Worker Absence and Shirking: Evidence from Matched Teacher ...

Worker Absence and Shirking: Evidence fromMatched Teacher-School Data

Steve Bradley�, Colin Greeny, Gareth Leevesz

April 12, 2005

Abstract

We utilise a unique matched teacher-school data set of absenteeism

records to quantify the impact of group interaction on the shirking be-

haviour of primary and secondary teachers. To resolve the identi�cation

problem our study focuses on teachers who move between schools, and it

is shown that group absence increases teacher shirking by approximately

6%. There is also evidence of multiple equilibria insofar as teachers who

move from �healthy�(low absence rate) schools to �sick�(high absence rate)

schools, for instance, exhibit a further increase in shirking.

KEYWORDS: Absenteeism, Shirking, Group Interaction E¤ects

JEL CODE: J29, J45, M54.

�Corresponding Author: Department of Economics, The Management School, LancasterUniversity, UK.

yCentre for Economic Policy Modelling, University of Queensland, Australia.zSchool of Economics, University of Queensland, Australia.

1

1 Introduction

Absenteeism is viewed as costly and disruptive for organisations. For instance,

Barmby et al (2002) demonstrate for nine industrialised countries that a sig-

ni�cant proportion of work hours are lost through absence ranging from 1.8%

to 6% of average weekly hours for Switerzerland and Sweden, respectively. As

we note later, �gures for the teaching profession fall within this range. Teacher

absence incurs direct costs through the organisation and payment of cover for

absent teachers. In addition, there are indirect costs in terms of adverse e¤ects

on other sta¤ and potentially on student performance (Ehrenberg et al 1986).

In contrast, psychologists have suggested that absenteeism in stressful situations

might be implicitly condoned by management because it is seen as a form of

temporary relief for the employee, and could therefore be regarded as an e¢ cient

response (Steers and Rhodes 1978).

Devising an appropriate policy response to absenteeism is therefore complex,

even more so when it is realised that worker absence can occur involuntarily be-

cause of physical or mental ill-health, or because of voluntary shirking behaviour

by workers (Barmby et al 1994, Johansson and Palme 2002). A recent report

by the Western Australian Government conjectures that up to 50% of absence

amongst public sector workers, including teachers, is in fact voluntary, which

implies a substantial degree of shirking (Auditor General 1997).1 Disentan-

gling shirking behaviour (i.e. voluntary absence) from involuntary absence is

1 Imants and van Zoelen (1995) suggested that only 20% of teacher absence could be directlyrelated to what they described as justi�able medical grounds.

2

extremely di¢ cult to do in practice. Previous research has attempted to do so

using absence spell length, however, as Driver and Watson (1989) argue this is

unreliable because long spells of absence could be voluntary and a sequence of

short spells could be indicative of recurring sickness.

Consequently, this paper does not explicitly try to distinguish between vol-

untary and involuntary absence. Instead, we take the approach of looking for

peer, or group interaction, e¤ects on individual absence behaviour, and also in-

vestigate the existence of multiple equilibria in this relationship. A signi�cant

positive association between individual and group behaviour in relation to ab-

senteeism rates increases the likelihood that we are observing shirking behaviour

rather than simply sickness. There are, however, serious identi�cation problems

when trying to establish whether group behaviour a¤ects individual behaviour,

which we discuss below, and to resolve this problem our paper focuses attention

on teachers who move between schools. Our investigation of multiple equi-

libria draws on the education literature which distinguishes between �sick�and

�healthy�schools (Imants and van Zoelen 1995), the former having a consistently

higher incidence of shirking. Our hypothesis is that teachers who move from

a healthy to a sick school are likely to exhibit a higher incidence of absence,

primarily because of the in�uence of group norms and behaviour on shirking

behaviour.

We test these hypotheses using a unique and previously unused database of

matched teachers and schools obtained from the Queensland government. Using

this data we estimate cross-sectional and panel �xed e¤ect models for the full

3

sample and models for teachers who move schools. The results of cross-sectional

models of absenteeism are also reported to help �ll the gap in the empirical liter-

ature on the correlates of shirking in the teaching profession and more generally

within the public sector. Our results suggest that shirking behaviour is in�u-

enced by prevailing group absence behaviour at the school. Moreover teachers

moving from healthy to sick schools exhibit larger increases in shirking than

those who move between other categories of school. A move from a healthy to a

sick school leads to an increase in absence on average of 1 day per quarter. This

suggests that schools with relatively poor norms with respect to absence behav-

iour have markedly greater impact on individuals absenteeism. It is also shown

that teachers on temporary contracts have absenteeism rates 30-40% lower than

permanent teachers. This is consistent with earlier evidence on contract status

e¤ects in the private sector (Ichino and Riphahn 2001).

The rest of this paper is set out as follows. Section 2 contains a review of

relevant literature, including a discussion of the notion of a sick school. Section 3

provides a detailed description of the data that are used in the empirical analysis.

In section 4 we provide a theoretical framework and econometric methodology

for the empirical work, which is followed by a discussion of the results in section

5. Section 6 concludes the paper.

4

2 Literature Review

A sizeable literature exists with regard to the in�uence of group interaction

e¤ects on a range of social and economic outcomes, such as crime, educa-

tional achievement, employment and ethnic skill di¤erences (Case and Katz

1991, Borjas 1995, Glaeser et al 1996, Aaronson 1998, Sacerdote 2001, Mar-

maros and Sacerdote 2002, Arcidiacono and Nicholson 2005). Of particular

relevance to the current study is Ichino and Maggi (2000) who found evidence

of signi�cant group interaction e¤ects on individual shirking behaviour within

a large Italian bank.

In the education literature, there have been many qualitative case studies of

the factors in�uencing teacher absenteeism (Eisenberg et al 1986, Dworkin et

al 1990, Imants and van Zoelen 1995), but little statistical analysis. Imants and

van Zoelen (1995) argued that it was worthwhile classifying schools in terms

of those that are healthy and those that are sick. Speci�cally, they found that

collegial sta¤ relationships and a supportive management style from the school

principal reduced the rate of teacher absence. Furthermore, a recent report for

Australia concluded that �motivation and commitment appear to be important

elements which restrict absence through sickness�, giving rise to a need for ab-

sent management programmes �that take account of the importance of schools�

leaders in promoting healthy work environments� (Department of Education,

Science and Training 2001).

