percentage points lower absence The coefficient on per capitaincome is significant at the 1 percent level and the income and sector variablestogether account for more than half of the variation in sector-country and sector-state absence rates When we run two separate regressions one for the countriesand one for the Indian states we obtain very similar coefficients on log income Inthe cross-country regression doubling income is associated with a 58 percentage-point decline in absence and in the Indian cross-state regression a 48 percentage-point drop

However the relationship between a countryrsquos per capita income and absenceis stronger in education than in health Among teachers doubling income isassociated with an 80 percentage-point absence decline (significant at the01 percent level) compared with only a 38 percentage point decline in healthworker absence (falling short of significance at even the 10 percent level)5

Again a very similar pattern holds in the cross-country and the Indian cross-state regressions

One possible explanation for the correlation between income and absence isthat exogenous variation in institutional quality in service provision drives human

5 The absence-income relationship in the health sector appears to hold more strongly for doctors thanfor other medical personnel Within India regressing doctor absence on state per capita income yieldsa much larger coefficient (in absolute value) significant at the 10 percent level whereas the coefficientis small and insignificant for health workers as a group

Nazmul Chaudhury et al 97

capital acquisition and thus income Another is that the overall level of develop-ment drives the quality of education and health delivery While it is impossible todisentangle these stories completely to the extent that the overall level of devel-opment influences provider absence one might expect low income levels to lead tohigh absence rates in both education and health On the other hand if educationis particularly important for human capital acquisition and thus income whilemedical clinics have a larger consumption component then exogenous variation inquality of education systems will lead to variation in income while the quality ofhealth care systems will be less correlated with income This pattern matches whatwe see in the data

It is intriguing that the relationship between income and absence is so similaracross countries and across Indian states and that it is so tight in each case Whilesalaries typically rise with GDP (although not proportionally) teacher salariesacross Indian states are relatively flat6 Thus across the states of India salaries forteachers and health workers in poor states are considerably higher relative to thecost of living and relative to workersrsquo outside opportunities than are salaries in richstates Nonetheless absence rates are higher in poor states The similarity betweenthe absence-income regression line across countries and the comparable line acrossIndian states despite the difference in the relationship between income andsalaries in the two samples suggests a limited role for salaries in influencing

6 Ministry of Human Resource Development India

Figure 1Absence Rate versus NationalState Per Capita Income

Source Authorsrsquo calculationsNote BNG Bangladesh ECU Ecuador IDN Indonesia PER Peru UGA Uganda Indiarsquosnational averages are excluded due to the inclusion of the Indian states For Indian states incomesare the official per capita net state domestic products

98 Journal of Economic Perspectives

absence over the existing salary range Of course it is important to bear in mindthat the samples of countries and states are very small and other factors couldinfluence these slopes

Teacher and health worker absence are correlated across countries and stateseven after controlling for per capita income The residuals from the two regressionsdepicted in Figure 1 (with an additional dummy added for Indian states) are highlycorrelated with each other with a correlation coefficient of 044 (significant at the5 percent level) This correlation could potentially be due to mismeasurement ofincome but it could also reflect spillover effects in social norms across sectors oromitted variables such as the quality of governance

Concentration of Absence

To understand and potentially design policies to counter high absence ratesit is useful to know whether absences are spread out among providers or concen-trated among a small number of ldquoghost workersrdquo who are on the books but nevershow up Since our survey included only two or three observations per worker wewould observe some dispersion in absence rates even if all workers had identicalunderlying probabilities of being absent The left panel of Table 2 shows thedistribution of absence observed in the data For comparison the right panel showsthe distribution that would be observed if the probability of absence in each visitwere equal to the estimated absence rate in the specific country-sector combina-tion so all workers had the same probability of being absent For example if allteachers in Indonesia had a 019 chance of being absent (which is the averageteacher absence rate there) then on any two independent visits we would expect36 percent (019 019) to be absent both times 656 percent (081 081) to bepresent both times and the remaining 308 percent to be absent once On the otherhand if absence were completely concentrated in certain providers we wouldobserve that 19 percent of the teachers are always absent 81 percent are alwayspresent and none are absent only once

Clearly the data match neither the extreme of all workers having identicalunderlying probabilities of absence nor of all absence being due to ghost workersbut an eyeball test suggests that absence appears to be fairly widespread with theempirical distribution surprisingly close to that predicted by a model with identicalabsence probabilities Teachers in Ecuador are an exception and appear to be theleading candidates for a ldquoghost workerrdquo explanation with a very high percentage ofteachers being present in both visits and more teachers absent in both visits than inone of the two visits

The exercise above while suggestive can technically only be used to test theextreme hypotheses of complete concentration of absence and perfectly identicalabsence rates among workers Glewwe Ilias and Kremer (2004) assume providersrsquounderlying probability of absence follows a beta distribution and estimate thisdistribution in two districts of Kenya using a maximum likelihood approach They

Missing in Action Teacher and Health Worker Absence in Developing Countries 99

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 8: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

capital acquisition and thus income Another is that the overall level of develop-ment drives the quality of education and health delivery While it is impossible todisentangle these stories completely to the extent that the overall level of devel-opment influences provider absence one might expect low income levels to lead tohigh absence rates in both education and health On the other hand if educationis particularly important for human capital acquisition and thus income whilemedical clinics have a larger consumption component then exogenous variation inquality of education systems will lead to variation in income while the quality ofhealth care systems will be less correlated with income This pattern matches whatwe see in the data

It is intriguing that the relationship between income and absence is so similaracross countries and across Indian states and that it is so tight in each case Whilesalaries typically rise with GDP (although not proportionally) teacher salariesacross Indian states are relatively flat6 Thus across the states of India salaries forteachers and health workers in poor states are considerably higher relative to thecost of living and relative to workersrsquo outside opportunities than are salaries in richstates Nonetheless absence rates are higher in poor states The similarity betweenthe absence-income regression line across countries and the comparable line acrossIndian states despite the difference in the relationship between income andsalaries in the two samples suggests a limited role for salaries in influencing

6 Ministry of Human Resource Development India

Figure 1Absence Rate versus NationalState Per Capita Income

Source Authorsrsquo calculationsNote BNG Bangladesh ECU Ecuador IDN Indonesia PER Peru UGA Uganda Indiarsquosnational averages are excluded due to the inclusion of the Indian states For Indian states incomesare the official per capita net state domestic products

98 Journal of Economic Perspectives

absence over the existing salary range Of course it is important to bear in mindthat the samples of countries and states are very small and other factors couldinfluence these slopes

Teacher and health worker absence are correlated across countries and stateseven after controlling for per capita income The residuals from the two regressionsdepicted in Figure 1 (with an additional dummy added for Indian states) are highlycorrelated with each other with a correlation coefficient of 044 (significant at the5 percent level) This correlation could potentially be due to mismeasurement ofincome but it could also reflect spillover effects in social norms across sectors oromitted variables such as the quality of governance

Concentration of Absence

To understand and potentially design policies to counter high absence ratesit is useful to know whether absences are spread out among providers or concen-trated among a small number of ldquoghost workersrdquo who are on the books but nevershow up Since our survey included only two or three observations per worker wewould observe some dispersion in absence rates even if all workers had identicalunderlying probabilities of being absent The left panel of Table 2 shows thedistribution of absence observed in the data For comparison the right panel showsthe distribution that would be observed if the probability of absence in each visitwere equal to the estimated absence rate in the specific country-sector combina-tion so all workers had the same probability of being absent For example if allteachers in Indonesia had a 019 chance of being absent (which is the averageteacher absence rate there) then on any two independent visits we would expect36 percent (019 019) to be absent both times 656 percent (081 081) to bepresent both times and the remaining 308 percent to be absent once On the otherhand if absence were completely concentrated in certain providers we wouldobserve that 19 percent of the teachers are always absent 81 percent are alwayspresent and none are absent only once

