Report No. 19377-RU Russia Targeting and the Longer-lerm Poor (In Two Volumes) Volume I1: Annexes May 1999 Poverty Reduction and Economic Management (ECSPE) and Human Development Networks (ECSHD) Europe and Central Asia Region Document of the World Bank Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized Public Disclosure Authorized
148
Embed
Report No. 19377-RU Russia Targeting and the Longer-lerm …documents.worldbank.org/curated/en/541541468759003203/pdf/multi... · RUSSIA TARGETING AND THE LONGER-TERM POOR VOLUME
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
Report No. 19377-RU
RussiaTargeting and the Longer-lerm Poor(In Two Volumes) Volume I1: Annexes
May 1999
Poverty Reduction and Economic Management (ECSPE) andHuman Development Networks (ECSHD)Europe and Central Asia Region
Vice President Johannes Linn (ECAVP)Country Director Michael Carter (ECCRU)Sector Directors Christopher Lovelace (ECSHD)
Pradeep Mitra (ECSPE)Program Team Leader Hjalte Sederlof (ECSHD)
RUSSIATARGETING AND THE LONGER-TERM POOR
VOLUME II ANNEXES
TABLE OF CONTENTS
1. PANEL CONSTRUCTION AND ATTRITIONJeanine Braithwaite and Elena Glinskaya
2. ECONOMIES OF SCALE AND POVERTY LINESAnna Ivanova and Jeanine Braithwaile
3. CROSS-TABULATIONSAnna Ivanova
4. PROXY MEANS TEST REGRESSIONSJeanine Braithwaile and Anna Ivanova
5. WELFARE AND LABOR MOBILITY
Elena Glinskaya and Jeanine Braithwaite
ANNEX ONE
PANEL CONSTRUCTION & ATTRITION: RUSSIA 1994-96
JEANINE BRAITHWAITE (ECSPE)
1. The basic source of data for this report is a panel constructed from three rounds(waves) of the Russian Longitudinal Monitoring Survey (RLMS), which are publiclyavailable on the world wide web'. Households were matched to generate a panel of 2.675households for which data were available for each of the three years. Over the course ofthe three years, some households dropped out of the survey and others were added. Thepanel is constructed of households which were present in all three rounds and reportedenough information for their poverty status to be ascertained. There is a known danger ofbias owing to attrition, as households on the extremes of the distribution (i.e. the very richor the very poor) are more likely to drop out than others, although Glinskaya andBraithwaite (1997) found that attrition was not a significant issue for the RLMS panel(see below).
PANEL CONSTRUCTION
2. The primary motivation behind the construction of the panel data set is that it beuseable and accessible to scholars, analysts, and policy makers in Russia. Most scholarsand policy analysts are not that accustomed to working with large data sets on personalcomputers, and the data collection activities of Goskomstat Rossii are highly centralizedand dependent on main-frame computing time for processing.
3. In contrast, this study's panel dataset is made available as a public good (i.e. nocharges are required to obtain the data) for anyone who is interested in various definitionsof poverty and the poverty line. The original data set was made public by UNC on anInternet (World Wide Web) site, and this data set follows a similar the open accesspolicy (without being available on the web but happily supplied on diskette). So thatRussian scholars and policy analysts can quicklly open up a dataset and duplicate the mainfindings of this study, the following "short-cuts" were adopted.
The panel file is on the household level only. The individual files provided by UNCare extremely large and unwieldy. Interested parties can quickly match-merge theWorld Bank's version of the RLMS data set with the individual files by using theoriginal identification numbers found as "dupaid" "dupbid" and "dupcid".
The website is http://www.cpc.unc.edu/projects/rlms.
* Results and all cross-tabulations are run on only those households which are in thepanel every year, although cross-sectional data are available from UNC. Technicallyspeaking, this means that the World Bank version of the data runs some risk forattrition bias, which occurs as households drop out of a panel over time. Questions ofattrition are explored below.
* In the World Bank version of the RLMS data set, several variables have beenredefined, and the Bank poverty standard is based on household consumption, nothousehold income as in the UNC case. Since the household consumption variable isso significant, differences between the Bank and UNC are presented below. All of thevariables created by the Bank which are critical for the analysis, such as the definitionof the unemployed, disabled, and pensioners, are included in the panel file.
* The most important created variables are: total consumption, per capita consumption.per Goskomstat equivalent adult consumption, the cross-sectional poverty dummyvariables, and the poverty transition variable. For convenience, the study uses theseterns in the following way to characterize the various possible combinations ofpoverty status of the households over the three years as embodied in the povertytransition variable:
Longer-term poor: poor in every year (p-p-p)Never poor: not poor in every year (np-np-np)Escapedfrom poverty: poor in first year or first atnd second year, then not poor afterwards
(p-np-np or p-p-np)Fell into poverty. not poor in first year or first and second year, then poor afterwards (np-p-p,
np-np-p)Mixed: other patterns.
4. Additionally, exact patterns are indicated by symbols in the table column heading,with "p" designating poor and "np" for not poor. A notation such as p-np-np wouldrepresent escaping from poverty after the first year, while np-np-p would represent a fallinto poverty after the second year.
5. Poverty can be measured on either a household, individual, or population basis.Much basic poverty information is on an individual level, which is equivalent to tdepopulation as a whole if the sample is self-weighting. The RLMS sample is evaluatedvery positively by Heeringa (199?), in a report found on the RLMS Website.
6. UNC did include a vector of weights to be used when scaling up to the populationlevel, but these weights were found to have a negligible effect--cross-tabulations with andwithout the weights were equal to several decimal places, and standard tests failed todistinguish between outcomes with and without these weights. Therefore, the weightswere not further used in this analysis. However, attrition is potentially a more seriousproblem than inflating to the population level correctly, particularly since there is no clearway to remedy the problem (Deaton 1997).
2
7. Poverty in this study was based on a modified version of the consumption variableprovided by UNC in the data files.' The UNC( consumption variable was comprised ofpurchases of goods and services and the imputed value of food produced on the privateplot and consumed during the survey recall period. The imputation was based onpurchase prices collected in a community questionnaire. The modifications consistedprimarily of excluding savings and operations in foreign currency from the Bank'sdefinition of consumption. Household savings were non-existent or extremely low forpoor households and were overall quite low on average.
8. The poverty line used was a household-specific one, based on the officialsubsistence minimum calculated by Goskomstat Rossii. The subsistence minimum isdifferentiated for children, active-age adults, and adults at or past the statutory retirementage (60 for men, 55 for women), although it is usually published in a per capita form.The age-specific subsistence minimums were multiplied by the number of householdmembers in each category and summed to create a household-specific poverty threshold.This was then compared to the household's total consumption to determine whether thehousehold (and its constituent members) were poor or not. In this sense, our poverty lineincludes an embedded equivalence scale (Annex Two).
9. In the publicly-available data set, UNC provided additional poverty variables forregionally-differentiated subsistence minimum, but these were not used for the basicconclusions in this study. These LUNC poverty variables are aggregated for 8 regions ofRussia and are said to reflect differences in regional prices (Lokshin and Popkin 1998).However, they can not be used for this study as they are not equal to the actual legalsubsistence minimums in use in Russia during the study period (for comparison, theofficial statistics are presented in Annex Three).
10. First of all, the UNC variables do not correspond to the local subsistenceminimums used by some oblasts to allocate local social assistance or in 3 cases, the socialassistance pilot benefit. There are currently 89 such local subsistence minimums ascalculated by Goskomstat Rossii according to methodological instructions issued by theformer Ministry of Labor. There are additional regional variations as some areas (e.g.Moscow city and oblast) have chosen their own specific local standard. Second, in spiteof known considerable price variation in Russia, when checked previously for the 1994data, the 1995 World Bank poverty assessment found that there was essentially no majordifference in the overall headcount between comparing consumption to 89 individuallines instead of one national line. This is because although the cost of living undoubtedlyvaries across Russia, so do salaries and other sources of income, such that nominalconsumption levels tend to mirror price variation. While in principle it would be best torepeat this analysis on the panel data, it would require obtaining very detailed informationfrom Goskomstat Rossii on the 89 individual CPIs for the fieldwork period in each of
Technical detail on the consumption aggregate and other panel construction issues is provided in Annex One, while equivalence, thepoverty line, and sensitivity analysis are presented in Annex Two. For ease of exposition, only one poverty line is used in thebody of the study, the official prozhitochniy minimum (subsistence minimum).
3
three years, which are not routinely published, let alone the computational time to deflatethe data by oblast.
11. For example, but while the UNC consumption variable, correctly, did not includeexpenditures on the purchase of a house or apartment or of consumer durables such asautomobiles, it did include household expenditures on feed, seed, fertilizer, and otheritems used on the private plot. Typically, such expenditures are separated out from thehousehold's consumption, since thiey are investment for next period's consumption, andare handled through accounting for net profit from agricultural activities. However, giventhe timing of the fieldwork of the survey (during the Autumn of each year), this particularconcern about household expenditure on feed, seed, and other items used on a private plotis not a major problem for the analysis, since the vast majority of such expenditures occurin the Spring planting season, not the Fall harvest season.
12. One of the biggest problems in comparing poverty findings among differentstudies is that the choice of basis is quite significant for outcomes. Much of the UNCwork to date on its data set has been on a household reported income basis, but mosthousehold respondents in surveys in Russia and in other countries (including thedeveloped market economies) do not report accurately their income level. However,consumption which is drawn mainly from expenditure questions is much higher and morereliable than income in most transition and developing economy contexts, owing to thepervasive nature of the informal sector (Braithwaite 1995).
13. In the RLMS, there is a substantial degree of under-reporting of official income(Table A-1).
4
Table A-1. Russia: Income Under-reporting, 1994-1996
Table shows what percent of households who reported that their consumption was in a given quintile and who also reported enoughincome to place in that same quintileIncome deciles Consumption deciles
First 20 % Second Third Fourth Last 20%
1994
First 20 % 44.87
Second 26.65
T'hird 27.5
Fourth 30.64
Last 20% 55.16
1995
First 20 % 47.13
Second 28.55
'I'hird 27.17
Fourth 28.79
Last 20% 52.27
1996
First 20 % 44.52
Second 28.47
Third 28.51
Fourth 30.29
Last 20% 51.54Source: Author calculations from the RLMS.
Attrition
14. The number of households surveyed in each of the rounds (cross-sectionals) was3,762, 3,594, and 3,562 for rounds 5, 6, and 7 respectively. The absolute decline in thenumber of households interviewed reflects the problem of attrition noted above, while therewere other problems with household location and identification numbers that reduced thenumber of households available for the panel to 2,675. Of this number, two or threehouseholds lacked some key variable(s).
15. Attrition did not seem to affect many basic poverty findings such as povertycorrelates among household composition, location, durables, etc. However, some rates onthe population level are somewhat different between panel and cross-sectional data (TableA-2).
5
Table A-2. Russia: Unemployment and Disability Rates
Unemployed (not working, Individual (% of unemployed Household (% of households lhaving at least onedoes not receive a pension out of total population) unemployved out of all hIouseholds)and wants to work)
round Scross-sectional 9.5 23panel 8.8 21.6
round 6crnss-sectional 9.4 22.3pane! 8.7 21
round -cross-sectional 10.6 24.4panel 9.9 23.2
Disabled (receiving Individual (% of disabled out Househiold (% of hiouseholds having at least onedisability pension) of total population) disabled out of all htouseholds)
round 5cross-sectional 2.4 6.5panel 2.4. 6.6
round 6cross-sectional 2.5 6.5panel 2.6 7.2
round -cross-sectional 2.6 7.0panel 2.7 7.3
16. Some concerns can be raised in relation to the possible bias of the presentedresults due to the panel sample attrition. To test the robustness of our measures we ran aseries of binary probit estimations on the sub-samples of observations for variousdemographic groups. For example, to test the possible bias in the poverty assessment forthe pensioners on the panel data we ran a model with the dichotomous dependent variablewhich is equal to one for the pensioners who stay in the panel through last three rounds ofthe survey and is equal to zero for the pensioners who fall out of the sample. Asexplanatory variables we use the polynomial of the log of the total householdexpenditure for the last round of the survey and the log of household size. Thepolynomial form of the household expenditure allows to capture possible non-linearity inattrition bias. Similar estimations are run for the other demographic groups of interest.The results of binary probit estimations are presented in Table A-3.
6
Table A-3: Russia: Binary probit estimation of the possible attrition bias. World BankPanel 1994-1996.
Families of Nuclear Single parent All familiespensioners families familiesCoefficient Coefficient Coefficient Coefficient
* Significant with 90% probability** Significant with 95% probability**"Significant with 99.5% probability
17. The main conclusion is that attrition bias does not have a significant effect on thepoverty assessment results conducted on the panel sample verses the cross-sectionalsample (the coefficients on the total household expenditure variables are insignificant forall demographic groups and they are jointly insignificant also). However, there is apossibility for larger households to exit out of the panel sample disproportionately (thecoefficient for the family size is significant for the nuclear families and for all Russiasample). Disproportionate exit could lead to bias in the poverty findings for largerhouseholds.
Table A-4. Russia: Binary probit estimation of the possible attrition bias. World BankPanel 1994-1996.
* Significant with 90% probability; " Significant with 95% probability;" Significant with 99.5% probability; (),("),("') Jointly significant
7
18. Possible attrition bias in shown in the probit estimation by location of householdresidence (Table A-4). In all three locations, larger families are less likely to stay in thepanel sample. Total household expenditure does not have any significant effect on theprobability of being in the sample for households from metropolitan areas of Russia, butthe joint significance of the total household expenditure polynomial coefficients in otherurban areas of Russia and in the rural areas indicate a possible bias in poverty results forthese locations.
19. The negative combined coefficient on the total household expenditure variablesfor rural Russia implies that the poor rural Russian households are more likely to stay inthe panel sample and thus the poverty metrics will be biased toward zero for thesehouseholds, i.e., the poverty rate of rural households can be overestimated if the analysisis done solely on the panel sample. The positive coefficient on the other urban areas ofRussia expenditure indicate on the opposite picture. For this category we can expect toobserve a positive bias in the poverty measurements, or that the poverty rate of otherurban areas can be underestimated in comparison with the rate calculated on the cross-sectional sample. We do not observe any biases for the major metropolitan areas ofRussia based on the total household expenditure.
20. We also tested for attrition in a general, dynamic sense, starting from the initialpoverty status of the household. There seems to be no major differences in probabilitiesof exiting by the initial poverty status. 29 percent of the household which were initiallynon-poor exited, and 32 percent of the poor households exited. However, there are somedifferences in probability of attrition for the households of different characteristics. Thereare also seems to be differences in probabilities of leaving the sample for the same typesof households, but with the different poverty status.
21. Households headed by more educated individuals are more likely to exit thesample. This is true both for the households which were initially classified as poor and aswell as for the non-poor. Households headed by the young individuals which were abovethe poverty line in 1994 were more likely to exit than non-poor households headed by theolder individuals Households residing in the major Metropolitan areas are more likely tobe lost from the sample over time.
22. Poor female-headed households have a higher probability of exiting than non-poorfemale-headed households. Another substantial difference is the probability of exiting forpoor and non-poor households headed by the older heads.