The only previous econometric analysis of teacher absenteeism is Ehrenberg

et al (1986), who focused on providing an explanation of the di¤erences in aver-

5

age rates of teacher absenteeism between school districts because of di¤erences

in sick pay provisions. Average absenteeism was positively associated with the

use of sick banks, whereby teachers accumulate unused sick leave, and negatively

related to buy-back provisions that allowed unused sick leave to be cashed in.

There was also some evidence to suggest that a larger fraction of teachers aged

over 55 reduced sick leave usage. Of the variables controlling for school en-

vironment, the proportion of students living in an urban area and the ethnic

composition of the student body were statistically insigni�cant but the average

level of student absenteeism was positively correlated with teacher absenteeism,

which implies that the e¤ect of group behaviour on individual behaviour may

extend beyond the in�uence of one�s colleagues.

More generally, previous research on absenteeism conducted by economists

has focused mainly on the private sector and investigated the impact of both

individual (supply side) and workplace (demand side) characteristics (Brown

and Sessions 1996, Delgado and Kniesner 1997, Barmby and Stephan 2000).

Demand side factors that have been discussed in the literature include �rm

size (Coles and Treble 1996). It is shown that larger �rms have higher rates

of absenteeism, which arises because of their ability to diversify the risk from

absence more easily (Barmby and Stephan 2000). Clearly, the same could apply

in a school environment. Supply-side factors, such as age and gender, have been

investigated using aggregate data by Paringer (1983) and Bridges and Mumford

(2001). Both studies found that older and single workers were more likely to be

absent, especially for men. However, Delgado and Kniesner (1997) found the

6

opposite e¤ects in a microeconomic study of London bus drivers.

The in�uence of the worker�s contract status on absenteeism has also been

investigated by Arai and Thoursie (2005) and others. For instance, Barmby et

al (1995) and Barmby (2002) show that full time workers, or those with longer

contracted hours, are more likely to be absent when compared to part time

workers. Similarly, Ichino and Riphahn (2001) show that absence rates increase

once a probationary period ends and employment protection legislation comes

into e¤ect.

3 Data

This study uses personnel records, including data on absenteeism, which was

obtained from Education Queensland�s human resource information system for

quarter 4 2001 to quarter 3 2002 for all teachers employed in the state school

system. The absenteeism data is recorded under a number of categories: fam-

ily/paternity leave, work injury leave that is covered by employee insurance,

time o¤ due to industrial disputes, and absence due to reasons of self-reported

illness. The absence days are recorded as the cumulative number of days per

quarter. Our data tracks the individual teachers throughout the time period,

thereby enabling us to construct a panel data set of teacher absenteeism. We

focus on absence due to reported illness. Teachers can take up to two weeks

sick leave per year. They can accumulate unused sick leave in a sick bank and

there is no limit on how much leave can be accumulated, but unlike some school

7

districts in the US, there are no buy-back provisions for unused sick leave.

In common with previous studies, we restrict our analysis to teachers who

were employed throughout the year, to remove possible bias related to the pro-

bationary teaching period and those teachers in the process of leaving the pro-

fession.

Table 1 shows summary statistics of teacher�s absence. The �gures are mean

values for the percentage of working time lost and percentage of teachers taking

absence. Figures for the UK from Bowers and McIver (2000) are presented for

comparison. It is interesting that although full-time teachers in Australia are

more likely to take absence, the average percentage of time lost through absence

is similar in the UK and Australia.

For each teacher we have information on personal characteristics, such as

age, gender, whether a worker has a disability, their tenure in the public sector

and ethnic background; the latter is comprised of two groups, teachers from a

Non-English Speaking Background (NESB) and other Australians.2 We also

have data on the workplace characteristics of teachers, drawn from another

source of school administrative data. From this we can determine the school

the teacher was working at, and along with occupational codes allows us to

distinguish between primary and secondary school teachers and the size of the

school.

Queensland teachers can be employed under three types of contracts - ca-

sual, permanent or temporary. Casual teachers are generally short-term contract

2Some teachers are indigenous Australians, however they represent a very small proportionof the teaching workforce and provide insu¢ cient observations for statistical analysis.

8

sta¤, usually �relief�teachers, who may enter and leave the teaching workforce on

a regular basis. Consequently, we exclude this group from our analysis. Tempo-

rary teachers are employed on �xed term contracts, generally of one years length,

which can be renewed, whereas permanent teachers are in e¤ect tenured. Addi-

tionally, teachers can be employed on a full-time or part-time basis. Unlike US

school systems where there can be considerable variation in employment terms

and conditions across school districts (Ehrenberg et al 1986), all Queensland

teachers operate under the same basic pay and working conditions.

Measures of school quality are constructed from student record data that

refer to all students in the Queensland state school system in 2001. For primary

school students we have records for all year 7 students (the �nal year of primary

school) and matching year 5 records (for 1999). For secondary students we have

all year 10 student records (the �nal compulsory year of high school) and year

12 records for those pupils who remain in education. The student dataset also

provides information on student test score performance. In our data we have

test scores for year 7 (with matching year 5 scores) and standardized year 12

test scores. The individual pupil test scores are averaged for each school to

provide an indicator of school quality. However, for secondary schools, year 12

test scores will not necessarily be an accurate representation of overall student

quality. This is because: (a) poorer quality or less motivated students are likely

to leave the school after the end of compulsory schooling (year 10); and (b)

some students may choose to move to more successful schools to complete their

high school. To allow for this our measure of secondary school quality utilises

9

school retention data, which refers to the proportion of students who began at

the school in year 7 and also completed at the same school in year 12.

Queensland covers a large area of land and geographically remote areas

have higher concentrations of indigenous Australians who are, in socioeconomic

terms, generally amongst the most disadvantaged individuals (Bradley et al

2005). Other schools are in regional centres and then there is the urban concen-

tration of population in south east Queensland centred around Brisbane. As we

have the postcode for each school, this enables us to identify urban, rural and

remote schools.