Clearly the data match neither the extreme of all workers having identicalunderlying probabilities of absence nor of all absence being due to ghost workersbut an eyeball test suggests that absence appears to be fairly widespread with theempirical distribution surprisingly close to that predicted by a model with identicalabsence probabilities Teachers in Ecuador are an exception and appear to be theleading candidates for a ldquoghost workerrdquo explanation with a very high percentage ofteachers being present in both visits and more teachers absent in both visits than inone of the two visits

The exercise above while suggestive can technically only be used to test theextreme hypotheses of complete concentration of absence and perfectly identicalabsence rates among workers Glewwe Ilias and Kremer (2004) assume providersrsquounderlying probability of absence follows a beta distribution and estimate thisdistribution in two districts of Kenya using a maximum likelihood approach They

Missing in Action Teacher and Health Worker Absence in Developing Countries 99

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 9: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

absence over the existing salary range Of course it is important to bear in mindthat the samples of countries and states are very small and other factors couldinfluence these slopes

Teacher and health worker absence are correlated across countries and stateseven after controlling for per capita income The residuals from the two regressionsdepicted in Figure 1 (with an additional dummy added for Indian states) are highlycorrelated with each other with a correlation coefficient of 044 (significant at the5 percent level) This correlation could potentially be due to mismeasurement ofincome but it could also reflect spillover effects in social norms across sectors oromitted variables such as the quality of governance

Concentration of Absence

To understand and potentially design policies to counter high absence ratesit is useful to know whether absences are spread out among providers or concen-trated among a small number of ldquoghost workersrdquo who are on the books but nevershow up Since our survey included only two or three observations per worker wewould observe some dispersion in absence rates even if all workers had identicalunderlying probabilities of being absent The left panel of Table 2 shows thedistribution of absence observed in the data For comparison the right panel showsthe distribution that would be observed if the probability of absence in each visitwere equal to the estimated absence rate in the specific country-sector combina-tion so all workers had the same probability of being absent For example if allteachers in Indonesia had a 019 chance of being absent (which is the averageteacher absence rate there) then on any two independent visits we would expect36 percent (019 019) to be absent both times 656 percent (081 081) to bepresent both times and the remaining 308 percent to be absent once On the otherhand if absence were completely concentrated in certain providers we wouldobserve that 19 percent of the teachers are always absent 81 percent are alwayspresent and none are absent only once

Clearly the data match neither the extreme of all workers having identicalunderlying probabilities of absence nor of all absence being due to ghost workersbut an eyeball test suggests that absence appears to be fairly widespread with theempirical distribution surprisingly close to that predicted by a model with identicalabsence probabilities Teachers in Ecuador are an exception and appear to be theleading candidates for a ldquoghost workerrdquo explanation with a very high percentage ofteachers being present in both visits and more teachers absent in both visits than inone of the two visits

The exercise above while suggestive can technically only be used to test theextreme hypotheses of complete concentration of absence and perfectly identicalabsence rates among workers Glewwe Ilias and Kremer (2004) assume providersrsquounderlying probability of absence follows a beta distribution and estimate thisdistribution in two districts of Kenya using a maximum likelihood approach They

Missing in Action Teacher and Health Worker Absence in Developing Countries 99

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 10: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

find that although a few teachers are rarely present the majority of absences appearto be due to those who attend between 50 percent and 80 percent of the time andthe median teacher is absent 14 to 19 percent of the time The results of a similarcalibration using the multicountry data in this paper also suggest that other than inEcuador absence is typically fairly widespread rather than being concentrated ina minority of ldquoghostrdquo workers Banerjee Deaton and Duflo (2004) conducted anintensive study in Rajasthan India in which health workers were visited weekly fora year and they also find that absences are fairly widely distributed there

How Much of Absence is Authorized

It is difficult to assess the extent to which absence is authorized Enumeratorsasked the facility-survey respondentmdashgenerally the school head teacher or primaryhealth care center directormdashthe reason for each absence but facility directors maynot always answer truthfully Thus for example in India the fraction of staffreported to be on authorized leave greatly exceeded that which would be predictedgiven statutory leave allocations (Kremer et al 2004) However even taking facility

Table 2Distribution of Absences Among Providers

Percentage of providers who were absentthis many times in 2 visits

(3 visits in India)

For comparison expected distribution ifall providers had equal

absence probability

0 1 2 3 0 1 2 3

TeachersBangladesh 734 235 32 mdash 706 269 26Ecuador 828 69 104 mdash 740 241 20India 491 327 135 48 422 422 141 16Indonesia 677 275 48 mdash 656 308 36Peru 810 173 17 mdash 792 196 12Uganda 630 296 74 mdash 533 394 73

Medical workersIndia 357 319 208 116 216 432 288 64Indonesia 461 410 129 mdash 360 480 160Peru 564 335 101 mdash 563 375 63Uganda 520 380 100 mdash 397 466 137

Notes The left side of this table gives the distribution of absences observed for each type of provider ineach country For example it shows that during two survey visits 734 percent of teachers in Bangladeshprimary schools were never absent 235 percent were absent once and 32 percent were absent duringboth visits The right side of the table provides for comparison the distribution that would be expectedif all providers in a country had an identical underlying absence rate equal to the average rate observedfor that country Bangladesh health workers are excluded because the first-round survey was carried outfor a different study making it impossible to match workers across rounds and show the empiricaldistribution

100 Journal of Economic Perspectives

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 11: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

directorsrsquo responses at face value it seems clear that two categories of sanctionedabsencemdashillness and official duties outside of health and educationmdashdo notaccount for the bulk of absence

Across countries illness is the stated cause of absence in 2 percent of teacherobservations and 14 percent for health worker observations (in other words itaccounts for around 10 percent of teacher absence and 4 percent of health workerabsence) Two countries of particular interest here are Uganda and Zambia whereHIV infection is prevalent However preliminary analysis by Habyarimana (2004)suggests that neither the demographic nor the geographic distribution of teacherabsences in Uganda correlates very well with what is known about patterns of HIVprevalence Uganda does not appear to be an outliermdashthat is it does not appear tohave much more absence than would be expected given its income levels In thecase of Zambia where HIV prevalence is high Das Dercon Habyarimana andKrishnan (2005) suggest that the disease may explain a large share of teacherabsence and attrition Interestingly however the absence rate they estimate forZambia is 17 percentmdashwhich is much less than predicted by the absence-incomerelationship we estimate across countries7

Some argue that teacher absence is high in South Asia because governmentspull teachers out of school to carry out duties such as voter registration electionoversight and public health campaigns But head teachers should have little reasonto underreport such absences and in India only about 1 percent of observations(4 percent of absences) are attributed to non-education-related official duties(Kremer et al 2004)

Correlates of Teacher Absence

What factors are correlated with teacher absence Although our sample in-cludes both low- and middle-income countries on three continents certain com-mon patterns emerge as shown in Table 3 The dependent variable is absencecoded as 100 if the provider was absent on a particular visit and 0 if he or she waspresent All regressions include district fixed effects To obtain estimates of averagecoefficients for the sample as a whole we use hierarchical linear model estimationin which a combined coefficient is estimated by averaging the coefficients fromordinary least squares regressions of absence in each of the countries weighted inaccordance with the precision with which they are estimated8 (By contrast apooled ordinary least squares regression with interaction terms for country-specific