23. Attrition is modeled below, and the Mills ratio for observing a household in thesample is included in the estimated equation. Sample weights were used in allregressions. The following computer output contains an estimation of equations relating"exit from the RLMS sample" to the set of family characteristics and to the indicators ofthe position in the distribution of expenditure in 1994. The first equation combines allobservations together, and includes controls for 5 initial expenditure quintiles.
The second and third equations are estimated for the "initially poor" household and "initially non-poor"households, respectively. By comparing the effect of the household characteristics on the exitprobabilities of poop and non-poor households one can say whether these characteristics have differentialeffects on the probabilities of exiting the sample.
gen not_att=0 if site5=.(924 missing values generated)
replace not_att= I if flag= = I¬_att=.(2675 real changes made)
tab not_att
not_att I Freq. Percent Cum.+-
0 l 1113 29.38 29.38 exited the sample during r6 or r7I 1 2675 70.62 100.00 stayed in the sample all 3 rounds
LABELSncatl_S # of small children in the hhncat2_5 # of 7-18 y.old children in the hh
# of 19-60 males in the hh omittedncat4_5 # of 19-55 females in the hhncat5_5 retired malesncat6 5 retired females
th 5 female headed hhrmh 5 retired female headed hhrth_5 retired female headed hhyh_5 young person headed hh
own aut5 = 1 if own autohhw han5 = I if handicaps in the hhhhw mat5 = I if a member on maternity leavehhw_une5 = I if a member is unemployed
hh head age groupIhhh_1 18-23Ihhh_2 24-29Ihhh_3 29-34Ihhh_4 35-39Ihhh_5 40-44Ihhh_6 45-49Ihhh_7 50-55Ihhhh_8 56-60
older then 60 omittedhh head education group
Ihhh_e_l ptu or lowerIhhh_e_2 technicum
university omittedregions, Moscow and St. Petersburg omitted
reg2
15
reg3reg4reg5reg6reg7reg8
Isettl_2Isettl_3Ie_pcq_1 lowest pc expenditure quintileIe_pcq_2le_pcq_3Ie_pcq_4
********* * *****
thie following is what I had written up:
The first observation is that probability of moving out of the sample are quite close: 0.32 for thehouseholds below the poverty line and 0.29 for the households above the poverty line. This suggests that,on average, no systematic attrition on the basis on income is happening. However, that might not be truefor the households with the particular characteristics (as we show below), and it is still necessary tocontrol for attrition.
18 percent of the 'initially' poor households exited poverty and did not return to poverty during theobserved period of time. 30 percent of households which were poor in 1994 stayed poor throughout allobserved periods. 41 percent of initially wealthy households did not experience a poverty spell throughoutthe time of the survey.
Probability ot leaving the sample in 1995 or in 1996Conditional on being below the poverty line in 1994 0.32
Probability ot leaving the sample in 1995 or in 1996Conditional on being above the poverty line in 1994 0.29
16
Annex Two
Economies of Scale & Poverty Lines: Russia 1994-96
ANNA IVANOVA (U.WISC.)JEANINE BRAITHWAITE (ECSPE)
I. The purpose of this annex is to lay out the most important issues on economies ofscale and the poverty line, since these two parameters are critical for our povertyconclusions. For our poverty and inequality analysis an important question was toidentify resources available to each household member taking into account possibleeffects of economies of scale and the composition of the household. It is extremelydifficult to answer this question without reliable data on the distribution of individualconsumption within the household (information which shows whether one householdmember such as an active-age male consumes more food and other goods than otherhousehold members such as children). This is sometimes called the "unitary household"problem, and results from the extreme difficulty of collecting (or observing) reliableinformation about individuals, especially children. It is possible to approximate resourcesavailable to individual household members when making certain assumptions about theallocation of resources within the household. There are three approaches which haveoften been used: the Engels curve method, the Rothbart approach, and using informationabout subjective perceptions of poverty.
2. Under all the approaches, we can assume that each member consumes an equalportion of available resources and that there are some economies of size in householdconsumption (resulting from the presence of public goods). Economies of size inhousehold consumption mean that the marginal increment to total household expendituresof an additional household member declines with each subsequent. In more prosaicterms, the idea is that a large family can "stretch" a stew to feed one more by addingpotatoes instead of meat, children can wear hand-me-downs, and that the total cost of rentand utilities is either fixed or varies little with an additional person, so that the marginalcost of the additional person is lower (because the average cost is the total cost divided byan increasing number of members).
3. Besides this effect of size, there are the questions of equivalence mentioned in thefirst paragraph. Equivalence in this sense means determining what fraction of theconsumption of one household member (usually taken as an adult male for convenience)is covered by the consumption of other members. For example, one could expect thatchildren would consume less food than parents, and that an employed woman would havemore expenditures on professional clothing, transportation, and meals consumed outsidethe home than a retired grandmother also living in the household. In this case, thechildren and the pensioner would have lower consumption than the adult female in this
household, and their consumption could be said to be equivalent to (for example) 50 and70 percent respectively of the adult female's consumption. Although in principle weagree with the idea that all household members do not consume exactly identical amountsof total household consumption, particularly in developed market economies, we are lessconvinced about the extent of equivalence in many transition economies.
4. There are several factual observations about relative prices and actual payments intransition economies, as well as cultural specifics, which raise questions as to the actual(as opposed to theoretical) extent of equivalence in household consumption in transitioneconomies. Armenia provides a good illustration of these considerations. When Armeniafirst began its economic reform program with the World Bank in 1994. no one paid anyrmoney period for rent or utility payments even though these fixed prices were essentiallyzero when compared to the hyperinflated prices of food. The rent and utility paymentcollection systems had broken down during the armed conflict with Azerbaijan (1992-94)and the country was almost completely blockaded from land or rail freight, and theairport needed reconstruction before it could handle significant air freight. Thepopulation was subsisting almost exclusively on international and private humanitarianassistance (World Bank 1996).
5. However, at that time, neither the international non-governmental organizationsnor the World Bank could find any indication of widespread or even pockets of grosschild malnutrition (wasting), and little indication of prolonged child malnutrition(stunting). How could this be? The answer is found in Armenian cultural values, whichput children above all others in the family and society. Grandparents were semi-starvingthemselves in order to give food to their grandchildren and parents were also restrictingtheir intake. CARE documented significant and widespread weight loss among theArmenian elderly.
6. In such a situation, the standard OECD assumption that equivalence for childrenis 50 percent of adult consumption, and for the elderly, 70 percent, is clearly wrong. Inthe Armenian case in these terrible years, children were consuming at least 100 percent ofadult consumption, and the elderly were consuming imluch less, arguably under 50 percentof adult consumption. Unfortunately, it is extremely dit'ficult to measure what actualindividual equivalence is in general, and certainly not in the Armenian case given thevery limited survey information available.
7. Furthermore, it is quite difficult to argue that there were significant economies ofsize in consumption in Armenia from declining marginal contribution of a householdmember to average cost of rent and utilities, since hardly anyone paid any rent or utilitybills during this period. Even for the cost of heating it would be difficult to positeconomies of size, since the usual practice for a group of families living in an apartmentbuilding was to rotate heating responsibilities in the following way. Seriatim, familieswould purchase coal or wood and heat one room of their flat, in which every other personin the building could sit in the semi-heated semi-darkness. The formula for cost sharing
2
was rotation, or to think of it in another way, a flat fee per household which was invariantto the number of household members.
8. Fortunately, the government's economic reform program was successful, thehyperinflation was stopped, the blockade slightly reduced to allow land transportationthrough Georgia, the decision to restart the nuclear reactor supplied the population withelectricity, and a new approach to ensuring utility payments collection was adopted. Nowin 1998, it could very well be the case that there are significant economies of size, drivenby the impact of the sharply increased (and actually paid!) cost of utilities and rent.
9. There are some similarities to the Russian case. During 1994-96, the poor inRussia spent about 75 percent of total consumption on food and even the average sharewas around 60 percent, reflecting relative prices of food and non-food goods. In January1992, the relative price of food skyrocketed as the Government increased prices toeliminate (or greatly reduce) an extensive system of generalized consumer subsidies.However, the price of other items, notably heating oil, gasoline, utilities, and rent, wereessentially frozen in real terms, making them relatively very inexpensive. Even so,households began to fail to pay their utility and rent bills.
10. According to our data for 1994-96, Russian expenditures were primarily on foodand people mostly avoided purchasing discretionary items like clothing and consumerdurables, whose relative price had increased sharply. Spending on fuel, utility, and rentwas almost zero, reflecting the extremely low controlled price of these items during the1994-96 time period as well as the tendency for households to avoid paying these bills.As a result, in this environment, the cost of an additional household member basicallyamounts to food cost, and one has to make very strong cultural assumptions that parentswill not spend as much to feed a child as they do for themselves. Given what we know ofRussian culture and the relative price of non-food child goods (such as the true cost of apediatrician which includes the controlled price plus the under-the-table "side" out-of-pocket payment), it seems highly unlikely that the equivalent consumption of childrenwould be very significantly below that of adults. This conclusion is supported by theRussian literature and by the official "subsistence minimum" methodology, whichsuggests that the equivalent consumption of children is 90 percent of adult consumption(see below).
11. Nonetheless, it would be useful to formalize the equivalence and size issues and touse our data to empirically verify these suggestive arguments about consumption intransition economies. Next, we lay out the model and explain one approach to estimatingsize elasticity (economies of scale (size) in consumption) in Russia.
yEquivalent resources per household member can be written as -6 where Y are total
noresources of the household, n is household size and 0 is a parameter indicating economiesof size. Another possibility would be to account for the difference in consumptionbetween adults, children and elderly assuming that all adults (children/elderly) consume
3
equal proportions of total consumption. Then we could express resources per equivalent
adult as , where nad is the number of adults in the household, nCh is the(n,,/ + an.,, + 8n,,,,)
number of children in the household, and ne,ld is the number of elderly in the household(the method that is used in the official poverty line for Russia) or we could also accountfor gender differentiation (e.g. resources per equivalent male adult).
12. An important question that arises then is how to chose parameters 0. cc and P. Weconcentrate here only on the discussion of parameter 0 since aX and ,B basically reflect thedifferences in nutrition requirements for children and elderly versus adults and for theseparameters we used values implicitly incorporated in the official (Ministry of Labor ofRussia) methodology for computation of poverty lines, namely cx= 0.9 and ,3=0.63. Inorder to estimate 0 we need to rank households according to their level of well-being suchthat we could identify families with the same level of well-being but different resources,size and composition. This is not an easy task and there is no consensus in the literatureon how to reliably perform this kind of estimation because of the difficulties ininterpreting and measuring well-being.
13. There are two basic approaches to assessing well-being: welfarist (comparingwell-being based solely on "'utility" levels e.g. self-evaluation of people) and non-welfarist (comparing well-being with little or no emphasis on utility, e.g. using specificcommodity deprivation). These two approaches combine in the "standard of living"approach, where a person's standard of living is viewed as depending solely onindividual consumption of private goods (although public goods can also be included)and current consumption is considered to be the preferable indicator of well-being. Thisapproach can be both: welfarist which emphasizes aggregate expenditure on all goods andservices and non-welfarist which emphasizes specific commodity forms of deprivatione.g. inadequate food consumption.
14. The "standard of living" approach has been more popular in developing countries.The popularity of this approach in developing countries "reflects the greater importanceattached to specific forms of commodity deprivation, especially food insecurity and thatemphasize is quite defensible from both welfarist and non-welfarist points of view"(Ravallion 1992). This emphasize seems also be applicable to Russia in which foodexpenditures on average comprise 60-70% of total household outlays. Therefore, ourpoverty analysis for Russia was based on the "standard of living" approach . Moreover,since as it can be difficult to choose between welfarist and non-welfarist approaches wetried to employ elements of both. We used some elements of the non-welfarist approach(Engles curve estimation, based on share of food in the total household expenditures andshare of a basic consumption bundle of food, rent and utilities, and clothes, out of thetotal household expenditures) when estimating the size parameter, and an element of thewelfarist approach (total expenditures per equivalent adult as an indicator of well-being)when computing poverty estimates.
4
CRITIQUE OF ENGEL'S METHOD
15. Engel's method for estimating 0 has been criticized for several reasons. First ofall, it was noted that it suffers from an identification problem. As Deaton emphasized(1997, p.268-269) two different cost-of-living functions which have differentimplications for household welfare, for example, c(U,p,n)=n6cx(p)UW(P) and c(U,p,n)=n8 0'4 IIU x(p)UO(P) (the latter reflecting the fact that "additional people may not affect costsproportionally but have larger or smaller effects the better-off is the household") willyield the same budget share equation in Engel's method. The demand functions will notbe affected by the presence of the second term 31lnU while welfare levels are effected(p.269). So ultimately some parameters of well-being will not be identifiable fromdemand behavior.
16. The following argument regarding Engel's method was also emphasized byDeaton and Paxson (1998) as paraphrased here: Economies of scale are more likely to beobserved in the presence of household public goods which could be shared within thehousehold and do not need to be replicated for each household member. Resourcesreleased by sharing could be spent on both private and public good (income effect). Onthe other hand, there will be also a substitution effect towards shared goods since nowthey are cheaper for members of the household. But if there is a private good that is noteasily substitutable, with low own and cross-price elasticities, the income effect willdominate and per capita consumption on good will increase. The best candidate for such aprivate good is food, especially among poor households whose consumption of food takesa high share in their total consumption. Then, with per capita resources held constant,food consumption per head should rise with household size. Failure of this prediction (ifany) is most likely among rich households whose food needs are well satisfied.
17. To continue the paraphrase: The Engel's method is a direct contradiction to this.Since it is a well-acknowledged fact that food share falls with increase in resources at firstit seems that the food share should also fall with increase in household size which wouldrelease some resources in the household (in fact, in its simplest version Engel's methodgives positive estimates of the economies of scale i.e. 0 < I only if coefficient onln(household size) is negative (provided that the coefficient on the ln(PCE) is negative asit is well-believed) since 0 = I - P.,1size1ppce and in order for 0 to be less than I f3hhsize/pce
should be positive but 3p,e is negative so should be Phhsize. But holding per capitaexpenditure constant food share would decrease with household size only if per capitafood expenditures would decrease too since food share can also be expressed as the ratioof per capita food expenditures to the per capita total expenditures which is not at all whatwe would expect in the presence of economies of scale as it was outlined above - percapita food expenditure should increase with increase in household size (end paraphrase).
18. However, the empirical evidence presented in the above paper (in "rich" countriessuch as US, and France as well as in "poor" countries, such as Pakistan and South Africa)is exactly the opposite: with total household expenditures per capita held constant,
5
expenditure per head on food falls with increasing household size which supports Engel'smethod but for the "wrong reason".
19. Despite the above mentioned critique and many others, the profession has yet tofind an accepted substitute. Although some recent findings involving self-evaluation ofpeople for assessing their level of well-being suggest an alternative way for estimation ofeconomies of scales, subjective measures can not be taken as a full-time substitute forobjective indicators such as consumption behavior as here the question on to which extentpeople know what is best for themselves arises.