In Table 2 we present summary statistics on variables used in the empirical

analysis separately for primary and secondary schools. It is clear that women

dominate primary school teaching by a ratio of 5 to 1, whereas the gender

disparity is much lower in secondary school teaching. Other covariates show

little variation between primary and secondary schools. Three quarters of all

secondary school teachers are non-specialists, the omitted case.3 Mean absen-

teeism rates are very similar for primary and secondary school teachers. Figure

1 displays the distribution of individual absence days. The majority of teachers

took 0 or 1 days absence per year, whilst there is a clear tail at the right hand

side of the distribution.

3The quali�cations required to be a teacher in Australia di¤er between primary and sec-ondary school sectors. Secondary school teachers generally complete an undergraduate coursefollowed by a graduate diploma in education. Conversely, primary school teaching involves anundergraduate course dedicated to primary school education. As a result subject specializationand post-graduate study information is only available for secondary school teachers

10

4 Theoretical and econometric framework

To analyse these data we draw upon the theoretical framework and econometric

methodology developed by Ichino and Maggi (2000) in their study of absenteeism

amongst the employees of an Italian bank. They suggest four broad determi-

nants of individual variations in absenteeism. First, �individual backgrounds�,

including personal characteristics and region of birth, which in the present con-

text also includes the contract status of the teacher. Second, the characteristics

of the school in which the teacher works and the locality in which they live,

for instance, the size of the school and the local unemployment rate, which are

labelled �locality� e¤ects. These factors are what Manski (1995) refers to as

contextual e¤ects, and exogenous changes in them have no multiplied e¤ect on

teacher behaviour. Third, teachers may be sorted between schools which have

di¤erent average shirking rates, which is linked to a fourth in�uence, that of

group interaction. In this case shirking by an individual teacher is in�uenced

by the prevalence of that behaviour in the group (Manski 1993, Manski 1995),

which gives rise to the idea of a social multiplier e¤ect. Ichino and Maggi (2000)

suggest that the social multiplier could arise in the shirking context via (i) a

peer monitoring mechanism, that is, if more teachers shirk then an individual

shirker is less likely to be reported thus the penalty is lower and hence shirking

is higher; (ii) stigma e¤ects, whereby shirking represents �acceptable�behaviour;

and (iii) if the school management has limited monitoring resources then the

likelihood of being caught shirking is reduced.

In view of these considerations, it is suggested that workers will compare

11

the bene�ts of shirking with the costs of shirking to maximise expected utility

(EU), as represented by Equation 1 (we drop the individual subscript, i, for

simplicity).

EUt = B(At; S; �t)� C(At;_A) (1)

where At is the individual teachers shirking propensity, S is a set of school

and locality factors, and � represents the teacher�s preference for shirking._A is

the average shirking rate of the school. The marginal penalty to shirking is

lower for higher values of_A for the reasons highlighted above. It follows that a

teacher�s propensity to shirk will be determined by the factors listed in Equation

2.

At = g(_A;S; �t) (2)

From the assumptions above it follows that @At=@_A > 0 and @At=@� > 0.

Chatterji and Tilley (2002) introduce the concept of the �presenter�, these are

workers with a low preference for shirking. Such workers derive satisfaction

from being able to maintain a presence at work, regardless of illness, or see it

is in their own best interest to carry out their duties. The teacher�s concern

for her class or the excess workload on colleagues, or perhaps their reputation

with the school management, counterbalances the decrease in utility from illness

or stress. Moreover, extra leisure time generates no increase in utility for this

type of individual, and thus the presenter is likely to have a low value of �. We

cannot observe � directly, however, it is likely that the preference for shirking

12

is highly correlated with the personal characteristics of the teacher.

As the literature suggests, workers on temporary contracts are likely to have

a lower value of � (i.e. be more like a presenter) in the hope of being o¤ered

a permanent contract, whereas part-time workers might have a low value of �

because their absence is more easily noticed by management.

The �rst model that we estimate is therefore given by Equation 3.

A = �P + �_A+ S + � (3)

Initially, we ignore the time dimension in our data and simply estimate

Equation 3 as a cross-sectional model. P is a vector of individual background

characteristics. In a subsequent model we exploit the longitudinal nature of our

data and rede�ne (3) as a panel data model with individual speci�c �xed e¤ects:

At = �+ �tP + �_At +

Xj

&jDjt + St + �t (4)

where � is an individual speci�c �xed e¤ect, Djt is a dummy variable to cap-

ture unobserved locality e¤ects. Ideally this variable would identify the school

that a teacher was in at time t, however, there are insu¢ cent teachers who move

between schools to identify all 1123 school �xed e¤ects, and therefore we are

forced to adopt a similar approach to Ichino and Maggi (2000) by including

�xed e¤ects for the 33 education districts. We argue that as education districts

are reasonably homogenous with respect to pay and policies these should ade-

quately capture the e¤ect of time invariant local characteristics. Finally, �tP

13

is simply a time dummy interacted with ostensibly �xed individual background

variables so that we can obtain estimates of their e¤ect.

There are two further issues that must be considered. First, because group

interaction e¤ects are endogenous, estimates of_A may well be biased or even

spurious. One source of endogeneity bias is the so-called �re�ection problem�

whereby average group behaviour a¤ects individual behaviour but each indi-

vidual teacher�s behaviour also a¤ects average behaviour, which means that

individual errors are correlated and hence our estimates will be biased. Another

source of endogeneity bias arises if relevant school covariates are omitted, in

which case the errors for teachers in the same school will be correlated and thus

the e¤ect of_A will again be biased. Spurious e¤ects arise where teachers behave

in a similar way because they share a common set of unobservable characteris-

tics which causes teachers to sort between schools. For instance, shirkers sort

into sick schools and presenters sort into healthy schools.

In short, there is an identi�cation problem, which we mitigate by investi-

gating the e¤ect of group interaction only for those teachers who move schools.