7 Although the Zambia study follows a methodology similar to those reported in this article it wascarried out by a different team using a different survey instrument so the results may not be strictlycomparable8 The error terms are clustered at the school level throughout this analysis Results using probits aresimilar A good reference for hierarchical linear model estimation and inference is Raudenbusch andBryk (2002)

Nazmul Chaudhury et al 101

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 12: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

effects would be swamped by India since we have so many more observationsthere) At the risk of oversimplifying the heterogeneity across countries we willfocus primarily here on the results for the sample as a whole However the finalcolumn indicates the heterogeneity across countries by indicating which of thecountry-specific regressions yielded a coefficient with the same sign and whether itwas statistically significant (Tables showing the regression results for each country

Table 3Correlates of Teacher Absence (HLM with District-Level Fixed Effects)(dependent variable visit level absence of a given teacher 0 present 100 absent)

Estimates for themulticountry sample

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 1942 0509 BNG ECU IND IDN PEREver received training 2141 4354 BNG ECU PERUnion member 2538 1258 ECU IND IDN PERBorn in district of school 2715 0833 BNG ECU IND IDN PER UGReceived recent training 0740 2070 BNG ECU UGATenure at school (years) 0033 0044 BNG IDN PERAge (years) 0021 0046 ECU IND UGAMarried 0742 0972 BNG IDN PER UGAHas university degree 1055 1162 ECU IDNHas degree in education 1806 2071 ECU INDHead teacher 3771 0888 BNG ECU IND IDN PER UGASchool infrastructure index

(0ndash5)2234 0438 BNG ECU IND IDN PER

School inspected in last 2 mos 0142 1194 BNG ECU IND UGASchool is near Min Education

office4944 2642 BNG ECU IND IDN

School had recent PTAmeeting

2308 1576 BNG ECU PER

Schoolrsquos pupil-teacher ratio 0095 0080 BNG ECU IDN PERSchoolrsquos number of teachers 0015 0113 ECU PER UGASchool has teacher recognition

program0168 3525 ECU PER

Studentsrsquo parentsrsquo literacy rate(0ndash1)

9361 1604 BNG ECU IND IDN PER

School is in urban area 2039 1441 ECU IND PERSchool is near paved road 0040 1106 BNG ECU IDN UGATeacher is contract teacher 5722 2906 ECU IDN PER (no contract teachers in

BNGUGA)Dummy for 1st survey round 2938 1874 BNG ECU IND PER UGAConstant 32959 1963 BNG ECU IND IDN PER

UGAObservations 34880

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Regressions alsoincluded dummies for the days of the week (not reported here)

102 Journal of Economic Perspectives

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 13: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

using the same specification are available appended to this article at the httpwwwe-jeporg website)

Teacher CharacteristicsIn most countries salaries are highly correlated with the teacherrsquos age expe-

rience educational background (such as whether the teacher has a universitydegree or a degree in education) and rank (such as head teacher status) Table 3provides little evidence to suggest that higher salaries proxied by any of thesefactors are significantly associated with lower absence Head teachers are signifi-cantly more likely to be absent and point estimates suggest better-educated andolder teachers are on average absent more often Of course it is possible that otherfactors confound the effect of teacher salary in the data for example if the outsideopportunities for teachers increase faster than their pay within the government paystructure the regression results presented here could be misleading

However the earlier discussion on cross-state variation in relative teacherwages in India provides another source of data on the impact of teacher salariesthat is not subject to this difficulty If higher salaries relative to outside opportuni-ties or prices led to much lower absence then one might expect absence to rise withstate income in India (because salaries relative to outside opportunities are lowerin richer states) or at least not to fall as quickly as in the cross-country data In factthey fall at the same rate as in cross-country data

The coefficients on teacher characteristics suggest that along a number ofdimensions more powerful teachers are absent more Men are absent more oftenthan women and head teachers are absent more often than regular teachers In anumber of cases better-educated teachers appear to be absent more These teach-ers may be less subject to monitoring

A degree in education is strongly negatively associated with absence in Bang-ladesh and Uganda but the association is positive in Ecuador In-service training isnegatively associated with absence in three countries but not in the global analysisMoreover recent training is not associated with reduced absence other than inEcuador The negative coefficient in Ecuador could be due to ldquoghost teachersrdquo whoattend neither schools nor training sessions

Theoretically teachers from the local area might be expected to be absent lessbecause they care more about their students or are easier to monitor or absentmore because they have more outside opportunities in the local economy and areharder to discipline with sanctions Empirically we find that teachers who wereborn in the district of the school are more likely to show up for work Local teachersare less likely to be absent in all six countries (two of them at statistically significantlevels) and the coefficient for the combined sample is also significantly negative

This result is robust to including school dummies suggesting that we areobserving a local-teacher effect rather than just perhaps something related to thecharacteristics of schools located in areas that produce many teachers Whileteachers born in the area are absent less there is no significant correlation between

Missing in Action Teacher and Health Worker Absence in Developing Countries 103

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 14: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

another possible measure of the teacherrsquos local tiesmdashthe duration of a teacherrsquosposting at the schoolmdashand teacher presence (except in Uganda)

School CharacteristicsWorking conditions can affect incentives to attend school even where receipt

of salary is independent of attendance and hence provides no such incentive Weconstructed an index measuring the quality of the schoolrsquos infrastructuremdasha sumof the five dummies measuring the availability of a toilet (or teachersrsquo toilet inIndia) covered classrooms nondirt floors electricity and a school library Theanalysis for the sample as a whole suggests that moving from a school with thelowest infrastructure index score to one with the highest (that is from a score ofzero to five) is associated with a 10 percentage point reduction in absence A onestandard-deviation increase in the infrastructure index is associated with a27 percentage-point reduction in absence If frequently absent teachers can bepunished by assigning them to schools with poorer facilities then the interpreta-tion of the coefficient on poor infrastructure becomes unclear To address thispossibility we also examine Indian teachers on their first posting because in Indiaan algorithm typically matches new hires to vacancies Even in this sample there isa strong negative relationship between infrastructure quality and absence

MonitoringThe lower teacher absence rate in the second survey round provides support

for the idea that monitoring could affect absence If even the presence of surveyenumerators with no power over individual teachers had an impact on absence itis plausible that formal inspections would also have such an impact

We examine two measures of the intensity of administrative oversight byMinistry of Education officials a dummy representing inspection of the schoolwithin the previous two months and a dummy representing proximity to thenearest office of the ministry while controlling for other measures of remotenesslike whether the school is near a paved road9 If ldquobadrdquo schools are more likely to getinspected the coefficient on inspections will be biased upwards On the otherhand if factors other than those we control for make schools more attractive bothto teachers and to inspectors the coefficient could be biased downward Having arecent inspection is significantly associated with lower teacher absence in India butnot in the other countries nor for the sample as a whole However the coefficienton proximity to the ministry office is somewhat more robust In three of the sixcountries schools that are closer to a Ministry of Education office have significantlylower absence even after controlling for proximity to a paved road in no countryare they significantly more often absent Of course proximity to the ministry could

9 The proximity variables in these regressionsmdashproximity to roads and to ministry officesmdashare definedslightly differently in each country Because of the great differences in population density in somecountries a road or office may be counted as ldquocloserdquo if it is within five kilometers whereas in othercountries the cutoff is 15 kilometers

104 Journal of Economic Perspectives

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 15: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

proxy for other types of contract with the ministry or for closeness to otherdesirable features of district headquarters

Past studies have suggested that local control of schools may be associated withbetter performance by teachers (King and Ozler 2001) One measure of thedegree of community involvement in the schools in our dataset is the activity levelof the Parent Teacher Association (PTA) As Table 3 shows there is not a signifi-cant correlation between absence and whether the PTA has met in the previous twomonths