20. "There are situations where personal judgments of well-being may be consideredsuspect, either because of miss-information or incapacity for rational choice even withperfect information." (Ravallion 1992). Moreover, when subjective and objectiveindicators contradict to each other (as it seems to be the case with Russia. for which mostrecent findings using subjective welfare measures suggest that 0=0.4 (Ravallion andLokshin 1998) rather than 0=1 as identified here) one has to answer the followingquestions "Are there reasons why consumption behavior is misguided, such as due to theintra-household inequalities?" (which contradicts an implicit assumption that we madewhen constructing equivalence scales, e.g. we assumed that all adults equally share inhousehold consumption which may not be the case in reality) "Is it an issue of imperfectinformation? Or is it a more fundamental problem, such as irrationality or incapacity forrational choice (such as due to simply being too young to know what is good for you),and not having someone else to make a sound choice?" (Ravallion 1992).
21. The subjective measure in Ravallion and Lokshin (1998) is more akin to a relativepoverty line which has been applied more often in developed rather than developingcountries. The question used for identifying the poor was as follows: " Please, imagine a9-step ladder where on the bottom, the first step, stand the poorest people, and on thehighest step, the ninth, stand the rich. On which step are you today?". Clearly, thismeasure reflects a relative position in the perceived distribution and not the actualposition in the actual distribution. In the case of Russia with highly unstable economyand constant reevaluation of existing standards in the society including the "standard ofliving" in the course of transition from socialist to market economy, this perception islikely to reflect not only past and current state but also expectations about the futurewhich are more often gloomy rather than bright.
22. There is ample evidence from the psychological literature to suggest that peoplehave a more difficult time evaluating their current situation when it is sharply differentthan the past. The World Bank has conducted participatory assessments in severaltransition countries now, and one clear theme is that most people feel very bitter that theyhave become so relatively poor since the transition. Indeed, the fact that so few of thesurvey respondents said that their households were in upper ranges of the distribution ofconsumption (income) as indicated by the absence of responses on the upper rungs of theCantril ladder suggests that people in Russia are having a very hard time recognizingwhere they really are in the current situation. If people had more accurate self-
6
perceptions, then the richest households would rate themselves in the top decile (topladder rung), but there were no responses in the highest decile at all (Ravallion andLokshin 1998).
23. It may be a question of perspective as most adult respondents are old enough tohave either experienced the Gulag years or to remember relatives lost to the purges & theGulag. It is hard to argue that such respondents (the majority of the adult respondents)are going to feel very inclined to answer the Cantril ladder question favorably. Stickingout above the crowd is a big taboo for most ordinary Russians.
24. Moreover, differences in perceptions are more likely to occur between oldergeneration who are more conservative and more reluctant to accept changes in the societyon the way to a market economy (this may also explain the difference in findings aboutpoverty among pensioners by objective and subjective indicators). Another problem withthis measure (as well as with income measures) is that when answering direct questionabout their poverty state, people may not tell the truth for several reasons (trying toconceal illegal or unreported income, stigma and embarrassment, hope that theinterviewer might be a source of private transfers, or other causes).
25. Given these concerns about the subjective approach, and not ignoring the recentassessments by Deaton (1997) and Deaton and Paxson (1998), we have still decided to trythe Engels curve estimation. The reasons for doing so are as follows:
* The identification problem can not be avoided and this is a clear drawback of Engelsmethod. But we could try to use a combination of measures such as share of food intotal expenditures and share of basic bundle (food, clothes and shelter) in totalexpenditures which allows us to infer more information about household's well-being(despite including in the latter such a public good as rent and utilities our estimates of0 did not change - both of them indicated 0=1).
* While in empirical finding by Deaton and Paxson, the direction of effect of householdsize on the demand for food in Engel's regression seems at first to contradict the mereidea of economies of scale, this actually requires certain assumptions which in turnmay not be uncontroversial.
* First, people are assumed to spend money released from "savings" from scaleeconomies on food which may not be the case even among poor households. Poorhouseholds could keep food spending constant and increase spending on clothes,using savings from economies of scale.
* Second, consumption is measured by money spent, not physical quantities consumed,and it may be the case that prices matter i.e. larger households may be able to acquirefood stuff at lower prices (bulk discounts) and then while their consumption does not
7
change or may even rise, the amount spent on food may fall.' This may happenbecause larger households are able to obtain better information about prices in themarket or may be able to use bulk purchases which are effectively cheaper (howeversuch bulk discounts may not be that important for Russia which lacks a well-developed retail trade network outside of Moscow or St. Petersburg).
* Of course, it would be interesting to see what would be the effect of household sizeon the share of the entire basic bundle (not only food) incorporating information onprices (say, if quantities consumed are available or can be computed from the surveydata then one could use them and evaluate total consumption at the same prices foreverybody ). Maybe in this case and using the share of the basic bundle rather thenjust food, using Engel's method would not provide contradicting results. Of course,this is a very tedious procedure and we did not have the opportunity to test thissuggestion. Instead, we used the amount spent and found that for Russia, householdsize has no effect on the amount spent on food, clothing and shelter which means thatif we replicate resources and people in the household, the per capita amount spent onfood, clothes and rent and utilities does not change.
26. Since currently we do not see any better method for estimating economies of sizeparameter in Russia and the "puzzle" of contradiction in Engel's method still needs to beresolved we based further analysis on the results obtained with this method.
ESTIMATING ECONOMIES OF SIZE: ENGEL'S METHOD.
27. Engel's method for estimating 0 (controlling for composition effects) was widelyused until recently. The Engels curve approach is quite straightforward, and is based onthe observation that ceteris paribus, richer households spend a lower percentage of theirtotal expenditures on food than do poorer households. From this, the argument wasextrapolated that the share of household expenditures on food could be used as anindicator of material well-being, and that all households spending the same share of totalexpenditure on food would have the same standard of living (Deaton and Muelbauer,1980). Following this approach, it is possible to generate estimates of the size parameter0, which was done below. Unfortunately, the Engel's method no longer commands thewidespread acceptance of only a few years ago (Lanjouw and Ravallion 1995 albeit withstrong caveats), and Deaton (1997) and Deaton and Paxson (1998) strongly suggestavoiding the method for estimation of the size parameter. Given the lack of consensusabout a suitable alternative to the Engel's method, we have used it to check our sizeparameter, but we must point out that more than the usual caveats now apply to suchest]imations.
28. We have estimated two specifications of Engel's regressions on the RLMS data.
This suggestion originated with Ruslan Yemtsov, whose comments here & elsewhere are appreciated.
where o is a vector of (oi - the budget share devoted to food, namely for Russian data (forexample, for round 5) it was computed as follows:
budget share devoted to food by household i in round 5 = (alcohln5 + breadn5 + dairyn5+ eatoutn5 + eggsn5 + fatn5 + fishn5 + fruitsn5 + hprgncn5 + meatn5 + ofoodn5 +potaton5 + sugam5 + vegetn5)/totexpn5
and
alcohln5 - total hh alcohol expenditures: nominal, round 5breadn5 - total hh bread expenditures: nominal, round 5dairyn5 - total hh dairy expenditures: nominal, round 5eatoutn5 - total hh dining out expends: nominal, round Seggsn5 - total hh eggs expenditures: nominal, round 5fatn5 - total hh fat expenditures: nominal, round 5fishn5 - total hh fish expenditures: nominal, round 5fruitsn5 - total hh fruit expenditures: nominal, round 5hprgncn5 - tot hh home prod gross, non cash, evaluated at prevailing market prices:nominl, round 5meatn5 - total hh meat expenditures: nominal, round 5ofoodn5 - total hh other food expends: nominal, round 5potaton5 - total hh potato expenditures: nominal, round 5sugarn5 - total hh sugar expenditures: nominal, round 5vegetn5 - total hh vegetable expenditures: nominal, round 5totexpn5 - total expenditures of the hh: nominal, round 5 (total household monetary foodand non-food expenditures excluding big purchases, purchases of luxury goods,bonds/stocks and savings plus value of home-produced food evaluated at prevailingmarket prices)
X* is a matrix containing the following variables in columns:* oa - a constant* ln(per capita expenditure) = ln(x,/ nj) where
9
x; - total household expenditures (e.g. for round 5 totexpn5)n,- household size (e.g. for round 5 hhsize5)
* ln(n,)* n j- proportion of household members in a given demographic group j
K -=8 number of demographic groups. Namely, the following demographic groups wereused (e.g. for round 5):
fO_13_5 - women in age group 0-13f14_25_5 - women in age group 14-25f26_p5 - women in age group 26-55felder_5 - women in age group 55 and oldermO_13_5 men in age group 0-13ml14_25_5 - men in age group 14-25m26_p_5 - men in age group 26-60melder_5 - men in age group 60 and older
The demographic group ml 4_25_5 was excluded from regression to avoidmulticolliniarity and should be viewed as a reference group when analyzing coefficientson other demographic variables.
29. An alternative specification was also estimated.
Specification II
) K-1~~~K-(n°-)Z ,5 {n,, + relative prices+ El
= a +,B* In( +, * (I - -*) ln(ni) + E±5 *,n,, + relative prices + ,=
wherenij numbers of household members in a given demographic group j
Let us denote the coefficient before ln(x,/n1) as Ppce and the coefficient before ln(n,)as 3hh,size then in specification I f3hhsize4=pce0 -0) and, therefore, 0= 1 - PhNiAN/ Inspecification II theta was estimated as follows:
10
E 17ij E a r7y F. E X*e
9=0*iJ=: O* J1 = i='1;0=0S*- i=' n, * - n, = 1 ,8st n,
evaluating mjj and ni at their sample mean points.
30. If the estimated 0 appears to be equal to I then there is no economies of size inhousehold consumption and a per capita consumption poverty standard is appropriate(composition effects are controlled for in this regression). But in order to make inferencesabout theta we need to compute its variance which could be done through the delta-method as 0 is a non-linear function of estimated coefficients.
This method yields the following formula for estimating variance of theta:
[ar() £99 £9dO0 i var(/3,1,,.,) cov(I30,h,i: i,.p((. ) ,var(0) X ,,, iLcov(,8,,,,; ( /r, ) var(=/).() £90
where derivatives and variances are evaluated at estimates of r31111i1e and 3pce.
For specification I we haveS ~ ~I oS A,I,.i
- =_pand 2
and for specification Il:
£99 I £9l9 +8I:6 +(#i272)=-- and =
31. To estimate coefficients P.,, and , we could use ordinary least squares (OLS)regressions provided that all the dependent variables were correctly measured.Unfortunately, per capita expenditures are most likely measured with some error. Thismeasurement error would bias the coefficient p,, towards zero if this measurement errorwere not correlated with the error terms in the regression (£j ). But almost certainly, theerror terms are correlated since the food share and per capita expenditure are computedusing the same information on total household expenditures. Since we do not knowwhether the correlation is negative or positive (it depends on which measurement error isbigger - for food or non-food items) we can not a priori predict the direction of bias in
11
,0pc. Therefore, we need to find an instrumental variable (IV) in order to obtain unbiasedestimates of the coefficients. Household income is highly correlated with expendituresand since it is measured in a different way from expenditure (excluding income fromhome produced food which is an imputed term in both income and expenditures andwhich could introduce common errors) it is proposed that measurement errors in cashincome and expenditure are not correlated. Therefore, we could use cash per capitaincome as an instrumental variable for per capita expenditures.
32. For testing the estimate of 0 we also need to estimate the variance of thecoefficient estimates, checking for heteroskedasticity, as regressions on cross-section dataare often found to be heteroskedastic (Deaton 1997, p. 27). To avoid pre-test bias,White's test on heteroskedasticity was performed on data from round 5 while the fullmodel was subsequently estimated on data from rounds 6 and 7. The White's test consistsof regressing squared residuals from the above regression (where instead of an OLSestimator, the IV estimator should be used) on all unique variables
in X 0 X instrumenting the natural log of per capita expenditure by ln(per capita cashincome) and testing whether this regression is vacuous i.e. whether all coefficients arezero.
33. For this purpose, the squared residuals were regressed on the following 66variables: 11 original variables in the matrix X* (a constant and 10 other variables), I0squared original variables (a square of a constant is a constant itself and, therefore, it doesnot represent a unique variable) and 45 cross-terms (a constant multiplied by any othervariable gives the variable itself and again these cross-terms do not represent uniquevariables, therefore, number of cross terms: I0!/(2!8!)=45)). Under the null hypothesis ofhomoskedasticity, nR' has a chi-Squared distribution with 64 degrees of freedom (n - thesample size for round 5 was 2,572 households). The results of the test are presentedbelow:
R-squared 0.11585 Mean dependent var 0.0445Adjusted R-squared 0.092238 S.D. dependent var 0.077125S.E. of regression 0.073482 Akaike info criterion -5.19538Sum squared resid 13.1427 Schwarz criterion -5.04163Log likelihood 3012.879 F-statistic 4.906544Durbin-Watson stat 1.971522 Prob(F-statistic) 0
34. Thus, R2 was equal to 0.12 and nR2 to 289.62 with corresponding probability oftype I error i.e. the probability of rejecting the null-hypothesis when it is true, equaledzero. Therefore, we can reject the null hypothesis and recognize this regression asheteroskedastic. Although OLS is inefficient in this case, it is still a consistent estimatorof coefficients. But correction of the standard errors is needed. Therefore, we usedWhite's heteroskedasticity consistent variance estimator when estimating the model ondata for round 6 and 7. The results of estimation are presented in Tables B-2 and B-3below.
Table B-2. Round 6. Regression of budget share devoted to food by houselholdTSLS 1/ Dependent Variable is SHFOOD 6
R-squared 0.050887 Mean dependent var 0.707292Adjusted R-squared 0.046984 S.D. dependent var 0.211948S.E. of regression 0.206909 Akaike info criterion -3.14647Sum squared resid 104.1167 Schwarz criterion -3.12035F-statistic 33.84636 Durbin-Watson stat 1.877081Prob(F-statistic) 0
Table B-3. Round 7. Regression of expenditure share devoted to food by household.TSLS 11 Dependent Variable is SHFOOD_7
R-squared 0.121638 Mean dependent var 0.676682Adjusted R-squared 0.117804 S.D. dependent var 0.216952S.E. of regression 0.203773 Akaike info criterion -3.17674Sum squared resid 95.12972 Schwarz criterion -3.1493F-statistic 31.0705 Durbin-Watson stat 1.788962Prob(F-statistic) 0
15
35. Analyzing these results we can estimate 0 as I - I,Iislze/p'pce= I - (-0.001 04)/( -0.13481)=0.99 for round 6 and 1-(-0.02574)/(-0.II146)=0.77 for round 7. To test whetherO is equal to 1 we have to compute the variance of theta (the tables presented thevariances Of Phhsize and Pc, only). The relevant parts of the variance-covariance matrix forthese two coefficients for rounds 6 and 7 are:
Table B-4: Variance-covariance matrix elements for testing hypothesis 0=1
Round 6C LNHHSZ6 LNPCEX6
C 0.039776 -0.00148 -0.00297LNHHSZ6 -0.00148 0.000197 9.70E-05LNPCEX6 -0.00297 9.70E-05 0.000233
Round 7C LNHHSZ7 LNPCEX7
C 0.030472 -0.0012 -0.00223LNHHSZ7 -0.0012 0.000199 6.95E-05ILNPCEX7 -0.00223 6.95E-05 0.000176
36. The corresponding estimates for the variance of theta calculated by the delta-tnethod are as follows: for round 6 var(0)=0.01 1 which yields the Wald test statistic of0.0089 for round 6 when testing the null hypothesis that 0=1 and for round 7 var(O)=0.014wvhich yields a Wald test statistic of 3.75. Under the null hypothesis of 0=1, the Wald teststatistics have a Chi-squared distribution with one degree of freedom (in case of a singlecoefficient the square root of the Wald test statistic is equivalent to the t-statistics whichare 0.094 and 1.93 in round 6 and 7 respectively). In both cases at 5% level ofsignificance we can not reject the null-hypothesis and can view 0 as indistinguishablefrom unity (Table B-5). This means that there are no significant economies of scale inconsumption.