We can be fairly certain of a less biased estimate of_A for this group because, as

Gaviria and Raphael (2001) argue, the �geographical boundaries�of interaction

are well de�ned within schools, insofar as the interaction between teachers is

less likely to be a¤ected by extraneous in�uences, unlike group interaction in

neighbourhoods, where individual behaviour is far more likely to be in�uenced

by factors other than group interaction.

Our model for movers is given by Equation 5:

14

At�At�1 = �(_At�

_At�1)+

Xj

(Djt�Djt�1)&j + (St�St�1)+ �t� �t�1 (5)

where Djt �Djt�1 is an indicator that a teacher moved education district.

Second, as Ichino and Maggi (2000) note, there may be multiple equilib-

ria, insofar as the existence of signi�cant group interaction e¤ects may generate

di¤ering equilibrium shirking norms at the school level This would imply, at

the least, a bimodal distribution of school absence rates. Figure 2 shows that

a bimodal distribution of shirking rates is not observed for our population of

schools, however, there are clearly some schools with low rates of shirking, a large

proportion with intermediate rates and a long tail of schools with high rates.

We use information on school absence rates and follow the suggestion in the

education literature that schools can be classi�ed as either sick or healthy. We

then use this categorisation to investigate the possibility that group interaction

e¤ects are magni�ed due to di¤erences in local attributes and school circum-

stances across school types. This model is speci�ed below and, once again, is

only estimated for movers.

At�At�1 = �1(_At�

_At�1)+

Xh

(Sct�Sct�1)�+Xj

(Djt�Djt�1)&j+ (St�St�1)+�t��t�1

(6)

where Sct refers to the category of school, that is �sick�, �healthy�, or in the

intermediate group (de�ned later). As a result the term (Sct � Sct�1) provides

15

a full matrix of dummy variables for all the possible types of moves between

these three categories of schools, including movements between the same type

of schools (i.e. healthy to healthy). We are particularly interested, however,

in whether there is a di¤erence in the behaviour for teachers who move from

sick to healthy schools (i.e. their propensity to shirk falls), compared to those

teachers who move from healthy to sick schools (i.e. their propensity to shirk

increases).

5 Results

Tables 3 to 5 report estimates of equation 3 for primary teachers, secondary

teachers and all school teachers, respectively. Initially we estimate equation 3 by

OLS. However, the record of teacher absenteeism in our data is a count of events

over a period of time, and therefore estimation using a count data approach is

more appropriate and also provides a comparison to the OLS estimates.

The results for each variable are reported as raw coe¢ cients with standard

errors and, for ease of interpretation, the incident rate ratio (IRR) is also pro-

vided for the count model estimates. The latter provides more direct comparison

of the relative e¤ects of covariates. An IRR of greater than one indicates that

the expected count in the exposed group is greater than the expected count in

the unexposed group. For instance an IRR of 1.15 indicates a 15% increase in

the absenteeism rate for the exposed group, whilst 0.85 indicates a 15% decrease

in the rate. Since there is a lack of detailed econometric evidence on the deter-

16

minants of shirking amongst teachers, we �rst discuss the impact of individual

background and locality factors before we move to the main focus of the paper,

the estimation of group interaction e¤ects.

5.1 The e¤ect of personal and locality characteristics on

shirking

The OLS results in Tables 3 and 4 show that there is a non-linear relationship

between age and shirking which declines quite sharply for primary and secondary

teachers but with slightly di¤erent turning points. For primary school teachers

the turning point is around the age of 28 whereas for secondary school teachers

it is age 32, after which shirking increases. We also estimated a poisson model

where the age variable was grouped into bands to investigate the non-linear

relationship in more detail. The �ndings (not reported) show that for primary

and secondary school teachers there is a substantial increase in absenteeism for

older age groups. For instance, secondary school teachers aged 31-40 have a 17%

increase in absenteeism compared to younger teachers (the base group) whereas

workers aged 51-60 have 77% greater absenteeism, a di¤erential of 60 percentage

points. Similarly, the di¤erential between primary school teachers aged 41-50

and those aged 61 or more is 59 percentage points. This is consistent the age

e¤ects found in aggregate evidence presented by Bridges and Mumford (2001).

These �ndings could be partly to do with the e¤ect of taking accumulated leave

before retirement (Ehrenberg et al 1986), or because of a gradual decline in en-

thusiasm for the job and hence an increase in shirking, given that we control for

17

health through the disability variable. There is no statistically signi�cant gen-

der di¤erence in the incidence of shirking for either primary or secondary school

teachers. Secondary school teachers with an Non-English speaking background

have absence rates - around 18% higher. As expected, the impact of having a

registered disability on the amount of absenteeism taken is very high, 59% and

55%, respectively, for primary and secondary teachers.

There are some interesting �ndings with respect to contract status and the

incidence of shirking. Workers on temporary contracts take signi�cantly (33%-

39%) fewer days of absence than their colleagues on permanent contracts, which

may be indicative of a presenteeism e¤ect in an attempt to secure a perma-

nent contract (Ichino and Riphahn 2001). A further aspect of the contractual

arrangement for teachers is that some are asked to perform higher duties in

return for more pay, perhaps as a precursor to promotion, but certainly in

recognition of teacher merit. We �nd that those primary and secondary teach-

ers exercising such duties have around 35% less absence, which indicates that

performance-related pay may reduce shirking. For secondary school teachers we

note that absenteeism varies by subject specialism.

Turning to the locality e¤ects, the e¤ects of school size and the average

quality of the pupil population are either statistically insigni�cant (Table 3)

or very small (Table 4). Some of the contextual e¤ects are slightly larger and

more signi�cant, for instance, the higher the average age of the teaching body

in the school, the greater an individual teachers absence or shirking. We �nd

that teachers in remote and rural schools take less absence than teachers in

18

urban schools. This could be related to a greater commitment amongst teachers

willing to take teaching posts in these localities, a wider sense of community

involvement, or simply an inadequate supply of casual teachers, which means

that teachers in non-urban areas are forced to present themselves for work to

ensure that the class is taught4 . Lastly, teachers in schools in an area with a

higher unemployment rate have higher levels of shirking, however, these e¤ects

are small and are only statistically signi�cant in the combined model (Table 5).