Community CharacteristicsTeachers are less frequently absent in schools where the parental literacy rate

is higher The coefficient on school-level parental literacy is highly significantlynegative for the sample as a whole as Table 3 shows each 10-percentage-pointincrease in the parental literacy rate reduces predicted absence by more than onepercentage point The correlation may be due to greater demand for educationmonitoring ability or political influence by educated parents more pleasant work-ing conditions for teachers (if children of literate parents are better prepared ormore motivated) selection effects with educated parents abandoning schools withhigh absence or favorable community fixed characteristics contributing to bothgreater parental literacy and lower teacher absence

The location of the community might also be thought to play a role in absenceand in India Indonesia and Peru schools in rural communities do in fact havesignificantly higher mean absence rates than do urban schools by an average ofalmost 4 percentage points (In the other countries the difference is not signifi-cant) But the dummies for whether a school is in an urban area and is near a pavedroad are both insignificant in all countries after controlling for other characteristicsof rural schools such as poor infrastructure These variables might have offsettingeffects on teacher absence because being in an urban area or near a road mightmake the school a more desirable posting but these factors could also make iteasier for providers to live far from the school or pursue alternative activities(Chaudhury and Hammer 2003)

Alternative Institutional FormsA number of alternative institutional forms have appeared in reaction to

dissatisfaction with the cost and quality of existing education institutions Theseinclude hiring contract teachers in regular government schools establishingcommunity-run nonformal education centers and using low-cost private schoolsAdvocates argue that such systems not only are much cheaper but also deliverbetter results We discuss evidence on absence below

Four of the six countries we examine make some use of contract teachers intheir primary school systems It has been hypothesized that these contract teacherswhose tenure in the teaching corps is not guaranteed may feel a stronger incentiveto perform well than do civil-servant teachers On the other hand contract teachersoften earn much less than civil servants in India for example public-school

Nazmul Chaudhury et al 105

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 16: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

contract teachers typically earn less than a third of the wages of regular teachersand in Indonesia nonregular teachers under different types of contracts earnbetween a tenth and a half as much as regular teachers In Ecuador by contrastcontract teachers appear to earn compensation similar to that of regular teachersbut without the same job security (Rogers et al 2004) Moreover the lack of tenurefor contract teachers could increase incentives to divert effort to searching forother jobs Empirically we find that contract teachers are much more likely to beabsent than other teachers in Indonesia and that in two other countries and in thecombined sample the coefficient is positive but is not statistically significant Vegasand De Laat (2003) find that in Togo contract teachers are absent at about thesame rate as civil-service teachers

Many argue that local control will bring greater accountability to teachers andhealth workers Nonformal education centers have been created by state govern-ments in India in areas with low population density that have too few students tojustify a full school with the aim of ensuring a school exists within a one-kilometerradius of every habitation These schools typically have a teacher or two from thelocal community who are not civil-service employees and are paid through grantsmade by the government to locally elected community bodies The teachers areemployed on fixed-term contracts that are subject to renewal by these bodies Oursample in India has 87 such schools and 393 observations on teachers in thesenonformal education centers We find that absence rates in the nonformal educa-tion centers are higher (28 percent) than in regular government-run schools (25percent) though this difference is not significant at the 10 percent level Thedifference remains statistically insignificant even after including village fixed effectsand other controls (as shown in Table 4)

Finally we examine private schools and private aided schools in Indian villageswith government schools Opposing forces are also likely at work in determiningwhether private-school teachers have higher or lower attendance rates than public-school teachers On the one hand private-school teachers often earn much lowerwages than do public-school teachers in India for example regular teachers inrural government schools typically get paid over three times more than theircounterparts in the rural private schools10 On the other hand private-schoolteachers face a greater chance of dismissal for absence In India 35 out of 600private schools reported a case of the head teacher dismissing a teacher forrepeated absence or tardiness compared to (as noted earlier) one in 3000 ingovernment schools in India

Empirically we find the absence rate of Indian private-school teachers is onlyslightly lower than that of public-school teachers However private-school teachersare 4 percentage points less likely to be absent than public-school teachers working

10 We calculate the total revenue of each private school based on total fees collected and find that evenif all the revenue was used for teacher salaries the average teacher salary in private schools would bearound 1600 rupees per month whereas the average public school teacherrsquos salary is around Rs 5000per month

106 Journal of Economic Perspectives

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 17: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

in the same village and 8 percentage points less likely to be absent after controllingfor school and teacher variables as shown in Table 4 This pattern arises becauseprivate schools are disproportionately located in villages that have governmentschools with particularly high absence rates Advocates of private schools mayinterpret the correlation between the presence of private schools and weakness ofpublic schools as suggesting that private schools spring up in areas where govern-ment schools are performing particularly badly opponents could counter that theentry of private schools leads to exit of politically influential families from thepublic school system further weakening pressure on public-school teachers toattend school

Private aided schools in India are privately managed but the government paysthe teacher salaries directly These teachers are government employees and enjoyfull civil service protection They thus represent an alternative institutional formwith private management but public regulation Raw absence rates in these schoolsare significantly lower than those in government-run public schools but there is nosignificant difference controlling for village fixed effects as shown in Table 4Overall our results suggest that while the alternative institutional forms are oftenmuch cheaper than government schools staffed by teachers with civil serviceprotection teacher absence is no lower in any of the publicly funded models InIndia private-school teachers do have lower absence than public school teachers inthe same village

Correlates of Absence among Health Workers

One important difference between absence in health and education is thathealth workers who are absent from public clinics seem more likely to be providingprivate medical care than absent teachers are to be offering private tuition In the

Table 4Absence Rate by School Type (India Only)

Teacherabsence

(unweighted)Number of

observations

Difference relative to government-run schools

Samplemeans

Regression withvillagetownfixed effects

Regression withvillagetownfixed effects controls

Government-run schools 245 34525 mdash mdash mdashNonformal schools 280 393 35 27 24Private aided schools 191 3371 54 13 04Private schools 252 9098 07 38 78

Notes Controls include a full set of visit-level teacher-level and school-level controls Significantdifferences are indicated by and for significances at 1 5 and 10 percent

Missing in Action Teacher and Health Worker Absence in Developing Countries 107

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 18: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

sample countries for which we have data on this question (India is excluded) an(unweighted) average of 41 percent of health workers say they have a privatepractice Actual numbers may be even higher since moonlighting is technicallyillegal in some countries By contrast while private tutoring is common in somecountries and among middle class urban pupils particularly at the secondary levelsit does not appear to be a major activity for the primary school teachers in oursample in which only about 10 percent of our sample teachers report holding anyoutside teaching or tutoring job

Table 5 shows correlates of absence among health workers Again the depen-dent variable is absence coded as 100 if the provider was absent on a particular visitand 0 if he or she was present As in the education sector the estimation incorpo-rates district fixed effects and uses hierarchical linear modeling

Health Worker CharacteristicsOf the individual health worker characteristics in our regressions the only one

that significantly and robustly predicts absence is the type of medical worker In

Table 5Correlates of Health Worker Absence (HLM with District-Level Fixed Effects)(dependent variable visit-level absence of a given HC staff member 0 present100 absent)

Estimates from themulticountry sample(excl Bangladesh)