37. As it was mentioned, another specification of the above regression (specificationIl) was estimated as well. For rounds 6 and 7, the corresponding estimates of 0 were 0.99and 0.92 (we have not computed the variance here).
38. Furthermore, similar regressions to the above were run with the share of the basicneeds bundle, namely, expenditures on food, shelter (rent and utilities) and clothes intotal household expenditures as the dependent variable. In this case, estimates for round 6and 7 were 0.97 and 0.74 and their variances were 0.013 and 0.022 respectively. Againthe Wald-test statistics which were 0.0651 and 3.07 (corresponding t-statistics are 0.26aind 1.75) show that at 5% significance level the hypothesis of 0 being equal to I can notbe rejected.
16
39. It should be also noted that in all cases the coefficient on the logarithm ofhousehold size was, insignificant which allows us to at least make conclusion about theeffect of household size on the demand for food and basic bundle in the household i.e.there is basically no effect. This result is different from what was found for othercountries (including some developing countries) in "Economies of scale, household size,and the demand for food" (Deaton and Paxson, 1998 mimeograph) where the food sharewas found to be negatively correlated with household size in most cases except in GreatBritain where the coefficient on logarithm of household size appeared to be insignificantas in our finding. It should also be noted that although in the above paper the Engel'sregression was used in a quite different context it was found that the coefficients in thisregression are not greatly affected by the choice of functional form for per capitaexpenditures and in most cases (except Pakistan and South Africa) instrumental variablesestimates are not significantly different from OLS estimates.
Table B-5 Russia: Engel's Estimates for Economies of ScaleSpecification I Specification 11
Regression coefficients Regression coefficients
Share of food was regressed on q Share of food was regressed on qdemographic variables, relative demographic variables, relative
*In all cases coefficient on ln(household sizc was insignificant)
17
POVERTY LINE USED IN THE STUDY
40. The measure of well-being which we used to identify the poor for Russian povertyy
assessment was as follows: resources per equivalent adultn,, + 09nh, +0.63n,,
where Y are total resources of the household, nad is the number of adults in the household,nCh is the number of children in the household, and nCId is the number of elderly in thehousehold and these resources were compared to the regional poverty line for adultsconstructed for eight regions of Russia as a population weighted average across 78official regional subsistence minimum for adults (Table B-6) (to match the sample)'. Itshould be noted though that depending on the choice of poverty line and parameter 0conclusions about poverty composition and rates may significantly vary.
41. It should be noted that the coefficients were used were constant, calculated fromthe official data for 1994. Goskomstat Rossii provided information on average povertylines for the 89 oblasts of Russia, but not the detailed information on how this average isbroken down into the child, adult, and elderly subcomponents for 1995 and 1996. Sincethese coefficients have changed very little over the time of the study, we simply appliedthe 1994 breakdown to 1995 and 1996. Strictly speaking, we should have requested theadditional information from Goskomstat Rossii and used slightly different breakdownsfor 1995 and 1996.
42. Relative prices were computed based on the regional poverty lines calculated bythe Ministry of Labor (prozhitochniy minimum). Eight regions were used for regressionanalysis:o Major metropolitan areas (Moscow and Moscow region, St. Petersburg and
Leningradsky region)* North and Northwestern regions* Central and Central black-earth regions
Volga and Volgo-Vyatsky regions* North Caucasus* Ural* West Siberian region* East Siberian region and Far East.
43. Relative prices and regional poverty lines calculations are presented in AnnexThree, pages 15-17.
44. Table B-7 below presents average household size of the poor and non-poor aswell as poverty rates for different values of 0 (a sensitivity analysis for 0). Lanjouw,Paternosto and Milanovic (1998) suggested that a critical value for 0 would be 0.7*
2 Division into regions
18
check, at which there would be reversals of policy advice to target children over theelderly. At such a 0, the poverty rate would be 25.* percent. It is important tounderstand that policy reversals are not only driven by estimates of 0 but also by wherethe poverty line is drawn. For example, at a very high poverty line (many of thepopulation would be poor), the share of elderly in the poor might be much higher, whileat a very low poverty line (very few of the population would be poor), the share ofchildren in the poor could be very high. This finding demonstrates the process of whichoverall changes in headcount fluctuate with applying higher at lower poverty lines (TableB-8). The table demonstrates that there is definitely some bunching of people around theimmediate poverty line, since a proportional change in the poverty line is exceeded by theresulting change in the headcounts. For example. reducing the poverty line by 10 percentresults in a 15 percent reduction in the resulting headcount. Although some bunching isclearly noticeable, the extent here seems to be in the distribution of consumption less thanobserved for the early rounds of the RLMS (World Bank, 1995).
19
Table B-6
Average regional Average regional Average regional Relative prices (base - Relative prices (base - Relative pricespoverty line, poverty line, Mintrud. poverty line, Mintrud, Volga & Volga-Vyatka), Volga & Volga- (base - Volga &
Notes: First column is baseline estimates for the Russian Povert) Assessment Update.
Columns labeled Poverty Line = show the headcounts that would result if a poverty line x percent lower or higher were used.
TIhe remaining columns show the percent change in headcounts (retferencing the baseline estimates in the flirst column) that resulted fromapplying a lower or higher poverty line
O,O;O'O O O O 0'0I0000 0 0 -. ID00000000000 o o o o o ooo expenditures on rent and utilities - '-ai
C )0 !~ 0 CD 0 0 C)a
* o'o olo o o o o o o a) Do o 010 0: 0 0 0 oo 0 expenditures on services ,
0D 0). 00 VlI c.n cn cn - 01
0 0'o ooIoooC)oo '° t ' N-a-a-a -a--a ' X . * ' oC exp enditures on other non-food items C
eD~~~~~~~~~~0 W 4 - C_, W W .N) 9 C ° ° -I4 -x ' '
-a-a _ t , _; v 1 *, ^ ' _ sTotal
- .0 0,0W010 0'pp:ao a0iO0 CD 0 _-- 00i 0 t9 .' I ! il- 0 cash food expenditures
00 0w 01 0o OD 0 0j 0O 0 0
g o .°lQ!°;Q C°P.° o ohome-produced foodcn .8h, arl I tn (M -bl OD 3
3~~~~~. 1. .| ':1J ') '~ ' 0_ic
- o oCoo o oio ol .Cj totalf tashfood oodexpendituresand o0ic) -M 4 ~-41 -4, -4 -4 -4J-. 41 aj
3 cno o C!>¢ N cn t' 0 home-produced food) 0- o- OiOl0trO oltI- --ot-I i - - -Di o 00 o}0I0000] o o! o! ol o iexpenditures on clothes - 0IS~~~~~~~!a 01a 0,000 :c 23<C X x- D+C 1 -4 i-i 4 1 -4 i i4 L,iii
j~~ooooooo~~~0 ooC) (DI~ 0 |ojoo 0ki 000 jt 0 expendituresonrentandutilities ol !i
0. x.~~~~~~~~~~~~~~~~~~~a,Ci)v Ni C;[ X ~ i ^1 C:) C:) C:): C:) ! t 0c I I . . i X
0, a' 0! O C> 0 00000 00 expenditures on services 0.o~~~~~~~D j to CO CO! !-,-4 4 0C.0,C
=D P,P _. p _ _ _ _; _ o; o O expenditures on other non-food items_ cn D .... O 0,_ o C.D 4 ) on oh i o 0
01
i Q . D D C, . . ao 6J .01 eJ in b 6 6sc1 bn '.? <n cash food expenditures ;jo L" C D: -t W. 4 _1 C CO L'0 ! -J J
0 0 0 ,o o total food (cash food expenditures and osr.o __ _ 9 home-produced food) I <
*0~~~~~~~~~~~~~~~~~~n X o o O ~~~~~~~~~~ex.pe-nd-i-tur-e-s on c-lothes <
o 0 4 .0'c
expenditures on rent and utilities c oe L~~~~CA)_ w cn _ x _ _ _,
0
o.. -1 ° 0 ° J °expenditures on services .-
0 1- - .O - -0- 0 -
a _j I C) p expenditures on other non-food items o.b . n ._ _ , _ _ .
Total :
t en; .v + cn t crl ~~~cash food expenditures !I
5KX 1Or O , t 1 shome-produced foodf- 0i1 C0 - t-
2 0 0 | O | o k o , total food (cash food expenditures and | IL"l CM 9 ) :-4 aw j home-produced food)
01 0 0 O p O , expenditures on clothes i W3 ~ 0 0) Co expenditures on
o co O CsC
m .--- _--- __ . - - ---- - - -
C 40 0 0) 0.I-..3 6 ~~~~~~~~~~~~~~~~~~expenditures on rent and utilities eV
_ _: _ o O 8 O expenditures on services
expenditures on other non-food items O |
_ i L1 Total... i . J J _L ,_ __ ._ ___
* V r i - - - - - r -- r- - - - . ----
*or-r . --r- L -I ~ I~ I ....- ! |Deciles of expenditures per GKSo g 0 .J co C?' 1 O en v CDequivalent adult3_ P0Po 'Po - ° 0 ,, cash food expenditures
0-401cm C." C." ur O ~ ~~~~~~~ 000 0- -½ ----(D > o .° l ° l ° ° l *° l *° jo tJ i home-produced food
-X -0 P&IP 0 0 0A homx 0) 6 Po I ' --4- IM _-'4 ---
ta o o 14I o o o~ | o o o o o ototal food (cash food expenditures and
Source: Author calculations from World Bank version of RLMS data.
Notes:(*) dF/dx is for discrete change of dummy variable from 0 to INumber of observations 2573
6
RuJs;: Incidence Analysis. 1994-1996
Ex-Ante Consumption (without receipt of a given transfer)(In percent of households poor by ex-ante consumption calculated as without the given tranfer)
1994 1995 1996Poor Not Poor Poor Not Poor Poor Not PoorReceived Not Received Not Received Not Received Not Received Not Received Not
Percentef anseholds Poer er Non-Poor by Ex-A ate Consumnptioe
Source: Author Calculations from the World Bank version of the RLMS dataset.
9
Russia: Income Under-reporting, 1994-1996
Table shows what percent of households who reported that their consumption was in agiven quintile also reported enough income to place in that same quintile.
1994 ConsumptionIncome First 20 % Second Third Fourth Last 20%First 20 % 44.87Second 26.65Third 27.5Fourth 30.64Last 20% 55.16
1995 ConsumptionIncome First 20 % Second Third Fourth Last 20%First 20 % 47.13Second 28.55Third 27.17Fourth 28.79Last 20% 52.27
1996 ConsumptionIncome First 20 % Second Third Fourth Last 20%First 20 % 44.52Second 28.47Third 28.51Fourth 30.29Last 20% 51.54
Memorandum Items: Average Household Consumption & Income
Average Average Income asHousehold Household Percent ofConsumption Income Consumption
b~~~~~~~~~~~ ~ .bo bg go . .b .bg .b bo . .O bo bg . .g .g. °og °ooo oOOgOg=
I~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ;0 .. 0 ...
a . . .o . . . . . . o o o o o o . . . . . . . . . . .
8 S S S S S S S S S S S S S S S S S S 8 S S S S S SS S @ 8 S S S 5 S S S S SS S S S S S S 4 Wo Wc Wo 8° S 9 ,oW oW oW CZ eW Wc C tW t0, qW Wa 0 qW, qW CO
Total number of Families 32,617 36,725 40,246Including:2 people 8,655 11,608 13,7593 people 9,116 11,589 11,2814 people 8,118 8,588 10,1545 people 6,728 4,940 5,052Average Size 4 3 3Nuclear Family 20,639 24,350 26,930Extended Family 5,128 4,692 4,614Complex Family 1,024 1,278 1,355Single Parent w/ children only 4,070 4,659 5,293Single Parent wl children+relatives 1,129 752 816Other 627 994 1,238Single person families 8,580 9,581 10,126Source: (Sotsial'noye polozheniye i uroveni zhizni naseleniya Rossiya 1997, p. 34.)
13
Russia: Pensioners by Type. 1990-1996(in Thousands End Year)
1990 1991 1992 1993 1994 1995 1996
All 32,848 34,044 35,273 36,100 36,623 37,083 37,827Old Age 25,659 27,131 28,390 29,021 29.095 29,011 29,081Disability 3,514 3,385 3,363 3,562 3,910 4,270 4,542Loss of Breadwinner 2,792 2,574 2,473 2,420 2,423 2,482 2,464Early Retirement 82 84 97 107 135 197 544Social I/ 470 870 956 990 1,060 1,123 1,196Source: Soisial'noye polozheniye i uroveni zhizni naseleniya Rossiya 1997, p. 197.
I/ Social Pensioners were fornerly called minimum pensioners. Social pensions are awarded to those who lack sufficientwork tenure for an old age pension.