This is an interesting �nding insofar as it contrasts with the time series evidence

which suggests an inverse relationship between unemployment and absence rates

(Kaivanto 1997).

In summary, we �nd that in general the variables re�ecting individual back-

ground and locality e¤ects are either quantitative or statistically unimportant

explanations of teacher shirking behaviour. However, there are some interesting

exceptions, such as disability, which is not unexpected, contract status and the

remoteness of the school.

5.2 The e¤ect of group interaction on shirking

Table 6 summarises our estimates of the group interaction e¤ect that are derived

from the cross-section, panel and mover models. The estimates of_A in columns 2

and 3 are reproduced from Tables 3 to 5. A general �nding is that the e¤ect of_A

is insigni�cant in the cross-sectional models for secondary school teachers but is

4There is some support for this latter argument as casual teachers as a proportion of thetotal teacher workforce is much lower in remote areas; 50% less than in urban areas in thecase of primary schools and 20% less for secondary schools

19

large and positive for the primary school teacher models. The estimates from the

combined teacher models are also statistically signi�cant. Column 4 presents

estimates of_A obtained from a pooled OLS estimation of Equation 4, which

follow a similar pattern to the cross-sectional estimates. However, when we

control for individual speci�c �xed e¤ects (column 5) two things happen. First,

for primary school teachers the impact of_A falls from 0.14 to 0.09. Second, the

estimate of_A for secondary school teachers is positive and signi�cant, and larger

in magnitude than the primary school e¤ect. These basic set of results suggest

the presence of group interaction e¤ects on shirking behaviour, and especially in

the case of secondary school teachers, are substantially changed by the inclusion

of controls for unobserved individual level heterogeneity:

The problem with these estimates, however, is that there is an identi�cation

problem for the reasons discussed earlier. We therefore estimate equation 5

for those teachers who move schools, with and without the dummy variables,

Djt. In the appendix (Table A1) we present summary information on movers

in comparison to stayers. The main di¤erences are that movers are slightly

younger, have a shorter tenure and are more likely to be on a temporary contract.

Inter-school movement may be part of a sorting process where, for instance,

teachers with high shirking propensities move to schools with higher average

absence levels. However, it was found that individuals in the highest quartile

of the absence distribution moved to schools that had a slightly lower average

absenteeism than their origin school (1.86 days - 1.68 days). For movers from

other quartiles there was no discernable change in average school absenteeism

20

between origin and destination. Hence, there does not appear to be systematic

sorting across schools by shirking propensity, at least within the limited time

frame in which we observe movement.

Column 6 and 7 of Table 6 provide estimates of the mover models. As these

models are estimated for the period of movement, they provide estimates of

the contemporaneous impact of group interaction e¤ects on individual shirking.

For primary school teachers the e¤ect of_At �

_At�1 is robust at around 0.07,

regardless of the inclusion of the district dummy variables, whereas for sec-

ondary school teachers the coe¢ cient drops to around 0.01 and is statistically

insigni�cant when district dummy variables are included. We have relatively few

secondary teacher movers and this may be causing this result. The combined

model pools both groups of teachers and we observe that the e¤ect of the group

interaction e¤ect falls by a third when we compared to the combined panel �xed

model (0.09 to 0.06). We can therefore claim that group interaction increases

shirking by approximately 6%. In other words, an increase of approximately 16

days absence per quarter by the peer group would increase individual teacher

absence by one day.in that quarter.

Up to this point we have ignored the motivation for teachers to move school.

This may impact upon our estimates of group interaction e¤ects insofar as

individuals whose shirking behaviour is improving over time may move to low-

shirking schools. If this occurs, this would tend to overstate the relationship

between individual and group absence behaviour and lead to an upwardly bi-

ased estimate of the e¤ect of_At �

_At�1 (Ichino and Maggi 2000). Our data

21

does not explicitly identify reasons for movement between schools. However, we

observe changes in a worker�s job classi�cation level and their wage rate, which

means that we can identify those teachers who received a promotion as part of

their school move and those who did not. We re-estimate our mover model sepa-

rately for the promoted and non-promoted sub-groups on our combined sample.

Estimates of_At�

_At�1 for the non-promoted sub-group are approximately 0.074

(t-value = 1.76) whilst the estimate for the promoted movers was 0.051 (t-value

= 2.04). This result suggests endogeneity e¤ects related to the in-sample sorting

of workers have not seriously biased our earlier estimates of group interaction

e¤ects.

5.3 Multiple equilibria

Irrespective of potential sources of bias, the size of the group-interaction e¤ects

illustrated in these models are not quantitatively large, however, we investigate

these e¤ects further by examining the issue of multiple equilibria. In particu-

lar, we use the healthy-sick school classi�cation to identify if group interaction

e¤ects are quantitatively di¤erent at schools with di¤erent norms regarding ab-

sence behaviour. This categorisation is based on whether school yearly absence

rates lie in the top 25%, intermediate 25%-75% or the lowest 25% of the school

absence distribution.5 The �rst group are termed �sick� schools and the lat-

ter group �healthy�schools. We have access to limited data on average school

absence levels for one quarter in the year prior to our main sample. We use

5As a robustness check, we duplicated these results using a 20%, 80% classi�cation. Thisdid not change the results substantively.

22

this information to compare the distribution of average absenteeism by school

between our sample and the previous year. We rank schools by average absen-

teeism in both periods and performed a Spearman rank correlation test. The

coe¢ cient of 0.40 indicates that the ordering of schools between the two periods

is not independent. Thus, there is a persistance in the distribution of relative

average school absenteeism.

Table 7 presents summary statistics for each type of school, and it is clear

that healthy schools have teachers who are younger with shorter tenure. These

schools are more likely to be in rural or remote areas, are smaller and mainly

in the primary school sector. Di¤erences between intermediate and sick schools

characteristics are less marked, although age and tenure appear to be monoton-

ically increasing across the bands.