Countries where coefficient has samesign as multicountry coefficientCoefficient

Standarderror

Male 0628 1475 INDTenure at facility (years) 0081 0382 IDN PERTenure at facility squared 0008 0011 IDN PERBorn in PHCrsquos district 1404 0873 BNG IDNDoctor 3380 0754 BNG IND IDN PER UGAWorks night shift 4267 1066 BNG IND IDN PER UGAConducts outreach 6617 0620 IND IDN PERLives in PHC-provided housing 0583 1507 BNG IDN PER UGAPHC was inspected in last 2 mos 1975 0624 BNG IND IDN PER UGAPHC is close to MOH office 0768 1999 BNG INDPHC has potable water 3352 0844 BNG IND IDNPHC is close to paved road 6076 3042 IND IDN PERDummy for 1st survey round 12457 11180 IDN PER UGAConstant 38014 1538 BNG IND IDN PER UGAObservations 27894

Notes Significant at 10 percent significant at 5 percent significant at 1 percentRegressions and HLM estimation also included dummies for days of the week (not reported here)Where applicable regressions also included dummies for urban area (Peru) and for type of clinic(Bangladesh India) Bangladesh is excluded from HLM because matching across the two survey roundswas not possible as first-round data are drawn from a separate survey

108 Journal of Economic Perspectives

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 19: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

every country doctors are more often absent than other health care workers andthe difference is significant in three countries and in the multicountry regressionDoctors have a marketable skill and lucrative outside earning capabilities at privateclinics In Peru for example 48 percent of doctors reported outside income fromprivate practice much higher than the 30 percent of nondoctor medical workers

Facility-Level VariablesHealth providers are less likely to be absent where the public health clinic was

inspected within the past two months in every country and the relationship issignificant at the 10 percent level in the combined sample Being close to a Ministryof Health office is (insignificantly) positively correlated with absence in the com-bined sample although it is correlated with lower absence in Indonesia

In India we find that for medical providers other than doctors attendance atlarger classes of facilities (community health centers) is much higher than insmaller subcenters where no doctor (and therefore no one of higher status) isassigned One interpretation is that doctors play a role in monitoring other healthcare workers Another interpretation is that primary health centers are in moreremote less attractive localities

In terms of working conditions the availability of potable water predicts lowerabsence at a statistically significant level in the combined sample as well as in IndiaIndonesia and Uganda However whether the public health clinic has toilets is notcorrelated with absence in any country

Another aspect of working conditions the logistics of getting to work and thedesirability of the primary health care centersrsquo location is also correlated withabsence in some countries In Bangladesh and Uganda providers who live inprimary health care center-provided housing (which is typically on primary healthcare centersrsquo premises) have much lower absence although this coefficient was notstatistically significant in the global sample In Indonesia although not in theglobal sample primary health care centers located near paved roads have muchlower absence rates

Providers who work the night shift were less likely to be absent for theirdaytime shifts Given the usually voluntary and episodic nature of night shifts thisvariable may proxy for intrinsic motivation Alternatively it is possible that nightshifts are assigned to less influential employees who are less likely to get away withabsence

Alternative Institutional FormsIn our sample there are no private medical facilities and we have data on

contract employment of medical personnel only in Peru In that countrycontract work is strongly associated with lower absence despite the fact that liketheir civil-service counterparts contract medical personnel are paid on salaryrather than on a fee-for-service basis This result is consistent with previousfindings on absence among Peruvian hospital personnel (Alcazar and Andrade2001)

Nazmul Chaudhury et al 109

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 20: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

Efficiency of Absence

While 19 percent absence among teachers and 35 percent absence amonghealth workers is clearly undesirable it is worth asking two questions to investigatethe extent to which this level of absence is a distributional issue an efficiency issueor both First are teachers and health care workers earning rents beyond what theywould obtain outside the public sector in the sense that the package of pay andactual work requirements is significantly more attractive than what these workerscould obtain in the private sector Because service providers (especially doctors)are typically better off than average any policy that results in taxpayer-funded rentsfor them will generally be regressive Second taking the value of the overallpackage of wages and perks for teachers and health workers as fixed is it efficientfor them to be compensated in part through toleration of absence

It seems clear that many primary school teachers in developing countries earnrents In India for example public-school teachers earn much more than theircounterparts either in the private sector or among contract teachers hired by thepublic sector and qualified applicants form long queues to be hired as governmentteachers Many health workers may also be earning rents but for high-skilled healthcare providers doctors in particular the case is not clear It seems possible that ifdoctorsrsquo wages were kept constant but they were prohibited from being absentmany would quit and enter private practice or even migrate to richer countries

In their intensive study of medical providers in rural Rajasthan BanerjeeDeaton and Duflo (2004) find evidence suggesting absence is inefficiently high inthe case of nurses who staff the smaller health subcenters They argue that efficientabsence would require facilities to be open on a fixed schedule so patients wouldknow when it was worth their while to travel to the clinic They find however thatfacilities are open at unpredictable times Of course it is hypothetically possiblethat clients know when providers are available or how to find them even ifresearchers cannot discern a pattern It is harder to prove inefficiency for high-skillhealth workers One interpretation of high absence rates among skilled healthworkers is that the government is paying them to locate in an undesirable rural areaand to spend part of their day serving poor patients at public facilities11 Inexchange the implicit contract between the government and providers allowsproviders to work privately during the rest of the day It is possible that this outcomerepresents fairly efficient price discrimination with the poor receiving care ingovernment facilities and the better-off seeing doctors privately In our datamedical personnel who ask to be posted in a particular place are absent less oftenwhich could be interpreted as consistent with the view that absence rates representa compensating differential

However it seems unlikely that the most efficient way to implement a contract

11 Chomitz et al (1999) find that many Indonesian doctors would require enormous pay premiums tobe willing to accept postings to islands off Java

110 Journal of Economic Perspectives

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 21: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

that allowed doctors to work part-time for the government would be through asystem in which providers were formally required to be present full-time but theseregulations were not enforced It is also not completely clear what public policygoals are served by subsidizing many types of curative care in rural areas to such anextent In the typical clinic in Peru for example only about two patients were seenper provider hour This ratio seems fairly low with health care being very expensiveto provide in these areas

In the case of education it is possible to reject the efficient absence hypothesiseven more definitively A necessary (but of course not sufficient) condition forhigh rates of teacher absence to be efficient is that teacher and student absence ineach school be highly correlated over time In fact as discussed further in Kremeret al (2004) the correlation is not that high students frequently come to schoolonly to find their teachers absent

Political Economy of Absence

An important proximate cause of absence among civil servant teachers andhealth workers is the weakness of sanctions for absence as indicated by ouruncovering only one case of a teacher being fired for absence in 3000 headmasterinterviews in India Technical means for monitoring absence do exist For exampleheadmasters could be required to keep good teacher attendance records and couldbe demoted if inspectors find their records are inaccurate Such rules are typicallyon the books but are not enforced Duflo and Hanna (2005) show that requiringteachers at nonformal education centers to take daily pictures of themselves andtheir students to qualify for bonuses can dramatically improve teacher attendanceand student learning In some of the countries we examine teacher and healthworker absence was reportedly less of an issue during the colonial period Absencehas reportedly also been reportedly low in some authoritarian countries such asCuba under Castro or Korea under Park although such claims are difficult toverify

Why doesnrsquot the political system generate demands for stronger supervision ofproviders Most of the countries in our sample are either democratic or havesubstantial elements of democracy Yet provider absence in health and education isnot a major election issue Apparently politicians do not consider campaigning ona platform of cracking down on absent providers to be a winning electoral strategy

One possible reason why provider absence is not on the political agenda is thatproviders are an organized interest group whereas clients particularly in healthare diffuse Those poor enough to use public schools and public clinics have lesspolitical power than middle class teachers and health workers In many countrieseven those who are moderately well off send their children to private schools anduse private clinics This pattern may create a self-reinforcing cycle of low qualityexit of the politically influential from the public sector and further deterioration ofquality (Hirschman 1970)