Table Russia: Official Poverty Headcounts and Composition, 1994-1996(in percent of subtotal population by region) 996
Poor by typeHeadcounts Composition Extremely poor Merely poor
Poverty counts by household characteristics in Russia (number of panel households)
Round 5S Round 6* _ Round 7**lCounts Very Very Non- Very Very Non- Very Very Non-(house- P Poor2 poor + 3 Total poor2 poor + 3 Total l Poor poor + Totalholds) poor poor poor poor poor poor poor poor
Poverty counts by household characteristics in Russia (number of panel households)
* Round 5 of the RLMS survey was conducted in Russia in November 1994 - January 1995** Round 6 of the RLMS survey was conducted in Russia in October - November 1995
$** Round 7 of the RLMS survey was conducted in Russia in October- December 19961 Very poor - households with total expcnditures (see explanation in # 13) below 50% of the official regionally
differentiated (see explanation in 4 14) subsistence minimum adjusted for economies of scale in the household(Ministry of Labour of Russia)
2 Poor - households with total expenditures (see explanation in # 13) below official regionally differentiated (seeexplanation in # 14) subsistence minimum adjusted for economies of scale in thc houschold (Ministry of Labour ofRussia)
3 Non-poor - households with total expenditures (see explanation in # 13) above or equal to official regionallydifferentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale in the household(Ministry of Labour of Russia)
4 Children - those below 14 years of age5 Elderly - men above 59 and women above 54 years of age6 Household head - as defined by UNC7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region (oblast)8 Metropolies - Moscow and St. Petersburg9 Single mothers - a category chosen instead of single parent since there was only I case of single father in the sample
as of round 7 and no such cases as of rounds 5 and 610 Reporting unemployment - those who do not report any work, receive neither pension nor disability benefit and wouldII Disabled - those who receive disability benefit
12 Pensioners - those who receive old-age and/or early retirement pension13 Total expenditures - total household monetary food and non-food expenditures excluding big purchases, purchases of14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as population weighted average
21
Poverty rates by household characteristics in Russia (% of panel households)
Round 5* Round 6** Round 7***Rates (0/6ooff Verv Vhouseholds) VerYPor Very Non- Very 2 Non- Very Vr-y Non-
poor or poor + po 3 Total oa1 Poor 2poor -i po 3 Total po'Poor'2 poor + por Totalpoor poor poor poopoor poor poor poor-
Poverty rates by household characteristics in Russia (% of panel households)
Round 5 of the RLMS survey was conducted in Russia in November 1994 - January 1995** Round 6 of the RLMS survey was conducted in Russia in October - November 1995* Round 7 of the RLMS survey was conducted in Russia in October - December 1996
Very poor - households with total expenditures (see explanation in # 13) below 50% of the official regionallyI differentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale in the household
(Ministry of Labour of Russia)
Poor - households with total expenditures (see explanation in # 13) below official regionally differentiated (see2 explanation in # 14) subsistence minimum adjusted for economies of scale in the houschold (Ministry of Labour of
Russia)Non-poor - households with total expenditures (see explanation in # I 3) above or equal to official regionally
3 differentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale in the household(Ministry of Labour of Russia)
4 Children - those below 14 years of age
5 Elderly - men above 59 and women above 54 years of age6 Household head - as defined by UNC7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region (oblast)
8 Metropolies - Moscow and St. Petersburg9Single mothers - a category choscn instead of single parent since there was only I case of single father in the sample
as of round 7 and no such cases as of rounds 5 and 610 Reporting unemployment - those who do not report any work, receive neither pension nor disability benefit and
11 Disabled - those who receive disability benefit
12 Pensioners - those who reccive old-age and/or early retirement pension
13 Total expenditures - total household monetary food and non-food expenditures excluding big purchases, purchases of14 Regionally differentiated subsistence miniitim - 8 regional poverty lines computed as population weighted average
25
Poverty composition by household chAracteristics in Russia (% of panel households)
Round 5* Round 6** Round 7***
Compousitionldso Verypo |Poor2 Non-poor3 Total Very poor' Poor2 Non-poor3 Total Very poor' Poor2 Non-poor' Total
Poverty composition by household characteristics in Russia (% of panel households)
* Round 5 of the RLMS survey was conducted in Russia in Novembcr 1994 - January 1995** Round 6 of the RLMS survey was conducted in Russia in October - November 1995
-* Round 7 of the RLMS survey was conducted in Russia in October - December 1996I Very poor - houscholds with total expenditures (see explanation in # 13) below 50% of the official regionally differentiated (see explanation
in # 14) subsistence minimum adjusted for economies of scale in thc household (Ministry of Labour of Russia)
2 Poor - households with total expenditures (see explanation in # 13) below official regionally-differentiated (see explanation in 4 14)
subsistence minimum adjusted for economies of scale in the household (Ministry of Labour of Russia)
3 Non-poor - households with total expenditures (see explanation in 4 13) above or equal to official regionally differentiated (see explanation
in # 14) subsistence minimum adjusted for economies of scale in the household (Ministry of Labour of Russia)
4 Children - those below 14 years of age5 Elderly - men above 59 and women above 54 years of age6 Household head - as defined by UNC
7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region (oblast)
8 Metropolies - Moscow and St. Petersburg9 Single mothers - a category chosen instead of singlc parent since there was only 1 case of single father in the sample as of round 7 and no
such cases as of rounds 5 and 610 Reporting unemployment - those who do not report any work, receive neither pension nor disability benefit and would like to work
II Disabled - those who receive disability benefit
12 Pcnsioners - those who receive old-age and/or early retirement pension
13 Total expenditures - total household monetary food and non-food expenditurcs cxcluding big purchases, purchases of luxury goods,bonds/stocks and savings plus value of home-produced food evaluated at prevailing market prices
14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as population weighted average across 78 officialregional subsistence minima so that to match survey sampie division of Russia into 8 regions
Poverty counts by household characteristics in Russia (number of computed panel individuals")
Round 5* Round 6** Round 7***CountsVeyer
(computed Very Very Non- Very Very Non- Very Very Non-indi- poor' Poor poor + Total p Poor poor + poor3 Total p Poor' poor + Total
Poverty counts by household characteristics in Russia (number of computed panel indiv1duabl )
* Round 5 of the RLMS survey was conducted in Russia in November 1994 -January 1995Round 6 of the RLMS survey was conducted in Russia in October - November 1995Round 7 of the RLMS survey was conducted in Russia in October - December 1996Computed individuals - computed across households weighted by household size
1 Very poor - households with total expenditures (see explanation in # 13) below 50% of the officialregionally differentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale inthe household (Ministry of Labour of Russia)
2 Poor - households with total expenditures (see explanation in # 13) below official regionally differentiated(see explanation in # 14) subsistence minimum adjusted for economies of scale in the household (Ministryof Labour of Russia)
3 Non-poor - households with total expenditures (see explanation in # 13) above or equal to official regionallydifferentiated (see explanation in #'14) subsistence minimum adjusted for economies of scale in thehousehold (Ministry of Labour of Russia)
4 Children - those below 14 years of age
5 Elderly - men above 59 and women above 54 years of age6 Household head - as defined by UNC
7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region (oblast)8 Metropolies - Moscow and St. Petersburg
9 Single mothers - a category chosen instead of single parent since there was only 1 case of single father inthe sample as of round 7 and no such cases as of rounds 5 and 6
10 Reporting unemployment - those who do not report any work, receive neither pension nor disability benefitand would like to work
11 Disabled - those who receive disability benefit12 Pensioners - those who receive old-age and/or early retirement pension
13 Total expenditures - total household monetary food and non-food expenditures excluding big purchases,purchases of luxury goods, bonds/stocks and savings plus value of home-produced food evaluated atprevailing market prices
14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as population weightedaverage across 78 official regional subsistence minima so that to match survey sample division of Russiainto 8 regions
33
Poverty rates by household characteristics in Russia (% of computed panel individuals*")
Round 5' Round 6- Round 7**_Rates (%of VrVeyVercomputed Very Very Non- Vs Very Non- Very ery Non-
indi- poor' Poor poor+ poor3 Total poor' oor + poor tal poor' Poor' poor + poor Totalviduals'***) poor poor poor
Poverty rates by household characteristics in Russia (% of computed panel individuals^)
Round 5 of the RLMS survey was conducted in Russia in November 1994 - January 1995Round 6 of the RLMS survey was conducted in Russia in October - November 1995Round 7 of the RLMS survey was conducted in Russia in October - December 1996Computed individuals - computed across households weighted by household size
1 Very poor - households with total expenditures (see explanation in # 13) below 50% of the officialregionally differentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale inthe household (Ministry of Labour of Russia)
2 Poor - households with total expenditures (see explanation in # 13) below official regionally differentiated(see explanation in # 14) subsistence minimum adjusted for economies of scale in the household (Ministryof Labour of Russia)
3 Non-poor - households with total expenditures (see explanation in # 13) above or equal to official regionallydifferentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale in thehousehold (Ministry of Labour of Russia)
4 Children - those below 14 years of age5 Elderly - men above 59 and women above 54 years of age6 Household head - as defined by UNC7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region (oblast)8 Metropolies - Moscow and St. Petersburg9 Single mothers - a category chosen instead of single parent since there was only 1 case of single father in
the sample as of round 7 and no such cases as of rounds 5 and 6
10 Reporting unemployment - those who do not report any work, receive neither pension nor disability benefitand would like to work
11 Disabled - those who receive disability benefit
12 Pensioners - those who receive old-age and/or early retirement pension13 Total expenditures - total household monetary food and non-food expenditures excluding big purchases,
purchases of luxury goods, bonds/stocks and savings plus value of home-produced food evaluated atprevailing market prices
14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as population weightedaverage across 78 official regional subsistence minima so that to match survey sample division of Russiainto 8 regions
37
Poverty composition by household characteristics in Russia (% of computed panel individuals****)
Round 5* Round 6** Round 7-Composition (%of
computed indi- Very poor' Poor2 Non-poor3 Total Very poor' Poor2 Non-poor3 Total Very poor' Poor2 Non-poor3 Totalviduals***)
Poverty transition (counts) by household characteristics in Russia (number of panel households)
* Household characteristics used are as of round 5 (see explanation in #1)1 Poverty transition - poverty state of the household in three subsequent rounds of the RLMS survey
(5, 6 and 7) which were conducted as follows:Round 5: November 1994 - January 1995Round 6: October - November 1995Round 7: October - December 1996
where2 np (non-poor) - households with total expenditures (see explanation in # 13) above or equal to
official regionally differentiated (see explanation in # 14) subsistence minimum adjusted foreconomies of scale in the household (Ministry of Labour of Russia)
3 p (poor) - households with total expenditures (see explanation in # 13) below official regionallydifferentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale inthe household (Ministry of Labour of Russia)
Example: np-p-np means that the household was non-poor (see explanation in #2 ) in round 5 (seeexplanation in #1), poor (see explanation in #3 ) in round 6 (see explanation in #1) and non-poor(see explanation in #2 ) in round 7 (see explanation in #1)
4 Children - those below 14 years of age5 Elderly - men above 59 and women above 54 years of age6 Household head - as defined by UNC7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region
(oblast)8 Metropolies - Moscow and St. Petersburg9 Single mothers - a category chosen instead of single parent since there was only 1 case of single
father in the sample as of round 7 and no such cases as of rounds 5 and 610 Reporting unemployment - those who do not report any work, receive neither pension nor disability
benefit and would like to work11 Disabled - those who receive disability benefit12 Pensioners - those who receive old-age and/or early retirement pension
13 Total expenditures - total household monetary food and non-food expenditures excluding bigpurchases, purchases of luxury goods, bonds/stocks and savings plus value of home-producedfood evaluated at prevailing market prices
14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as populationweighted average across 78 official regional subsistence minima so that to match survey sampledivision of Russia into 8 regions
45
Poverty transition (rates) by household characteristics* in Russia (% of panel households)
2 and more pensio-nd rsio2 6.4 4.8 7.8 3.2 6.4 11.4 55.3 4.8 100.0ners'
Total 7.0 13.0 7.8 5.7 8.1 10.3 42.4 5.7 100.0
48
Poverty transition (rates) by household characteristics* in Russia (% of panel households)
* Household characteristics used are as of round 5 (see explanation in #1)1 Poverty transition - poverty state of the household in three subsequent rounds of the RLMS
survey (5, 6 and 7) which were conducted as follows:Round 5: November 1994 - January 1995Round 6. October - November 1995Round 7: October - December 1996
where2 np (non-poor) - households with total expenditures (see explanation in # 13) above or equal.
to official regionally differentiated (see explanation in # 14) subsistence minimum adjustedfor economies of scale in the household (Ministry of Labour of Russia)
3 p (poor) - households with total expenditures (see explanation in # 13) below officialregionally differentiated (see explanation in # 14) subsistence minimum adjusted foreconomies of scale in the household (Ministry of Labour of Russia)
Example: np-p-np means that the household was non-poor (see explanation in #2 ) in round 5 (seeexplanation in #1), poor (see explanation in #3 ) in round 6 (see explanation in #1) and non-poor (see explanation in #2 ) in round 7 (see explanation in #1)
4 Children - those below 14 years of age5 Elderly - men above 59 and women above 54 years of age6 Household head - as defined by UNC7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region
(oblast)8 Metropolies - Moscow and St. Petersburg9 Single mothers - a category chosen instead of single parent since there was only 1 case of
single father in the sample as of round 7 and no such cases as of rounds 5 and 610 Reporting unemployment - those who do not report any work, receive neither pension nor
disability benefit and would like to work1 1 Disabled - those who receive disability benefit12 Pensioners - those who receive old-age and/or early retirement pension
13 Total expenditures - total household monetary food and non-food expenditures excludingbig purchases, purchases of luxury goods, bonds/stocks and savings plus value of home-produced food evaluated at prevailing market prices
14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed aspopulation weighted average across 78 official regional subsistence minima so that to matchsurvey sample division of Russia into 8 regions
49
Poverty transition (composition) by household characteristics* in Russia (% of panel households)
Poverty transition'Composition (% of - I . Ihouseholds)nze p-p-p' p-np-np1 p-p-np1 np-p-p1 np-np-p np-np-np1 p-np-p1 I Total
Poverty transition (composition) by household characteristics* In Russia (% of panel households)
* Household characterstics used are as of round 5 (see explanation in #11)1 Poverty transition - poverty state of the household in three subsequent rounds of the RLMS survey (5,
6 and 7) which were conducted as follows:Round 5: November 1994 - January 1995Round 6: October - November 1995Round 7: October - December 1996
where2 np (non-poor) - households with total expenditures (see explanation in # 13) above or equal to official
regionally differentiated (see explanation in # 14) subsistence minimum adjusted for economies ofscale in the household (Ministry of Labour of Russia)
3 p (poor) - households with total expenditures (see explanation in # 13) below official regionallydifferenbated (see explanation in # 14) subsistence minimum adjusted for economies of scale in thehousehold (Ministry of Labour of Russia)
Example: np-p-np means that the household was non-poor (see explanation in #2 ) in round 5 (see explanationin #1), poor (see explanation in #3 ) in round 6 (see explanation in #1) and non-poor (see explanationin #2 ) in round 7 (see explanation in #1)
4 Children - those below 14 years of ageLi 5 Elderly - men above 59 and women above 54 years of age
6 Household head - as defined by UNC7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region (oblast)8 Metropolies - Moscow and St. Petersburg9 Single mothers - a category chosen instead of single parent since there was only 1 case of single
father in the sample as of round 7 and no such cases as of rounds 5 and 610 Reporting unemployment - those who do not report any work, receive neither pension nor disability
benefit and would like to work11 Disabled - those who receive disability benefit12 Pensioners - those who receive old-age and/or early retirement pension13 Total expenditures - total household monetary food and non-food expenditures excluding big
purchases, purchases of luxury goods, bonds/stocks and savings plus value of home-produced foodevaluated at prevailing market prices
14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as populationweighted average across 78 official regional subsistence minima so that to match survey sampledivision of Russia into 8 regions