In Table 8 we present the estimates of Sct � Sct�1 from equation 6, along

with the estimate of_At �

_At�1. First, we note that the group interaction e¤ect

is robust to the inclusion of the new absence categorisation dummy variables.

Of the dummies that represent moves between the school categories only moves

to �sick�schools have a statistically signi�cant e¤ect on individual absence rates.

What is perhaps even more important is the fact that this e¤ect is large. For

instance, a move from a healthy to a sick school is associated with an increase

of over one days absence in the following quarter when compared with a move

from one sick school to another. Similar, but smaller, e¤ects are apparent for

movements from intermediate to sick schools. These results suggest that there

are additional e¤ects on individual shirking as a result of moving to schools

23

with particularly poor shirking behaviour, and imply the existence of multiple

equilibria in the relationship between individual behaviour and group behaviour.

6 Conclusion

In this paper we have sought to analyse to what extent individual teacher ab-

sence behaviour is in�uenced by group behaviour. To do this we have used a

unique database of all primary and secondary teachers employed in the Queens-

land public sector over a period of one year. To address problems related to

the identi�cation of group interaction e¤ects we focused on those teachers who

moved school during this period. Our results suggest that shirking is a function

of observable teacher characteristics, such as disability, contract status and the

remoteness of the school, however, group interaction e¤ects also play a signif-

icant role in determining individual shirking. In addition, teachers who move

from schools with relatively good or average absenteeism records to those with

poor absenteeism records exhibit further increases in shirking. This suggests

that schools with relatively poor norms with respect to absence behaviour have

a markedly greater in�uence on individuals shirking behaviour.

Overall, our results support previous research which demonstrates that in-

dividual absence behaviour is related to levels of absenteeism within the work

unit (Ichino and Maggi 2000). However, in our data it appears that group

interaction e¤ects are particularly strong in high shirking work units. This is

particularly striking given that our data covers a system where work policies and

24

procedures are highly standardised. Thus, although explicit di¤erences in an

organisation�s policy can a¤ect the level of absenteeism (Ehrenberg et al 1986),

in this case variations in management and personnel factors at the school level

seem a more likely explanation. As a result, further research would do well

to consider how absenteeism, and speci�cally the group-interaction e¤ect, re-

sponds to human resource practices, such as changes in management sta¤ and

the manner in which absence management policies are applied. In summary, our

results suggest that even in large bureaucratic organisations, such as state-based

school systems, sta¤ relationships and management style do a¤ect the degree of

teacher absence.

Acknowledgement

The authors would like to thank Education Queensland for access to their data.

The opinions expressed in this paper are those of the authors and do not nec-

essarily represent the views of Education Queensland. The authors would also

like to thank Geraint Johnes, Rob Simmons, two anonymous referees and the

editor of the journal for very helpful comments on a previous draft of this paper.

REFERENCES

Aaronson, D. 1998 Using sibling data to estimate the impact of neighbourhoods

on children�s educational outcomes, Journal of Human Resources, 33, 915-946.

25

Arai, M. and Thoursie, P. S. 2005. Incentives and selection in cyclical absen-

teeism, Labour Economics, 12, 269-280..

Arcidiacono, P. and Nicholson, S. 2005 Peer e¤ects in medical schools, Journal

of Public Economics, 89, 327-350.

Auditor General. 1997. Get Better Soon: The Management of Sickness in the

W.A. Public Sector. O¢ ce of the Auditor General, Perth.

Barmby, T. Ercolani, M. G. and Treble, J. G., 2002. Sickness Absence: An

International Comparison. The Economic Journal 112, F315-F331.

Barmby, T. Sessions, J. and Treble, J. G., 1994. Absenteeism, E¢ ciency Wages

and Shirking. Scandinavian Journal of Economics 94, 561-566.

Barmby, T. Orme, C. D. and Treble, J. G., 1995. Worker Absence histories: a

Panel Data Study, Labour Economics, 2, 53-65

Barmby, T. and Stephan, G., 2000. Worker Absenteeism: Why Firm Size May

Matter. The Manchester School 68, 568-577.

Borjas, G. 1995, Ethnicity, Neighbourhoods and Human-Capital Externalites.

American Economic Review, 85, 365-390.

Bowers, T. and McIver, M., 2000. Ill Health Retirement and Absenteeism

Amongst Teachers. Research Report RR235, DfES, UK. .

Bradley, S., Draca, M., Green, C. and Leeves, 2005. An investigation of the mag-

nitude of educational disadvantage amongst indigeneous and non-indigenous

26

minority groups in Australia, Lancaster University Management School Dis-

cussion Paper 2005/008, Lancaster, UK.

Bridges, S. and Mumford, K., 2001. Absenteeism in the UK: A Comparison

Across Genders. The Manchester School 69, 276-284.

Brown, S. and Sessions, J. G., 1996. The Economics of Absence: Theory and

Evidence. Journal of Economic Survey 10, 23-53.

Case, A. C. and Katz, L. F., 1991. The company you keep: The e¤ects of

family and neighbourhood on disadvantaged youths, NBER Working Paper

No. 3705, Washington, USA.

Chatterji, M. and Tilley, C. J., 2002. Sickness, Absenteeism, Preseteeism and

Sick Pay. Oxford Economic Papers 54, 669-687.

Coles, M. G. and Treble, J., 1996. Calculating the Price of Worker Reliability.

Labour Economics 3, 169-188.

Delgado, M. A. and Kniesner, T. J. 1997. Count Data Models with Variance of

Unknown Form: An Application to a Hedonic model of Worker Absenteeism.

The Review of Economics and Statisitics 79, 41-49.

Department of Education, Science and Training., 2001. Teachers in Australian

Schools: A Report from the 1999 National Survey. Australian Government

Publishing Service, Canberra.

Driver, R. W. and Watson, C., 1989. Construct Validity of Voluntary and In-

voluntary Absence. Journal of Business and Psychology 4, 109-118.

27

Dworkin, A. G., Haney, C. A., Dworkin, R. J. and Telschov, R. L., 1990. Stress

and Illness Behaviour among Urban Public School Teachers. Educational Ad-

ministration Quarterly 26, 60-72.