Missing in Action Teacher and Health Worker Absence in Developing Countries 111

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 22: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

The centralization of education and health systems in most developingcountries may contribute to weak accountability Voters in a particular electoralconstituency selecting a member of parliament may prefer that their representa-tives use their political influence to obtain a greater share of education funds fortheir constituencymdashfor example by building new schools theremdashrather than inimproving the overall quality of the system The free-rider problem among politi-cians would be ameliorated if policy were set in smaller administrative units

But moving from a formal civil service system to control by local elected bodieswould come at a price In the civil service system in place in the countries we examineproviders have weak incentives but the opportunity for corruption by politicians issomewhat limited If local elected bodies provided oversight teachers would havestronger incentives but local politicians would also have greater opportunity to appointfriends cronies or members of favored ethnic or religious groups

Disentangling the many features of civil service systems may be difficult Ifteachers are to be paid on a common pay scale many will earn substantial rentsHeterogeneity in local labor market conditions and in the compensating differen-tials needed to attract skilled personnel to different regions will typically be greaterin developing countries than in developed countries Since education employs agreater proportion of the educated labor force in developing countries thandeveloped countries heterogeneity in skill levels among this group will almostcertainly be greater than in developed countries Once a system is in place in whichmany teachers earn above-market wages there will be pressures for strong civilservice protection to protect those rents In the absence of such civil serviceprotection those with the right to hire and fire teachers will be able to extract rentsfrom those teachers who would otherwise receive them It is therefore understand-able that even teachers who do not personally expect to be absent often would favorcivil service rules that make it difficult for inspectors or headmasters to fireteachers Once such rules are in place those teachers who want to be absent areable to do so and this may contribute to a culture of absence This could create amultiplier effect by influencing norms potentially creating a culture of absence(Basu 2004)

Conclusion

With one in five government primary-school teachers and more than a third ofhealth workers absent from their facilities developing countries are wasting con-siderable resources and missing opportunities to educate their children and im-prove the health of their populations Even these figures may understate theproblem since many providers who were present in their facilities may not bedelivering services Our results complement a large recent literature that argues thatcorruption and weak institutions in developing countries reduce private investmentand thus growth Poorly functioning government institutions may also impair provi-sion of education and health Reduced levels of education and health could substan-

112 Journal of Economic Perspectives

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 23: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

tially reduce long-run growth as well as short-run welfare since public human capitalinvestment accounts for a large fraction of total investment in many countries

Faced with high absence rates policymakers have two challenges How caneducation and health policy be adapted to minimize the cost of absence How canabsence be reduced

On the first point policies in education and health should be designed totake into account high absence rates For instance doctor absence may bedifficult to prevent but possible to work around Very high salaries (combinedwith effective monitoring) may be required to induce well-trained medicalpersonnelmdash doctors in particularmdashto live in rural areas where they will find fewother educated people and where educational opportunities for their childrenwill be limited To conserve on the permanently posted rural workers whoexhibit such high absence rates health policy might shift budgets towardactivities that do not require doctors to be posted to remote areas This couldinclude immunization campaigns vector (pest) control to limit infectious dis-ease health education providing safe water and providing periodic doctor visitsrather than continuous service (Filmer Hammer and Pritchett 2000 2002)Doctors could be used in hospitals and where medical personnel are likely toattend work more regularly (World Bank 2004) and governments or nongov-ernment organizations could make efforts to reduce the cost of getting patientsto towns and hospitals

On the second pointmdashhow to reduce absencemdashour results can provide onlytentative guidance Conceptually there seem to be three broad strategies formoving forward One approach would be to increase local control for example bygiving local institutions like school committees new powers to hire and fire teach-ers However the high absence rates among contract teachers in several countriesand among teachers in community-controlled nonformal education centers inIndia suggest that these alternative contractual forms alone may not solve theabsence problem

The second approach would be to improve the existing civil service systemIn Ecuador for example identifying and eliminating ghost teachers could go along way More generally our analysis suggests a range of possible interventionsthat might be worth testing Some such as upgrading facility infrastructure andconstructing housing for doctors would involve extra budget outlays but wouldnot require politically difficult fundamental changes in systems Others such asincreasing the frequency and bite of inspections could be implemented usingexisting rules already on the books More politically difficult may be changes inincentive structures In the accompanying article in this journal Banerjee andDuflo review evidence from a number of randomized evaluations of incentiveprograms linked to teacher attendance and to student performance Howeveras discussed above teachers and health workers are likely to be particularlyresistant to approaches that leave lots of room for discretion by those imple-menting the system for fear that attempts to reduce absence may unfairlypunish teachers who are victims of circumstances or leave discretion in the

Nazmul Chaudhury et al 113

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 24: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

hands of those who may use it for private benefit Technical approachesallowing objective monitoring of teacher attendance such as the camera mon-itoring system explored by Duflo and Hanna (2005) may hold promise if theycan help assure teachers and health workers that those who are not frequentlyabsent will not be unfairly subject to sanction

The final approach would be to experiment more with systems in whichparents choose among schools and public money follows the pupils This choicecould either be within the public system or could encompass private schools Asimilar approach could be employed in health with money following patients asopposed to facilities

It is unclear whether political pressure will occur for any of these reformsThere is some evidence that surveys that monitor and publicize absence levelssuch as surveys we conducted can focus policymakersrsquo attention on the issuemdasheven if the problem of absence is already well known to students and clinicpatients In Bangladesh for example the Ministry of Health cracked down onabsent doctors after newspaper reports highlighted the results of the healthsurvey described in this paper (ldquo24 of 28 Docs Shunted Outrdquo 2003) This typeof one-time crackdown may not necessarily be effective but the providerabsence problem documented here clearly warrants greater attention frompolicymakers and civil society

Excessive absence of teachers and medical personnel is a direct hindrance tolearning and health improvements especially for poor people who lack alterna-tives But provider absence is also symptomatic of broader failures in ldquostreet-levelrdquoinstitutions and governance Until recently these failures have received much lessattention from development thinkers and policymakers than have weaknesses inmacro institutions like democracy and high-level governance Yet for many peoplea countryrsquos success at economic and social development will be defined by whetherit can improve the quality of these day-to-day transactions between the public andthose delivering public services whether they are teachers doctors or policeofficers In service delivery quality starts with attendance

y We are grateful to the many researchers survey experts and enumerators who collaboratedwith us on the country studies that made this global cross-country paper possible We thankSanya Carleyolsen Julie Gluck Anjali Oza Mona Steffen and Konstantin Styrin for theirinvaluable research assistance We are especially grateful to the UK Department for Interna-tional Development for generous financial support and to Laure Beaufils and Jane Haycockof DFID for their support and comments We thank the Global Development Network foradditional financial assistance as well as the editors of this journal and various seminarparticipants for their many helpful suggestions We are grateful to Jishnu Das and co-authorsfor allowing us to replicate their student assessments to Jean Dregraveze and Deon Filmer forsharing survey instruments to Eric Edmonds for detailed comments and to Shanta Devarajanand Ritva Reinikka for their consistent support The findings interpretations and conclusionsexpressed here are entirely those of the authors and they do not necessarily represent the viewsof the World Bank its executive directors or the countries they represent

114 Journal of Economic Perspectives

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 25: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

References

Alcazar Lorena and Raul Andrade 2001 ldquoIn-duced Demand and Absenteeism in PeruvianHospitalsrdquo in Diagnosis Corruption Rafael DiTella and William D Savedoff eds WashingtonDC Inter-American Development Bankpp 123ndash62