Poverty transition (counts) by household characteristics* in Russia (number of computed panelindividuals"*)
Poverty transition'Counts
(ncvduatse np-p-np p-p-p p-np-np p-p-np1 np-p-p' np-np-p' np-np-np' p-np-p1 Total
Poverty transition (counts) by household characteristics' in Russia (number of computed panelindividuals")
Household characteristics used are as of round 5 (see explanation in #1)Computed individuals - computed across households weighted by household size
1 Poverty transition - poverty state of the household in three subsequent rounds of the RLMSsurvey (5, 6 and 7) which were conducted as follows:Round 5: November 1994 - January 1995Round 6: October - November 1995Round 7: October- December 1996
where2 np (non-poor) - households with total expenditures (see explanation in # 13) above or equal to
official regionally differentiated (see explanation in # 14) subsistence minimum adjusted foreconomies of scale in the household (Ministry of Labour of Russia)
3 p (poor) - households with total expenditures (see explanation in # 13) below official regionallydifferentiated (see explanation in # 14) subsistence minimum adjusted for economies of scalein the household (Ministry of Labour of Russia)
Example: np-p-np means that the household was non-poor (see explanation in #2 ) in round 5 (seeexplanation in #1), poor (see explanation in #3 ) in round 6 (see explanation in #1) and non-poor (see explanation in #2 ) in round 7 (see explanation in #1)
4 Children - those below 14 years of age5 Elderly - men above 59 and women above 54 years of age6 Household head - as defined by UNC7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region
(oblast). Metropolies - Moscow and St. Petersburg9 Single mothers - a category chosen instead of single parent since there was only 1 case of
single father in the sample as of round 7 and no such cases as of rounds 5 and 610 Reporting unemployment - those who do not report any work, receive neither pension nor
disability benefit and would like to work11 Disabled - those who receive disability benefit12 Pensioners - those who receive old-age and/or early retirement pension13 Total expenditures - total household monetary food and non-food expenditures excluding big
purchases, purchases of luxury goods, bonds/stocks and savings plus value of home-produced food evaluated at prevailing market prices
14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed aspopulation weighted average across 78 official regional subsistence minima so that to matchsurvey sample division of Russia into 8 regions
57
Poverty transition (rates) by household characteristics* in Russia (% of computed panel individuals')
Poverty transition'Rates (% of computed
individuals*) np-p-np' p-p-pl p-np-npl p-p-np' np-p-p' np-np-p' np-np-np' p-np-p1 Total
Poverty transition (rates) by household characteristics^ in Russia (% of computed panel individuals*)
Household characteristics used are as of round 5 (see explanation in #1)Computed individuals - computed across households weighted by household size
1 Poverty transition - poverty state of the household in three subsequent rounds of the RLMSsurvey (5, 6 and 7) which were conducted as follows:Round 5 November 1994 - January 1995Round 6: October - November 1995Round 7: October - December 1996
where2 np (non-poor) - households with total expenditures (see explanation in # 13) above or equal
to official regionally differentiated (see explanation in # 14) subsistence minimum adjustedfor economies of scale in the household (Ministry of Labour of Russia)
3 p (poor) - households with total expenditures (see explanation in # 13) below ofFicialregionally differentiated (see explanation in # 14) subsistence minimum adjusted foreconomies of scale in the household (Ministry of Labour of Russia)
Example np-p-np means that the household was non-poor (see explanation in #2 ) in round 5 (seeexplanation in #1), poor (see explanation in #3 ) in round 6 (see explanation in #1) and non-poor (see explanation in #2 ) in round 7 (see explanation in #1)
4 Children - those below 14 years of age5 Elderly - men above 59 and women above 54 years of age6 Household head - as defined by UNC7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski
region (oblast)8 Metropolies - Moscow and St. Petersburg9 Single mothers - a category chosen instead of single parent since there was only I case of
single father in the sample as of round 7 and no such cases as of rounds 5 and 610 Reporting unemployment - those who do not report any work, receive neither pension nor
disability benefit and would like to work1 1 Disabled - those who receive disability benefit12 Pensioners - those who receive old-age and/or early retirement pension13 Total expenditures - total household monetary food and non-food expenditures excluding
big purchases, purchases of luxury goods, bonds/stocks and savings plus value of home-produced food evaluated at prevailing market prices
14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed aspopulation weighted average across 78 official regional subsistence minima so that tomatch survey sample division of Russia into 8 regions
61
Poverty transition (composition) by household characteristics* in Russia (% of computed panel individuals*)
Poverty transition'Compustediondivduls of IComposition (% of dus np-p-np | ppp1 I p-np-np1 p-p-np' np-p-p' np-np-p' np-np-np1 p-np-p1 I Total
2 and more pensio-2 and er 1eo- 14.3 6.1 17.7 9.3 12.4 17.6 22.0 17.1 16.0ners'lTotal 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0
Poverty transition (composition) by household characteristics* in Russia (% of computed panel individuals**)
* Household characteristics used are as of round 5 (see explanation in #1)
Computed individuals - computed across households weighted by household size
1 Poverty transition - poverty state of the household in three subsequent rounds of the RLMS survey (5,
6 and 7) which were conducted as follows:
Round 5: November 1994 - January 1995Round 6: October - November 1995Round 7: October - December 1996
where2 np (non-poor) - households with total expenditures (see explanation in # 13) above or equal to official
regionally differentiated (see explanation in # 14) subsistence minimum adjusted for economies of
scale in the household (Ministry of Labour of Russia)
3 p (poor) - households with total expenditures (see explanation in # 13) below official regionally
differentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale in the
household (Ministry of Labour of Russia)Example: np-p-np means that the household was non-poor (see explanation in #2 ) in round 5 (see explanation
in #1), poor (see explanation in #3 ) in round 6 (see explanation in #1) and non-poor (see explanation
in #2 ) in round 7 (see explanation in #1)4 Children - those below 14 years of age5 Elderly - men above 59 and women above 54 years of age
6 Household head - as defined by UNC7 Metropolitan - Moscow and Moscow region (oblast), St. Petersburg and Leningradski region (oblast)
8 Metropolies - Moscow and St. Petersburg9 Single mothers - a category chosen instead of single parent since there was only 1 case of single
father in the sample as of round 7 and no such cases as of rounds 5 and 6
10 Reporting unemployment - those who do not report any work, receive neither pension nor disability
benefit and would like to work11 Disabled - those who receive disability benefit12 Pensioners - those who receive old-age and/or early retirement pension
13 Total expenditures - total household monetary food and non-food expenditures excluding big
purchases, purchases of luxury goods, bonds/stocks and savings plus value of home-produced food
evaluated at prevailing market prices
14 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as population
weighted average across 78 official regional subsistence minima so that to match survey sample
division of Russia into 8 regions
Poverty counts by individual characteristics in Russia (number of panel individuals)
Round 5^ Round 6** Round 7***Counts (individuals) Very Very Non- Very Very Non- Very Very Non-
poor' Poor2 poor+ poor3 Total poor' Poor' poor poor3 Total poor, Poor2 poor+ poo Total
* Round 5 of the RLMS survey was conducted in Russia in November 1994 - January 1995Round 6 of the RLMS survey was conducted in Russia in October - November 1995Round 7 of the RLMS survey was conducted in Russia in October - December 1996
1 Very poor - households with total expenditures (see explanation in # 7) below 50% of the official regionally differentiated (seea'> explanation in # 8) subsistence minimum adjusted for economies of scale in the household (Ministry of Labour of Russia)
2 Poor - households with total expenditures (see explanation in # 7) below official regionally differentiated (see explanation in # 8)subsistence minimum adjusted for economies of scale in the household (Ministry of Labour of Russia)
3 Non-poor - households with total expenditures (see explanation in # 7) above or equal to official regionally differentiated (seeexplanation in # 8) subsistence minimum adjusted for economies of scale in the household (Ministry of Labour of Russia)
4 Reporting unemployment - those who do not report any work, receive neither pension nor disability benefit and would like to work5 Children - those below 14 years of age6 Elderly - men above 59 and women above 54 years of age7 Total expenditures - total household monetary food and non-food expenditures excluding big purchases, purchases of
luxury goods, bonds/stocks and savings plus value of home-produced food evaluated at prevailing market prices8 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as population weighted average
across 78 official regional subsistence minima so that to match survey sample division of Russia into 8 regions
Poverty rates by individual characteristics in Russia (% of panel individuals)
Round 5- Round 6** Round 7***Rates (individuals) Very Very Non- Very Very Non- Very Very Non-
p Poor2 poor + poor3 Total poor' poor2 poor + pOoe Total poor' Poor2 poor + poor] Totalpo porpoporpo 3poor poor poor
* Round 5 of the RLMS survey was conducted in Russia in November 1994 - January 1995Round 6 of the RLMS survey was conducted in Russia in October - November 1995
*^* Round 7 of the RLMS survey was conducted in Russia in October - December 19961 Very poor - households with total expenditures (see explanation in # 7) below 50% of the official regionally
differentiated (see explanation in # 8) subsistence minimum adjusted for economies of scale in the household(Ministry of Labour of Russia)
2 Poor - households with total expenditures (see explanation in # 7) below official regionally differentiated (seeexplanation in # 8) subsistence minimum adjusted for economies of scale in the household (Ministry of Labour ofRussia)
3 Non-poor - households with total expenditures (see explanation in # 7) above or equal to official regionallydifferentiated (see explanation in # 8) subsistence minimum adjusted for economies of scale in the household(Ministry of Labour of Russia)
4 Reporting unemployment - those who do not report any work, receive neither pension nor disability benefit andwould like to work
5 Children - those below 14 years of age6 Elderly - men above 59 and women above 54 years of age7 Total expenditures - total household monetary food and non-food expenditures excluding big purchases, purchases of
luxury goods, bonds/stocks and savings plus value of home-produced food evaluated at prevailing market pricesB Regionally differentiated subsistence minimum - 8 regional poverty lines computed as population weighted average
across 78 official regional subsistence minima so that to match survey sample division of Russia into 8 regions
Poverty composition by individual characteristics in Russia (% of panel individuals)
Round 5* Round 6- Round 7***Composition (individuals) Very Non- Very Non- Very Non-
poor1 Poor' poor 3 Total p oor' 2po poor 3 Total poor' Poor2 poor' Total
Round 5 of the RLMS survey was conducted in Russia in November 1994 - January 1995Round 6 of the RLMS survey was conducted in Russia in October - November 1995Round 7 of the RLMS survey was conducted in Russia in October - December 1996
1 Very poor - households with total expenditures (see explanation in # 7) below 50% of the official regionallydifferentiated (see explanation in # 8) subsistence minimum adjusted for economies of scale in the household(Ministry of Labour of Russia)
2 Poor - households with total expenditures (see explanation in # 7) below official regionally differentiated (seeexplanation in # 8) subsistence minimum adjusted for economies of scale in the household (Ministry of Labour ofRussia)
3 Non-poor - households with total expenditures (see explanation in # 7) above or equal to official regionallydifferentiated (see explanation in # 8) subsistence minimum adjusted for economies of scale in the household(Ministry of Labour of Russia)
4 Reporting unemployment - those who do not report any work, receive neither pension nor disability benefit andwould like to work
5 Children - those below 14 years of age6 Elderly - men above 59 and women above 54 years of age7 Total expenditures - total household monetary food and non-food expenditures excluding big purchases,
purchases of luxury goods, bonds/stocks and savings plus value of home-produced food evaluated at prevailingmarket prices
8 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as population weightedaverage across 78 official regional subsistence minima so that to match survey sample division of Russia into 8regions
Poverty transition (counts) by individual characteristics* in Russia (number of panel individuals)
Poverty transition'Counts (individuals) f np-p-np' P-P-P p-np-np' p-p-np' np-p-p' np-np-p' np-np-np' p-np-p' Total
Household characteristics used are as of round 5 (see explanation in #1)1 Poverty transition - poverty state of the household in three subsequent rounds of the RLMS survey
(5, 6 and 7) which were conducted as follows:Round 5: November 1994 - January 1995Round 6: October - November 1995Round 7: October- December 1996
where2 np (non-poor) - households with total expenditures (see explanation in # 13) above or equal to
official regionally differentiated (see explanation in # 14) subsistence minimum adjusted foreconomies of scale in the household (Ministry of Labour of Russia)
3 p (poor) - households with total expenditures (see explanation in # 13) below official regionallydifferentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale inthe household (Ministry of Labour of Russia)
Example: np-p-np means that the household was non-poor (see explanation in #2 ) in round 5 (seeexplanation in #1), poor (see explanation in #3 ) in round 6 (see explanation in #1) and non-poor(see explanation in #2 ) in round 7 (see explanation in #1)
4 Reporting unemployment - those who do not report any work, receive neither pension nor disabilitybenefit and would like to work
5 Children - those below 14 years of age6 Elderly - men above 59 and women above 54 years of age7 Total expenditures - total household monetary food and non-food expenditures excluding big
purchases, purchases of luxury goods, bonds/stocks and savings plus value of home-producedfood evaluated at prevailing market prices
8 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as populationweighted average across 78 official regional subsistence minima so that to match survey sampledivision of Russia into 8 regions
80
Poverty transition (rates) by individual characteristics' in Russia (% of panel individuals)
Poverty transition'
Rates (individuals) np-p-np1 p- p 1 p-np-np' p-p-np1 np-p-p' np-np-p' n-np-np' p-np-pI Total
Household characteristics used are as of round 5 (see explanation in #1)1Poverty transition - poverty state of the household in three subsequent rounds of the RLMS survey
(5, 6 and 7) which were conducted as follows:Round 5. November 1994 - January 1995Round 6: October - November 1995Round 7 October - December 1996
where2 np (non-poor) - households with total expenditures (see explanation in # 13) above or equal to
official regionally differentiated (see explanation in # 14) subsistence minimum adjusted foreconomies of scale in the household (Ministry of Labour of Russia)
3 p (poor) - households with total expenditures (see explanation in # 13) below official regionallydifferentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale in thehousehold (Ministry of Labour of Russia)
Example np-p-np means that the household was non-poor (see explanation in #2 ) in round 5 (seeexplanation in #1), poor (see explanation in #3 ) in round 6 (see explanation in #1) and non-poor(see explanation in #2 ) in round 7 (see explanation in #1)
4 Reporting unemployment - those who do not report any work, receive neither pension nor disabilitybenefit and would like to work
5 Children - those below 14 years of age6 Elderly - men above 59 and women above 54 years of age7 Total expenditures - total household monetary food and non-food expenditures excluding big
purchases, purchases of luxury goods, bonds/stocks and savings plus value of home-producedfood evaluated at prevailing market prices
8 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as populationweighted average across 78 official regional subsistence minima so that to match survey sampledivision of Russia into 8 regions
83
Poverty transition (composition) by individual characteristics in Russia (% of panel individuals)
Total 100.0 100.0 1000 100.0 1000 100.0 100.0 1000 1000
Household characteristics used are as of round 5 (see explanation in #1)1 Poverty transition - poverty state of the household in three subsequent rounds of the RLMS survey
(5, 6 and 7) which were conducted as follows.
Round 5: November 1994 - January 1995Round 6: October - November 1995Round 7: October- December 1996
where2 np (non-poor) - households with total expenditures (see exp anation in # 13) above or equal to
official regionally differentiated (see explanation in # 14) subsistence minimum adjusted foreconomies of scale in tne household (Ministry of Labour of Russia)
3 p (poor) - households with total expenditures (see explanation in # 13) below official regionallydifferentiated (see explanation in # 14) subsistence minimum adjusted for economies of scale in thehousehold (Ministry of Labour of Russia)
Example. np-p-np means that the household was non-poor (see explanation in #2 ) in round 5 (seeexplanation in #1) poor (see explanation in #3 ) in round 6 (see explanation in #1) and non-poor(see explanation in #2 ) in round 7 (see explanation in #1)
4 Reporting unemployment - those who do not report any work, receive neither pension nor disabilitybenefit and would like to work
5 Children - those below 14 years of age6 Elderly - men above 59 and women above 54 years of age7 Total expenditures - total household monetary food and non-food expenditures excluding big
purchases, purchases of luxury goods, bonds/stocks and savings plus value of home-produced foodevaluated at prevailing market prices
8 Regionally differentiated subsistence minimum - 8 regional poverty lines computed as populationweighted average across 78 official regional subsistence minima so that to match survey sampledivision of Russia into 8 regions
86
ANNEX FOUR
PROXY MEANS TESTS FOR RUSSIA 1994-98
JEANINE BRAITHWAITE (ECSPE)
ANNA IVANOVA (U.WISC.)