Ehrenberg, R.G., Ehrenberg, R.A., Rees, D. I. and Ehrenberg, E.L., 1989. School

District Leave Policies, Teacher Absenteeism, and Student Absenteeism. Jour-

nal of Human Resources 26, 72-105.

Eisenberg, R., Huntington, R., Hutchinson, S. and Sowa, D., 1986. Perceived

Organisational Support. Journal of Applied Psychology 71, 500-507.

Gaviria, A. and Raphael, S. 2001 School-based peer e¤ects and juvenile behav-

iour, Review of Economics and Statistics, 83, 257-268.

Glaeser, E.L., Sacerdote, B. and Scheinkman, J. A. (1996) Crime and social

interactions, Quarterly Journal of Economics, May, 507-548.

Ichino, A. and Maggi, G. 2000 Work environment and individual background:

Explaining regional shirking di¤erentials in a large Italian �rm, Quarterly

Journal of Economics, August, 1057-1090.

Ichino, A. and Riphahn, R. T. 2001 The e¤ect of employment protection on

worker e¤ort: a comparison of worker absenteeism during and after probation,

IZA Discussion Paper 385, 1-18, IZA Bonn, Germany.

Imants, J. and van Zoelen, A., 1995. Teachers� Sickness Absence in Primary

Schools, School Climate and Teachers�Sense of E¢ ciency. School Organiza-

tion 15, 77-87.

28

Johansson, P. and Palme, M., 2002. Assessing the E¤ect of Public Policy on

Worker Absenteeism. Journal of Human Resources 37, 381-409.

Kaivanto, K. 1997 An alternative model of pro-cyclical absenteeism, Economics

Letters, 54, 29-34.

Manski, C., 1993 Identi�cation of Endogenous Social E¤ects: The Re�ection

Problem, Review of Economic Studies, 60, 531-542.

Manski, C, 1995. Identi�cation Problems in the Social Sciences, Harvard Uni-

versity Press, Cambridge.

Marmaros, D. and Sacerdote, B., 2002. Peer and Social Networks in Job Search,

European Economic Review, 46, 870-879.

Paringer, L., 1983. Women and Absenteeism: Health or Economics. American

Economic Review 73, 123-127.

Sacerdote, B., 2001 Peer E¤ects with Random Assignment: Results for Dart-

mouth Roommates, Quarterly Journal of Economics, ,681-704.

Steers, R. M. and Rhodes, S. R., 1978. Major In�uences on Employee Atten-

dance: A Process Model. Journal of Applied Psychology 63, 391-407..

29

0.0

5.1

.15

Den

sity

0 5 10 15 20

Figure 1 − Total Number of Sickdays Drawn in the Year

fi

0.0

5.1

.15

.2.2

5D

ensi

ty

0 10 20 30

Figure 2 − Average School Sickdays in the Year

fi

Table 1 Teacher absence rates, Australia and the UK Australia 2002 UK 1999 Primary Secondary Percentage time lost - All 3.12 2.87 3.15 Percentage time lost - FT 3.11 2.86 - Percentage time lost – PT 3.12 2.87 - Percentage taking absence - FT 86.65 65.19 Number of teachers 30,277 270,901 Source: MOHRI data Bowers and McIver (2000). Table 2 Means and standard deviations

Primary Secondary

Variable mean standard deviation mean

standard deviation

Individual absence per annum 6.22 13.16 6.07 13.74 Peer group absence per annum 6.10 3.08 5.93 1.98 Age 40.91 9.83 41.61 9.84 Age2 1770.5 821.8 1828.7 828.1 Tenure 4316.5 3760.5 4480.4 3704.1 Male 0.18 0.38 0.40 0.49 NESB 0.06 0.24 0.08 0.27 Disability 0.06 0.23 0.07 0.25 Temporary 0.08 0.27 0.06 0.23 Part time 0.16 0.37 0.07 0.26 Higher duties 0.03 0.18 0.08 0.26 School enrolment 491.6 256.3 956.8 494.8 Proportion female 0.82 0.10 0.60 0.06 Average age 40.93 3.82 41.62 2.63 Average tenure 4325.4 1368.1 4480.7 986.4 Rural 0.21 0.40 0.17 0.37 Remote 0.10 0.30 0.06 0.24 Unemployment rate 8.58 4.04 8.96 4.29 Average literacy (year 7) 681.3 35.8 - - Retention rate - - 76.9 17.2 Number of observations 16966 10280

Table 3 The determinants of shirking amongst primary school teachers OLS Poisson

Coefficient s.e. Prob value IRR s.e.

Prob value

_A 0.140 0.033 0.000 1.021 0.005 0.000 Age -0.226 0.092 0.014 0.983 0.013 0.182 Age2 0.004 0.001 0.001 1.000 0.000 0.010 Tenure 0.000 0.000 0.100 1.000 0.000 0.281 Male 0.021 0.272 0.937 1.005 0.041 0.904 NESB 0.244 0.434 0.574 1.035 0.066 0.594 Disability 3.872 0.755 0.000 1.590 0.117 0.000 Temporary contract -1.839 0.212 0.000 0.673 0.029 0.000 Part-time contract -3.203 0.220 0.000 0.494 0.028 0.000 Higher duties -2.188 0.305 0.000 0.665 0.044 0.000 Average literacy (year 7) 0.000 0.003 0.940 1.000 0.001 0.999 School enrolment 0.000 0.000 0.596 1.000 0.000 0.748 Proportion female 0.315 1.049 0.764 1.039 0.189 0.834 Average age 0.069 0.042 0.100 1.012 0.007 0.083 Average tenure 0.000 0.000 0.685 1.000 0.000 0.689 Rural -0.354 0.270 0.189 0.943 0.044 0.203 Remote -0.922 0.320 0.004 0.829 0.052 0.003 Unemployment rate 0.015 0.026 0.569 1.003 0.004 0.561 Constant 4.916 3.173 0.121 Number of observations 16966 16966 R2/Psedo R2 0.030 0.069 Wald chi2(18) 701.83 Log Likelihood -1562.4

Note: Standard errors are corrected for heteroscedasticity.