Alcazar Lorena F Halsey Rogers NazmulChaudhury Jeffrey Hammer Michael Kremerand Karthik Muralidharan 2005 ldquoWhy areTeachers Absent Probing Service Delivery inPeruvian Primary Schoolsrdquo Unpublished paperWorld Bank and GRADE Peru

Banerjee Abhijit Angus Deaton and EstherDuflo 2004 ldquoWealth Health and Health Ser-vices in Rural Rajasthanrdquo American Economic Re-view 942 pp 326ndash30

Basu Kaushik 2004 ldquoCombating Indiarsquos Tru-ant Teachersrdquo BBC News World Edition Novem-ber 29 Available at httpnewsbbccouk2hisouth_asia4051353stm

Begum Sharifa and Binayak Sen 1997 ldquoNotQuite Enough Financial Allocation and the Dis-tribution of Resources in the Health SectorrdquoWorking Paper No 2 HealthPoverty InterfaceStudy BIDSWHO

Bruns Barbara Alain Mingets and RamahatraRakotomalala 2003 ldquoAchieving Universal Pri-mary Education by 2015 A Chance for EveryChildrdquo World Bank

Chaudhury Nazmul and Jeffrey S Hammer2003 ldquoGhost Doctors Doctor Absenteeism inBangladeshi Health Centersrdquo World Bank PolicyResearch Working Paper No 3065

Das Jishnu Stefan Dercon James Habyari-mana and Pramila Krishnan 2005 ldquoTeacherShocks and Student Learning Evidence fromZambiardquo Working paper World Bank

Ehrenberg Ronald G Daniel I Rees and EricL Ehrenberg 1991 ldquoSchool District Leave Poli-cies Teacher Absenteeism and StudentAchievementrdquo Journal of Human Resources 261pp 72ndash105

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2000 ldquoWeak Links in the Chain ADiagnosis of Health Policy in Poor CountriesrdquoWorld Bank Research Observer 152 pp 199ndash224

Filmer Deon Jeffrey S Hammer and Lant HPritchett 2002 ldquoWeak Links in the Chain II APrescription for Health Policy in Poor Coun-triesrdquo World Bank Research Observer 171 pp 47ndash66

Glewwe Paul Michael Kremer and SylvieMoulin 1999 ldquoTextbooks and Test Scores Evi-

dence from a Prospective Evaluation in KenyardquoWorking paper Harvard University

Habyarimana James 2004 ldquoMeasuring andUnderstanding Teacher Absence in UgandardquoUnpublished paper Georgetown University

Hirschman Albert O 1970 Exit Voice andLoyalty Responses to Decline in Firms Organizationsand States Cambridge Mass Harvard UniversityPress

King Elizabeth M and Berk Ozler 2001ldquoWhatrsquos Decentralization Got To Do With Learn-ing Endogenous School Quality and StudentPerformance in Nicaraguardquo World Bank

King Elizabeth M Peter F Orazem and Eliz-abeth M Paterno 1999 ldquoPromotion with andwithout Learning Effects on Student DropoutrdquoWorld Bank

Kingdon Geeta Gandhi and Mohd Muzammil2001 ldquoA Political Economy of Education in In-dia I The Case of UPrdquo Economic and PoliticalWeekly August 3632 pp 3052ndash063

Kremer Michael Karthik MuralidharanNazmul Chaudhury Jeffrey Hammer and F Hal-sey Rogers 2004 ldquoTeacher Absence in IndiardquoWorld Bank

Pandey Priyanka 2005 ldquoService Delivery andCapture in Public Schools How Does HistoryMatter and Can Mandated Political Representa-tion Reverse the Effect of Historyrdquo MimeoWorld Bank

Pratichi Education Team 2002 ldquoThe Deliveryof Primary Education A Study in West BengalrdquoPratichi New Delhi

Pritchett Lant H and Deon Filmer 1999ldquoWhat Educational Production Functions ReallyShow A Positive Theory of Education Spend-ingrdquo Economics of Education Review 182 pp 223ndash39

PROBE Team 1999 Public Report on Basic Ed-ucation in India New Delhi Oxford UniversityPress

Raudenbusch Stephen W and Anthony SBryk 2002 Hierarchical Linear Models Applica-tions and Data Analysis Methods Thousand OaksCalif Sage Publications

Rogers F Halsey Jose Roberto Lopez-CalixNancy Cordoba Nazmul Chaudhury JeffreyHammer Michael Kremer and Karthik Mu-ralidharan 2004 ldquoTeacher Absence and Incen-tives in Primary Education Results from a NewNational Teacher Tracking Survey in Ecuadorrdquoin Ecuador Creating Fiscal Space for Poverty Reduc-tion Washington DC World Bank chapter 6

Sen Binayak 1997 ldquoPoverty and Policyrdquo in

Missing in Action Teacher and Health Worker Absence in Developing Countries 115

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 26: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

Growth or Stagnation A Review of Bangladeshrsquos De-velopment 1996 Rehman Shoban ed DhakaCenter for Policy Dialogue and the University ofDhaka Press Ltd pp 115ndash60

ldquo24 of 28 Docs Shunted Out for Absence DGHealth Surprised at Surprise Visit to NICVDrdquo2003 Daily Star October 2 4128 p A1

Vegas Emiliana and Joost De Laat 2003 ldquoDoDifferences in Teacher Contracts Affect Student

Performance Evidence from Togordquo WorldBank

World Bank 2003 World Development Report2004 Making Services Work for Poor People Wash-ington DC Oxford University Press for theWorld Bank

World Bank 2004 ldquoPapua New Guinea Pub-lic Expenditure and Service Deliveryrdquo WorldBank

116 Journal of Economic Perspectives

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 27: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

Table A-1Teachers Mean Differences in Absence Rate by Selected Characteristics

Bangladesh Ecuador India Indonesia Peru Uganda

Male 06 03 52 38 40 14Received training 31 90 126 56 07 137Union member 06 36 56 03 15 24Born locally 03 54 42 27 25 45Received recent training 09 54 30 15 19 91Longer-term employee 03 13 37 06 00 56Older than median 01 16 61 35 11 86Married 95 09 120 10 08 80Contract teacher mdash 60 05 63 69 mdashHas bachelorrsquos diploma 92 32 01 01 36 193Has degree in education 89 00 134 60 73 74Head teacher 26 17 71 94 124 213School inspected recently 39 53 45 37 27 58School is near Ministry of

Education office49 44 13 110 07 74

School had recent PTAmeeting

01 81 48 12 22 31

Studentsrsquo parents have highliteracy rate

33 80 48 63 21 17

School has goodinfrastructure

19 24 82 20 57 32

School is near paved road 05 72 69 05 111 10School has high pupil-

teacher ratio56 74 07 14 09 28

School is in urban area 29 19 23 30 61 32School is large 57 16 32 39 25 05School has teacher

recognition program11 57 36 07 30 46

Notes Significant at 10 percent significant at 5 percent significant at 1 percent Table gives thedifference in mean absence rates between the indicated category and its complement For example itshows that male teachers in India have an absence rate that is 52 percentage points higher than that offemale teachers and that the difference is significant at the 1 percent level

Nazmul Chaudhury et al A1

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 28: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

Table A-2Health Workers Mean Differences in Absence Rate by Selected Characteristics