1. A proxy means test is a method to estimate household consumption or welfarewithout requiring extremely detailed information about household income. In countriessuch as Russia or Chile where there is a large informal sector, it can be very difficult andadministratively very costly to verify true household money income. Furthermore, inRussia and other countries, a very significant part of household food consumption comesfrom food grown on private garden plots. It can be very difficult to estimate the true value(impute the value correctly) of home-produced goods, since typically, they are producedwith "costless" family labor and their quality may be different than for example fooditems which are produced for sale.
2. In proxy means tests, rather than trying to measure total income perfectly,information is collected on items which are much easier to measure and verify, such as thenumber of children in the family, etc. These variables should be ones which are known tocorrelate with poverty in the country, and ideally, which are easy to measure and thusrequire little administrative cost to verify. The first large-scale use of proxy means-testingoccurred in Chile in the late 1970s and 1980s, in a program called the Ficha CAS (card forsocial assistance). Since 1994, Costa Rica and Columbia have adopted proxy means-testsfor some of their social assistance programs, Mexico is about to start a proxy means-testprogram, and Argentina and Venezuela are actively considering the idea.
3. Within the region, independently of the Latin American experience, Armenia hasadopted a proxy means test in its Paros program for the distribution of humanitarianassistance. However, the scoring formula used in the Paros program was not based on anysort of econometric estimates and may contain some inaccuracies. Rather than use thescoring formula of the Paros program for Russia, it would be much better to estimate anew scoring formula. Indeed, in each country usinlg a proxy means test, a unique scoringformula is estimated, based on data on household expenditures and characteristics.
4. This paper presents several different proxy means regressions for Russia or oblastsof Russia. Results differed for many reasons, the most important of which were the highlydifferent sources of data used and the different time periods to which the regressionspertained. Furthermore, the results are presented in chronological order, from the veryfirst proxy means test results for data from early 1994 of the RLMS (these data are notuised elsewhere in this report), through social assistance pilot data for 1997-98, and finally,for the three rounds of the World Bank version of the RLMS data set for 1994-96.
I
1. SIMULATION FOR RUSSIA AND VOLGOGRAD
5. A data base is required for estimating an appropriate scoring formula for Russia.For this first illustrative example, household data for Russia for October 1993-February1994 (Round IV) from the Russian Longitudinal Monitoring Survey (RLMS) are used.
6. The first step in any study of household behavior is to decide on a best measure ofhousehold consumption, which is usually considered to be household expenditures plusthe market value of any goods produced and consumed by the household.' Here we facethe first technical limitation of Round IV of the RLMS. Unlike Rounds l-II, in Round IVhouseholds were asked only one question about the value of goods (mostly food) producedand consumed at home--they were asked to estimate its worth. In other householdsurveys, such as one done in Ukraine in 1995 and 1996, households were asked to providedetails about their food production & consumption in physical units by each type of food.and then survey researchers imputed the value of the food (by using average purchaseprices). This approach will also be used in the survey of the social assistance pilots, sinceit produces much more reliable estimates of household production and consumption offood. Preliminary work with the Round IV data of the RLMS suggest that householdsmay have under-estimated the value of the food they produced on their own private plots,perhaps in many cases because the household simply did not know the correct marketprice for its food.
7. For this illustrative example, however, it was not possible to adjust or correct theRound IV results, so they will be used with caution. Thus the measure of householdwelfare used here is total household expenditures, which include the household's ownestimate of the value of food and other goods it produced and consumed and the value ofall household reported purchases of food, non-food goods. services, and miscellaneousother purchases. Total household expenditures are then divided by the number ofhousehold members to generate per capita expenditures.2 Thus below per capita isunderstood to be "per household member."
8. The next step was to try to measure the correlation between household welfare (percapita expenditures) and other easily-measured variables. The technique used wasstepwise least squares regression, in which variables are excluded if they are notsignificantly correlated with per capita expenditures. Technically speaking, this techniqueshould only be used for variables which are thought to be independent (exogenous) andnot directly correlated with each other. I-lowever, since we are only trying to find proxies
Although many statistical publications rely on houschold income as measure of household welfare, there aremany studies which demonstrate that if household cxpenditurc data are available, they should be used inpreference to data on household income, whichl tenlds to bc under-reported, especially on the high end of theincome distribution. Household income tends to appear frequently in statistical publications of manycountries because typically, monthly estimates of household income are made by central statistical agencies,while household expenditures data are more commonily available on an annual (or sometimes quarterly)basis.
2 There is an extremely lively debate in the economics literature about whether per capita measurements are asdesirable as "equivalent adult" measurements which reflect economies of scale in consumption (decliningmarginal cost of additional family members), but for the purposes of the pilot, we will set aside this debate.
2
(substitutes) for poverty which are more easily measured than household expenditures(rather than trying to decide what determines poverty) we can ignore this caveat in thiscontext. Especially, we will ignore the caveat because we will regress per capitaexpenditures on a variety of variables, including official money income (wages plustransfer payments such as pensions, allowances, and stipends).3
9. During the first round of regressions, some surprising and possibly doubtful resultsoccurred. For example, the presence of a female household head4 seemed to correlate withhigher levels of per capita expenditures than those of a male household head, while thepresence of a private plot seemed to reduce the level of per capita expenditures. The firstresult may have occurred because there are few female household heads and they may beable to estimate more accurately their expenditures. The second result undoubtedlyreflects the technical limitation about self-production from the questionnaire noted above.Since it is possible to spend a very long time on econometric work without resolvingissues of this sort, it was decided to simply drop variables that seemed to have aproblematic relationship to per capita expenditures, even if the variable seemed to besignificantly correlated with per capita expenditures.
10. The final specification is one in which all the coefficients estimated are significantat the 10 percent level, and all but one are extremely significant. We estimated
C = a + bY + cX
where C is per capita expenditures (exp_p in the table below)
Y is per capita official income (jbmi_p in the table below, consisting of the sum ofhousehold wage income and transfer income, including pensions, child allowances,stipends, local social assistance, unemploymenlt benefits, and other cash and in-kindtransfers)'
Xi is a set of other easily measured variables, including the number of children inthe family (CHILDN), the number of elderly (ELDERLYN). whether the household islocated in a city other than Moscow (DLOC I), whether the household is a rural household(DLOC2),6 whether the household has a refrigerator (RREFIGDA), whether the
In this sense, we are estimating a houiselhold consumption function, where C = a + bY + cX, where C isconsumption (as measured by per capita expenditures), Y is per capita money income, and X, is a vector ofindependent variables independent from income. However. since in practice most of our easily measured X,variables are not likely to be completely independent from income, we would be violating one of therequirements of the regression model if we were seeking to determine causality.
4A female headed household was defined as one in which there was an active-aged female (over 15 and under 55)but no active-aged male or a household in which there were no active-aged adults, an elderly female but noelderly male.
5For in-kind transfers, this represents the estimate of the person answering the questionnaire on the value of thein-kind transfer. For wages, respondents were asked to estimate the monetary value of any wages paid in-kind.
6 In this definition, villages of the urban type are classified as rural.
3
household has a car (RCARDA), and the number of unemployed household members(DLFS1)
It. Although there are other variables that could have been included in the set of Xisuch as age of the household head, in the stepwise regression some variables were foundeither to not be significantly correlated with per capita expenditures (age of householdhead, whether the person rents or own the apartment, age of household head squaredwhich approximates a life cycle effect, education of household head) or were found to beperversely correlated' and so were omitted. The estimated coefficients, standard errors, tstatistics, and significance for this specification are shown in the following table.
Table D-1. Stepwise Regression Results for Russia, Oct. 1993-Feb. 1994
One way to interpret this table is to use its information to construct a scoring formula for estimating percapita household expenditures.
PCEest = 50618 + 0.63*(per cap Official Income) - 6660*(Numnher c?f Children) -5539*(Numher of peopleover 65 in the household) - 16340 (suibtr-act only if living in a city other than Moscow) + 9343 (add onb' ijhousehold has a refrigerator) - 12694 (subtract only if household is living in a rueral area)+ 5439 (add onlyif household has a car) - 5350*(Nu,mber of UneinploVed Hlousehold mnembers).
12. This estimated score can then be compared to the official subsistence minimum todetermine whether the household was poor. Let's take some concrete examples and keepin mind that these data pertain to 1993-94. Family A consists of 4 members: a husband,wife, and two children. Family B consists of two pensioners, but only one of them is over65. Both families have a car, both families have a refrigerator. Family B lives in a ruralarea but Family A lives in a city outside of Moscow. Conventional wisdom might suggestthat Family B is poor, while Family A is not. After all, in Family B, there is one pensionerwho receives the average old age pension but who is older than 65, while the otherpensioner receives only the minimum old-agc pension. In Family A, the husband has areasonably good job, but isn't paid that often, so he earns about one-half of the averagewage (which is about 5 times the minimum wage). His wife was not able to reclaim herjob after her three years of maternity leave expired, and she is registered unemployed andtakes care of her youngest child, aged 4, rather than spend money on day-care. Which
7 The estimated coefficients had signs in directions opposite to what we know about poverty in Russia from othersources. For example, access to land was estimated to be negatively correlated with per capita expenditures,reflecting the technical problem of undervalued production & consumption of food.
4
family is poor? Here we define poverty as having an estimated per capita expenditure lessthan the per capita subsistence minimum, which in December 1993 was ruble 42,800.
13. In December 1993, the average and minimum wages were ruble 141.200 and14,600 per month respectively, while the average and minimum old-age pensions wereruble 41,900 and 26,300, according to Goskomstat. Of course, some pensioners do notreceive the minimum old-age pension. These are the so-called social pensioners who donot have sufficient work tenure to qualify for the minimum old-age pension. Let'scompare the case of an elderly female pensioner living alone in a town outside of Moscowwho receives the social pension (estimated at about 70 percent of the level of theminimum old-age pension)--call her Family C.
Table D-2: Heuristic Example of Three Families
Previously Family Faim ily FamilyEstimated A B CCoefficients
Estimated per capita consumption 41337 68319 40159Per capita subsistence miiiinimuI 42800 42800 42800Result of proxy means test POOR NOT POOR
POOR
14. To update the formula for Russia at a more recent date, it would be necessary tomultiply all the coefficients and the constant term by a number which reflected the changein prices from the end of 1993 to the period required: for example, the beginning of 1997.The problem with this sort of mechanical updating is that household behavior might havechanged during the period, and we are keeping these relationships constant. Further, therelationship between household expenditures and some of our variables might change withinflation. As a shortcut, we could divide our nominal data (per capita money income) bythe change in prices from December 1993 to January 1997 to create an updated formulafor Russia. In Russia, prices were approximately 9.02 times higher in January 1997 thanthey were in December 1993, according to the Russian Consumer Price Index (CPI).Since we already have a coefficient for per capita money income, it is simplest to dividethis number (0.62) by 9.02 to generate a new coefficient for per capita money income(0.6877). All the other coefficients would remain unchanged. The number that resultswould be as if "in the prices of December 1993" and would have to be compared to the
5
1993 December subsistence minimum to determine whether the household was poor in1997.
15. To customize the formula for Volgograd or some other area of Russia. we coulduse the change in the consumer price index for that specific region in our updating. Forexample, in Volgograd, prices changed by 8.82 times, so the coefficient for per capitaofficial income would be 0.07031 and all the other coefficients would remain unchanged.
Table D-3: Accounting for Inflation
Variables to use in scoring formula Russia updated to Jan Volgograd updated to1997 Jan 1997
jbmi_p (Per Capita Official Income) 0.06877 0.07031
16. There is very little difference between Russia as a whole and Volgograd, becausethe rate that prices changed in Volgograd was essentially the same rate as (average) forRussia. Areas with prices which increased more or less rapidly than Volgograd wouldhave different coefficients.
17. Note that the same families which were poor or not poor in 1993 would still bepoor or not poor based on their 1997 income adjusted for Volgograd. Here we make someassumptions about average and minimum wages and pensions (that the relationships arethe same in 1997 as they were in 1993) since the author does not have any data other thanaverage wages for November 1997 close at hand.
Table D-4: Proxy Means Tests Results for Volgograd J.anuary 1997(based on average wages & pensions for Russia in November 1997)
Volgograd Coefficients Family A Faillily B Family C
jbmi_p* 0.07031 104375 260999 140909CHILDN -6660.29 2 0 0ELDERLYN -5538.59 0 1 1DLOCI -16340.2 I 0 1RREFIGDA 9342.581 1 I 0DLOC2 -12694.5 0 1 0RCARDA 5438.918 1 1 0DLFSI -5350.54 1 0 0Constant 50618.78* Coefficient for nominal per capita official incomiie includes correction (division by 1997/1993factor) for inflation in Volgograd fromii Janulary 1997 to December 1993
Estimated per capita consumption 37727.57 65517.99 38647.3Per cap sub min 42800 42800 42800Result of proxy means test POOR NOT POOR POOR
6
II. SOCIAL ASSISTANCE PILOT PRELIMINARY RESULTS
18. Very preliminary results of the targeting experiments conducted in three oblasts ofRussia in 1998 as part of the Government's social reform program, supported by theWorld Bank through the Social Protection Adjustment Loan (SPAL). are available andused below. However, these data were obtained to early in the project cycle to be trueestimates of the targeting potential of the three methods tested. The final wave of fieldwork was conducted in September 1998 and data are being cleaned and processed. Thesedata will be ready at the end of November. and the simulations in this section will beupdated.
19. In each of the three pilots, household per capita income or per capita potential orestimated income or consumption was compared to the regional poverty line. Theregional poverty line was based on the Ministry of Labor methodology and local prices, asreported to the social assistance pilots team. The eligibility thresholds were a percentageof the per capita regional poverty line (50 percent in Volgograd and Voronezh, 35 percentin Komi).
20. In Komi, the methodology was said to determine the "Economic Potential" of ahousehold. In Voronezh, the methodology was intended to estimate the potential totalincome (sovokupniy dokhod) of a household. In Volgograd, the clients were informedthat their eligibility would depend on their "potential consumption." Although the namesdiffered, the mechanics of each methodology were fairly similar. Based on informationsubmitted by households on their official (wage plus transfer) income, demographiccharacteristics, and durable goods and/or assets, each methodology in essence tried toestimate how much more the household was or could consume than its reported income.
21. For example, in Komi, the "'module of economic potential" valuation depended toa great deal on whether or not the household had an imported or other automobile orexcess living space (more space than the 'norm'), based on assumptions and hidden expertvaluations. This potential income was added to the household's reported income todetermiiine eligibility. In Voronezh, an agricultural area, the methodology was intended toestimate agricultural income, based on the household inventory of livestock. Expertvaluation was used to generate average prices for the various kinds of livestock.
22. Volgograd was the only methodology that was openly derived from householdsurvey data. Using annual data for 1996, household per capita consumption wascalculated, then estimated by step-wise linear regression analysis. The correlationsbetween consumption and variables (estimated beta coefficients) such as the number ofchildren or whether the household had a private plot of land were estimated. Thesecorrelations were then used in a formula (which was updated for inflation) to determineestimated household consumption.