Table 4 The determinants of shirking amongst secondary school teachers OLS Poisson

Coefficient s.e. Prob value IRR s.e.

Prob value

_A 0.046 0.083 0.578 1.007 0.013 0.614 Age -0.386 0.125 0.002 0.966 0.017 0.049 Age2 0.006 0.002 0.000 1.001 0.000 0.002 Tenure 0.000 0.000 0.150 1.000 0.000 0.162 Male -0.295 0.283 0.297 0.948 0.045 0.258 NESB 1.185 0.689 0.085 1.185 0.108 0.064 Disability 3.497 0.975 0.000 1.551 0.158 0.000 Temporary contract -2.217 0.384 0.000 0.616 0.053 0.000 Part-time contract -2.682 0.430 0.000 0.530 0.070 0.000 Higher duties -2.217 0.239 0.000 0.638 0.033 0.000 Retention rate -0.021 0.007 0.005 0.996 0.001 0.006 Science -1.207 0.274 0.000 0.789 0.044 0.000 Humanities 0.201 0.419 0.631 1.040 0.073 0.577 School enrolment 0.000 0.000 0.071 1.000 0.000 0.064 Proportion female -2.073 2.268 0.361 0.681 0.271 0.333 Average age 0.161 0.093 0.084 1.031 0.017 0.070 Average tenure 0.000 0.000 0.130 1.000 0.000 0.129 Rural -1.229 0.348 0.000 0.795 0.053 0.001 Remote -1.369 0.379 0.000 0.751 0.058 0.000 Unemployment rate 0.031 0.035 0.375 1.006 0.006 0.298 Constant 8.075 4.277 0.059 Number of observations 10280 10280 R2/Psedo R2 0.030 0.069 Wald chi2(18) 336.96 Log Likelihood -64419

Note: Standard errors are corrected for heteroscedasticity.

Table 5 The determinants of shirking, primary and secondary teachers combined OLS Poisson

Coefficient s.e. Prob value IRR s.e.

Prob value

_A 0.140 0.029 0.000 1.022 0.004 0.000 Age -0.304 0.073 0.000 0.974 0.010 0.009 Age2 0.005 0.001 0.000 1.001 0.000 0.000 Tenure 0.000 0.000 0.028 1.000 0.000 0.092 Male -0.204 0.192 0.288 0.969 0.030 0.300 NESB 0.603 0.377 0.110 1.093 0.058 0.097 Disability 3.728 0.585 0.000 1.579 0.093 0.000 Primary 0.735 0.275 0.007 1.143 0.055 0.006 Temporary contract -1.965 0.185 0.000 0.653 0.026 0.000 Part-time contract -3.037 0.188 0.000 0.504 0.026 0.000 Higher duties -2.227 0.180 0.000 0.641 0.025 0.000 School enrolment 0.000 0.000 0.172 1.000 0.000 0.106 Proportion female -0.174 0.883 0.844 0.943 0.147 0.707 Average age 0.073 0.035 0.035 1.014 0.006 0.025 Average tenure 0.000 0.000 0.319 1.000 0.000 0.286 Rural -0.485 0.199 0.015 0.917 0.033 0.016 Remote -0.965 0.225 0.000 0.811 0.038 0.000 Unemployment rate 0.031 0.020 0.115 1.005 0.003 0.094 Constant 5.783 1.970 0.003 Number of observations 27246 27246 R2/Psedo R2 0.03 0.067 Wald chi2(18) 1047.64 Log Likelihood -170272

Note: Standard errors corrected for heteroscedasticity. Table 6 Estimated effects of group interactions

Cross-sectional models Panel Movers

School OLS Poisson Pooled OLS

Fixed effects with Djt without Djt

Primary 0.141

(0.033) 0.021

(0.005) 0.141

(0.042) 0.091

(0.016) 0.071

(0.026) 0.071

(0.025)

Secondary 0.045

(0.083) 0.006

(0.013) 0.010

(0.037) 0.138

(0.040) 0.010

(0.108) 0.068

(0.089)

Combined 0.147

(0.029) 0.022

(0.004) 0.139

(0.035) 0.091

(0.015) 0.055

(0.021) 0.062

(0.021) Note: Robust standard errors in parantheses.

Table 7 Sick and healthy school characteristics (selected) Type of school: Healthy Intermediate Sick Average school absence days per annum 1.72 4.88 9.97 Primary school 0.94 0.81 0.81 School enrolment 105.6 440.9 472.9 Remote 0.28 0.11 0.09 Rural 0.44 0.28 0.21 Staff Proportion on temporary contracts 0.10 0.08 0.08 Proportion female 0.84 0.78 0.76 Average age (years) 37.93 40.95 42.21 Average tenure (days) 3130.91 4218.57 4667.24 Number of observations 290 571 287

Table 8 Testing for multiple equilibria

Estimated effects Coefficient s.e. t-statistic _A 0.12 0.05 2.42 School transition: origin to destination Healthy – Healthy 0.59 0.43 1.36 Healthy – Intermediate 0.37 0.44 0.85 Healthy – Sick 1.12 0.56 2.00 Intermediate – Sick 0.80 0.48 1.65 Intermediate – Intermediate 0.10 0.26 0.39 Intermediate – Healthy 0.57 0.56 1.02 Sick – Intermediate 0.09 0.49 0.18 Sick – Healthy 0.20 0.63 0.32

Table A1 A comparison of movers and stayers Movers Stayers Individual absence per annum 5.24 6.13 Age 39.10 41.23 Tenure 2750.53 4436.83 Male 0.27 0.26 NESB 0.06 0.07 Disability 0.05 0.06 Temporary contract 0.28 0.06 Part-time contract 0.17 0.13 Higher duties 0.17 0.05 School size 613.73 662.10 Proportion female at school 0.74 0.74 Average age at school 40.80 41.18 Average tenure at school 4191.17 4393.61 Rural school 0.21 0.19 Remote school 0.12 0.09 Unemployment rate 8.55 8.72 Average literacy (year 7) 676.56 681.15 Retention rate 74.50 75.96 Observations 968 29309