India Indonesia Bangladesh Peru Uganda

Male 20 41 26 78 67Longer-term employee 109 19 114 15 38Born locally 158 53 131 94 87Contract employee 55Employee is doctor 45 23 175 08 150Employee works at night shift 61 201 06 37 92Employee provides outreach services 91 48 14 11 68Employee resides in PHC housing 31 72 49 69 89Facility inspected recently 22 106 33 25 14Facility is near Ministry of Health office 02 56 50 82 02Facility has toilet 01 55 53Facility has water 38 02 12 143 124Facility is near paved road 25 286 150 97 05Facility in urban area 44PHC 22CHC 51

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Table givesthe difference in mean absence rates between the indicated category and its complement For exampleit shows that male health workers in India have an absence rate that is percentage points lower than thatof female teachers and that the difference is significant at the 1 percent level

A2 Journal of Economic Perspectives

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 29: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

Table A-3Correlates of Teacher Absence (OLS and HLM District-Level Fixed Effects)(dependent variable visit-level absence of a given teacher 0 present 100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]Ecuador

[3]India

[4]Indonesia

[5]Peru

[6]Uganda

[7]All countries

Male 3518 0669 2327 2174 2037 2356 1942[3030] [2696] [0580] [1775] [2103] [2005] [0509]

Ever received training 2929 23859 2661 6176 1532 5565 2141[3086] [7575] [0963] [3211] [11133] [3113] [4354]

Union member 0097 6112 0405 4174 0395 1631 2538[2704] [2617] [0731] [2978] [2246] [2529] [1258]

Born in district ofschool

261 4722 1713 3117 0031 02 2715[3829] [2969] [0607] [1746] [2559] [2343] [0833]

Received recenttraining

2017 7979 0402 242 2262 2045 074[3173] [2924] [0713] [1870] [2472] [2695] [2070]

Tenure at school(years)

0029 0116 002 0106 0263 0721 0033[0178] [0186] [0041] [0133] [0187] [0291] [0044]

Age (years) 0173 0206 0038 004 0165 0317 0021[0207] [0145] [0034] [0155] [0153] [0177] [0046]

Married 4615 0309 0651 0928 1165 4904 0742[5877] [2445] [0835] [3207] [1698] [2237] [0972]

Contract teacher 5509 0687 8250 3432 5722[4426] [1407] [3556] [3343] [2906]

Has university degree 4271 3675 1503 073 1048 11773 1055[2953] [2407] [0589] [2530] [3331] [6572] [1162]

Has degree ineducation

28601 7492 1758 4277 6831 16266 1806[5836] [3802] [1014] [5438] [4682] [4239] [2071]

Head teacher 3326 0724 4482 7326 6205 5849 3771[3515] [5606] [0719] [3691] [8921] [4756] [0888]

School inspected inlast 2 mos

2227 0522 2435 1867 0657 386 0142[2218] [5316] [0685] [2307] [2356] [3121] [1194]

School is near MinEducation office

2963 11105 1535 5454 012 1071 4944[2554] [4217] [0773] [3199] [3066] [3569] [2642]

School had recentPTA meeting

1248 4261 0962 1816 4880 1092 2308[2486] [4515] [0707] [2479] [2518] [3038] [1576]

Studentsrsquo parentsrsquoliteracy rate (0ndash1)

1248 10313 5132 22634 24295 6883 9361[4659] [13446] [1663] [16143] [11303] [10810] [1604]

School infrastructureindex (0ndash5)

2126 4648 1352 104 1991 3197 2234[2090] [2682] [0382] [1817] [1751] [2771] [0438]

School is near pavedroad

1338 4116 0784 3083 3317 1264 0040[3760] [6353] [0964] [4103] [8523] [4103] [1106]

Schoolrsquos pupil-teacherratio

0063 0440 0014 0153 0008 0145 0095[0046] [0255] [0017] [0112] [0126] [0097] [0080]

School is in urbanarea

1285 2769 0341 1436 1189 5103 2039[2014] [5516] [0837] [3131] [6171] [3577] [1441]

Schoolrsquos number ofteachers

0215 0267 0046 0282 0192 0112 0015[0652] [0443] [0144] [0349] [0130] [0317] [0113]

School has teacherrecognition program

4062 7029 1098 7524 525 3462 0168[7848] [4724] [0827] [2866] [3574] [3597] [3525]

Dummy for 1st surveyround

0416 7543 2709 1794 4356 3037 2938[2512] [2790] [0839] [2125] [2264] [4460] [1874]

Constant 59096 1996 31215 47941 33524 3037 32959[15449] [25291] [2763] [20410] [14712] [11096] [1963]

Observations 771 1163 30825 2137 1172 1624 34880R-squared 009 021 006 006 011 014

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errorsclustered at the school level are given in brackets for OLS regressions in columns 1ndash6 Regressions also includeddummies for the days of the week

Missing in Action Teacher and Health Worker Absence in Developing Countries A3

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives

Page 30: Missing in Action: Teacher and Health Worker Absence in …siteresources.worldbank.org/INTPUBSERV/Resources/47… ·  · 2009-01-16University, Cambridge, Massachusetts. Karthik Muralidharan

Table A-4Correlates of Health Worker Absence (OLS and HLM District-Level FixedEffects)(dependent variable visit-level absence of a given medical staff member 0 present100 absent)

Country-specific regressions Global HLM

[1]Bangladesh

[2]India

[3]Indonesia

[4]Peru

[5]Uganda

[6](ex Bangl)

Male 3404 2624 211 0934 1121 0628[6541] [0662] [2119] [2929] [2958] [1475]

Tenure at facility(years)

1467 0469 0682 105 0706 0081[1473] [0126] [0501] [0863] [0608] [0382]

Tenure at facilitysquared

0046 0009 0029 008 0001 0008[0073] [0005] [0023] [0059] [0024] [0011]

Born in PHCrsquos district 13479 0237 2328 2959 8263 1404[4609] [0649] [2114] [4295] [3055] [0873]

Contract employee 7058[2649]

Doctor 15499 3226 3512 0325 15551 3380[6714] [0854] [2481] [3113] [4662] [0754]

Works night shift 489 4921 1717 4013 4851 4267[5829] [0672] [3278] [3076] [3352] [1066]

Conducts outreach 1286 6297 4874 1422 7677 6617[5525] [0671] [2995] [4027] [3246] [0620]

Lives in PHC-providedhousing

10223 0912 2334 5027 564 0583[5162] [1063] [2638] [5298] [3400] [1507]

PHC was inspected inlast 2 mos

5989 0356 4114 1357 3149 1975[5545] [0676] [2895] [2802] [2815] [0624]

PHC is close to MOHoffice

4641 2598 5054 4311 0945 0768[5261] [1550] [2132] [3191] [4604] [1999]

PHC has toilet 4163 0863 11162[11713] [0777] [13534]

PHC has potable water 10283 269 8106 1871 8233 3352[9450] [0840] [4815] [5598] [4486] [0844]

PHC is close to pavedroad

8865 0874 32652 4811 0599 6076[9386] [0775] [11357] [4185] [4480] [3042]

Dummy for 1st surveyround

4697 27659 8664 5574 12457[0674] [1596] [4903] [2761] [11180]

Dummy for 2nd surveyround

3648[0735]

Constant 25866 36723 74061 44076 51087 38014[16876] [2074] [12927] [17566] [11649] [1538]

Observations 339 26127 1767 1123 1264 27894R-squared 012Number of providers 9493 1094 607 747

Notes Significant at 10 percent significant at 5 percent and significant at 1 percent Robust standard errors inbrackets Bangladesh regression uses only one round of data and is therefore a simple cross-section Regressionsinclude dummies for days of the week (not reported here) Where applicable regressions also include dummies forurban area (Peru) and for type of clinic (Bangladesh India)

A4 Journal of Economic Perspectives