23. It seems that the actual methodologies employed in the three pilot oblasts were aseffective in identifying the poor as simulations performed. Based on the criteria of 1/2 ofthe subsistence minimum in Volgograd and Voronezh, and 35 percent (also called GDD)in Komi, the non-poor were rather well identified by the pilots in practice, and the poor
7
somewhat less so (Table D5). The estimated proxy means tests were usually slightlybetter at predicting the poor than the actual methodologies, but not strikingly so except inthe case of Komi.
Table D-5: Actual and Estimated Identification Rates in Pilot Oblasts & Russia
Percent Total Percent Poor Identified Percent Non-Poor IdentifiedIdentified
Evs,rimaed By Proxv Means Test RegressionisVolgograd 71 % 36% 77%Vcronezh 80 % 21% 90 %Kcmi 78 % 37% 91 %Russia 70 % 57 % 77 %Notes: In Volgograd and Voronezh. the poverty standard x% as actual orestimated consumption compared to one-halfof the oblast's subsistence minimum. In Komi. the poverty standard was actual or estimated consumption comparedto 35 percent of the oblast's subsistence minimumri (also called GDD for guaranteed per capita income).More detailed inobrmation presentcd in Appendix tabics.Estimate for Russia from Braithwaite, Grootaert, and Milanovic (1998).
24. The identification rates are quite high overall, both in the actual cases, and in theproxy means estimates. All do as well or significantly better than the simulation for allRussia. Unfortunately, neither the actual methodologies nor the proxy means estimates doa very good job of identifying the poor. The all-Russia simulation indicates that it iseasier for the regression-based methodologies (proxy means tests) to distinguish the non-poor, and this is even more true for the actual methodologies employed in the three pilotoblasts.
25. The reason for this is not the fault of any one methodology, but rather a generalindication of how difficult it is to distinguish the poor from the non-poor in Russia(Braithwaite 1995, Klugman 1997, Klugman and Braithwaite 1998) and in other FSUcountries more generally (Braithwaite, Grootaert, and Milanovic 1998). Unfortunately,the lack of very sharp poverty correlates translates into actual methodologies and proxymeans estimated methodology with high rates of exclusion (poor who do not receive thebenefit). And even worse, in spite of the high degree of accuracy of the three actual pilotmethods in identifying the non-poor, inclusion (payments of a benefit to the non-poor;also called leakage) rates are still quite high (Table D6). Exclusion is calculated thenurnber of those who had per capita consumption below the eligibility standard and didnot re.ceive a benefit (or would not have received in the proxy means tests simulations)divided by the total number eligible. Inclusion is calculated as the number of non-poor(based on their actual per capita consumption) who none the less received a benefit,divided by the total number of beneficiaries per oblast.
8
Table D-6: Actual and Estimated Exclusion & Inclusion Rates in Pilot Oblasts
Estitnated by Proxy Means RegressioniVolgograd 64 % 79 %Voronezh 79 % 74 %Komi 63 % 46 %Notes: In Volgograd and Voronezhi, thc povcrto stanidard was actual or estimated conlsumilptioIn compared to one-hal-oi'thc oblast's subsistcnce minilum. In Koomi. the povertx standard was actual or estimated consumption comparedto 35 percent of the oblast's subsistence minimulLIm (also called GDD for guaranteed per capita income).More detailed information presenited in Appeindix tables.
26. The reasons underlying the high rates are likely to be different. For the high ratesof exclusion, it is clear that neither the actual methodologies nor the simulated proxymeans tests can do a very good job of identifying the poor. The ability of either the actualmethods or of the simulations to identify the poor does depend on the poverty line used(Appendix table). The methodologies all do better in identifying the poor when thepoverty line is very low, but performance declines as the poverty line is raised to 50 or100 percent of the local subsistence minimum, and comes at a cost of higher rates ofinclusion.
III. PROXY MEANS SIMULATIONS FOR RuSSIA
27. Proxy means test simulations have been conducted for Russia and neighboringcountries in a World Bank research project (Braithwaite, Grootaert and Milanovic 1998;Grootaert and Braithwaite 1998). These simulations were based on round IV of theRussian Longitudinal Monitoring Survey, for which the fieldwork was conducted in early1994. In this research, we identified the determinants of poverty and welfare in the FSU,and also explored the poverty profile both in terms of headcount and depth of poverty.The findings clearly suggested a link between such easily identified household attributesas location, the number of children and elderly members, and whether the household isfemale-headed and the poverty status of the household. Certain traits, such as the link tothe formal labor market and a household enterprise, were associated with higher levels ofwelfare.
28. Under the previous Soviet social welfare system, most benefits were categoricalones. For example, all males aged 60 and over received some sort of pension (regardlessof whether they continued to work), which was also the case for all females aged 55 andabove. Starting in 1992, all children under the age of 16 (or 18 if they were full-timestudents) were eligible for a general child allowance. Certain categories of people,particularly the disabled (Groups I, II, and III) received diverse benefits, such as free or
9
reduced-price utilities and transportation services. Universal benefits such as generalizedconsumer subsidies were removed in the course of stabilization programs. but may havetainted some of the categorical programs as well.
29. Since many of those who received such categorical benefits were demonstrated toactually be the non-poor (see various World Bank poverty assessments) while universalconsumer subsidies were shown to be fiscally unsustainable and highly inequitable.categorical targeting received significant and warranted criticism from external andinternal advisors and policy makers. However, the problem with categorical targetingmay have been in the poor choice of categories more so than the idea of using an indicatoror combination of indicators (a proxy means test) to identify the poor. The choice ofcategories was dictated by political considerations (relating to the labor theory of valueand whether a person was perceived as being able to work or not). not by a careful studyof who was poor and what determined poverty.
30. In this section, we try to determine whether a combination of indicators canidentify the poor, which in turn would provide the necessary information for effectivetargeting of cash or in-kind benefits, or for active labor market policies. In practice, in theFSU, and particularly in Russia and Ukraine. increasingly benefits are being awarded toapplicants who meet a categorical filter amd an income-test. Typically. this means-test isbased only on official income. Unfortunately in the FSU. official income alone is aparticularly poor predictor of household welfare, due to the pervasive informal sector andthe general unwillingness of households to disclose such sensitive information.
31. Preliminary evidence from the housing allowance subsidy programs in Ukraineand Russia, which are based on official income (wages plus transfer income) suggest thatthis official income-test has a very high error of exclusion (those who are actually poor arenot receiving the benefit). Partly this originates from the very different goal of theseprograms, which is to promote housing privatization, and partly it may originate from alack of consideration of other factors related to poverty which are not captured in officialincome.
32. In order to improve means-testing where it currently exists, and to revise andupdate the categorical approach overall, we estimate an expanded welfare equation withvariables added for official income (wages and social transfers) and for ownership ofhousehold durable. Owing to the obvious endogeneity of these variables, no causalinterpretation should be assigned to the coefficients. The sole purpose here is to determinetheir predictive power. All of the regresses included (household durable, official income,family composition/demographic characteristics, location, unemployment status) are allfairly easy to identify by social workers, either through direct observation, declaration, orverification through documentation or a home visit. The model was estimated withforward stepwise regression.
33. The data in Table D-7 show that the proxy means test was able to identify correctlyapproximately 65-75 percent of the populations, with all three countries having betterpredictions for the non-poor than for the poor. Only about 60 percent (57-62) of the poorwere identified correctly, but this still represent a significant improvement over the
10
previous single-indicator/categorical approach used to allocate benefits such as old-agepensions and student stipends.8
34. At first glance, the five best predictors for the FSU countries seem to be morerelated to the non-poor side of the spectrum (wage income, car, color TV, householdbusiness, university education, land ownership) as to the poor (transfer income). Even so,the five best predictors achieved almost the same degree of accuracy as did the completemodel, identifying only slightly less of the poor (54-57 percent) and the populations (64-70 percent). Even the addition of the next five (best ten total) predictors shows a mixtureof factors associated with higher welfare (stereo, car, household enterprise) as with lowwelfare (number of children. transfer income, rural location, other urban location, numberof unemployed, inactive head). The addition of the next five best predictors does little toimprove the fit, raising the overall error rate only slightly (64-73 percent) and the errorrate for the poor a bit more (57-58 percent) than was observed by using only the five bestpredictors.
35. Given the presence of so many variables associated with the higher end of thewelfare distribution and the higher identification rates for the non-poor, we repeated thesecond simulation done for the Eastern European countries, and found vastly divergentresults. If there was some way to screen out the upper portion of the distribution, howwell would the proxy means test distinguish among the poor and non-poor in the lowerhalf of the distribution? For the Eastern European simulation, we assumed that the screenwould correctly identify the upper half of the distribution, since the identification rates forthe non-poor were all above 90 percent. Although this was a reasonable assumption forEastern Europe, in the original expanded regression for the FSU countries, only 70-80percent of the non-poor were correctly identified, thus making this assumption a bit morequestionable. However, for consistency, we simply re-ran the expanded welfareregression via forward stepwise regression on the half of the FSU samples with welfarebelow the median.
8 Analysis of individual countries (Russia, Kyrgyz Republic) in World Bank poverty assessments and comparativeanalyses found that in general, only child allowances were well-targeted transfers in FSU countries. Allother transfers were regressive or highly regressive.
11
Table D7: Stepwise Targeting Regressions (All Observations)Former Soviet Union
Estonia Kyrgyz Republic RussiaBest Five Predictors _
Wage income Wage income Wage incomeCar Car Transfer incomeColor TV Washing machine Color TVHigher education Color TV RefrigeratorTransfer income Land ownership Household enterprise
Second Best Five PredictorsStereo Number of children Inactive headHousehold enterprise Renter CarNumber of unemployed Household enterprise Location: other urbanInactive head Location: rural Location: ruralNumber of children Location: other urban Sewing machine
All Variables - % Correct PredictionsPoor 61.9 57.1 56.9Non-poor 77.1 68.6 75.5All 74.5 64.0 68.9Note: Dependent variable is the log of per equivalent adult expenditure. The regresses are the same as in the welfare andpoverty regressions with the addition of wage and transfer income and consumer durable.
36. The results in Table D-8 demonstrate that such an assumed screen wouldsornewhat improve the identification of the poor in Estonia (from 62 percent to 66 percentcorrectly identified) but would improve the identification of the poor much more in Russiaand Kyrgyz Republic, increasing to 80 and 83 percent respectively. Of course, there is acost to this--the few non-poor which remained in the below-median sample were eitherpoorly identified (Estonia), extremely poorly identified (Russia), or virtually unidentified(Kyrgyz Republic). This suggests that a proxy means test system could perform ratherwe]ll in Russia and Kyrgyz Republic, and acceptably well in Estonia, provided that aneffiective mechanism could be found to screen out the upper portion of the welfaredistribution. In all three cases even without the screen, the proxy means test wouldrepresent a significant improvement over the old categorical approach.
12
Table D8: Stepwise Targeting Regressions (Observations Below Median)Former Soviet Union
Estonia Kyrgyz Republic RussiaBest Five Predictors
Wage income Land ownership Wage incomeTransfer income Wage income Color TVColor TV Car Transfer incomeInactive head Motorcycle Education: primaryNumber of unemployed Renter Refrigerator
Seco1d IBest Five PredcictorsL.and ownership Washing machine' Education: higherLocation: rural Land ownershipL.ducation: voc.-tech Household enterpriseCar ReniterWashing imachine Number of elderly
,4// Vzaritahles - % Correct Predlictionsfloor 65.5 83.1 79.5Non-poor 61.1 8.7 21.8All 65.1 81.3 72.9.'oiL l)ependent variah1c is the log ol'per equivalent aidult expenditure. 'I'he regresses are the same as in the welvfare andpoeCrty regressions wkith the addition of' age and transler income and consumer durable.
O()nl six variablcs mct the: cntr criteriii.
37. Further, in all three countries, the five best predictors alone did as good a job inidentifying the poor (Kyrgyz Republic. Russia) or almost as well (Estonia) as did the fullmodel, implying that only a few key data would be required for collection. As in EasternEurope, the set of predictors which emerges as the best for identifying the poor (given thatthe upper 50 percent of the distribution was screened out of consideration) is more or lessthe same as which resulted from estimation over the full sample. Interestingly enough, forKyrgyz Republic, using the below-median observations resulted in only six variablesmeeting the entry criteria for the forward stepwise regression: land ownership, wageincome, car, motorcycle, renter status, and washing machine. For Russia and Estonia,more than 10 variables entered into the forward stepwise specification.
38. Wage income was still the best indicator in Estonia & Russia, but was displaced tosecond in Kyrgyz Republic by land ownership (which itself was in the top five for the fullsample). In Estonia, ownership of a car and higher education have been replaced by
13
inactive head and number of unemployed, which seems logical enough, but in KyrgyzRepublic, color TV and washing machine have been replaced by renter status andmotorcycle, of which the latter is a bit more difficult to rationalize except to speculate thatit served well to identify the few non-poor households with below-median welfare. InRussia, moving from the full sample to the restricted one meant that household enterprisewas replaced by primary education of household head, which repeats the logic of theEstonia findings that a factor more associated with poverty would become moresignificant with the restricted sample.
39. Given the fact that restricting the observations to those below the median bothsignificantly improved identification of the poor in Kyrgyz Republic and Russia but verydramatically worsened the identification of the non-poor prompted an additionalexperiment with other regresses. in an ultimately futile attempt to improve the predictionsof household consumption. Adding such "kitchen sink" variables as housing amenities(hot water, central heating, etc.) and an additional dummy variable for self-employedhouisehold head, resulted in error rates which were virtually identical to those for theoriginal specification for Estonia and Kyrgyz Republic and which were only marginallybetter (2-3 percent) for Russia, and are therefore not further considered.
40. Overall, the acceptability of the proxy means test for the FSU countries depends onthe reasonability of the assumed screening device. Unlike in Eastern Europe, 95 percentor more of the non-poor can not be assumed to be removed from consideration through aninventory of their consumer durable and other factors. Only approximately 70-75 percentof the non-poor could be removed at best in the FSU. Once the non-poor are removedfrom consideration, virtually the same information collected could be used to furtherrefine the identification of the poor and non-poor in the remaining portion of the welfaredistribution, resulting in identification rates of 65-82 percent. Although the FSUperformance is not quite as impressive as in Eastern Europe, it is still a significantimprovement over the previous system of categorical indicators, which was plagued byvery large leakage to the non-poor.
IV. PROXY MEANS TEST RESUlTJI1S FOR PANEL, 1994-96
41. The World Bank version of the panel based on the UJNC dataset was used toestimate proxy means test results formulas, and error rates were checked for the predictedoutcome versus the observed details. Of course, attrition bias could also affect theregression results. Not withstanding possible attrition bias, the results of the proxy meanstest for the panel are quite good for identifying the non-poor and acceptable for identifyingthe poor (Table D9 and D10).
42. This annex has presented four different. sets of proxy means test regressions foridentifying the poor and the non-poor in Russia. First, a model for Russia and Volgogradwas set up and estimated. Second, results from simulations and actual outcomes from thefor World Bank social assistance pilots were compared. Third, simulations for Russiawere compared to two neighboring countries, Estonia and Kyrgyz Republic. Finally, withthese results in mind, proxy mean tests for Russia using our panel (World Bank version ofthe RLMS dataset) were estimated.
15
Table D1O: PROXY MEANS TEST RESULTS FOR PANEL, 1994-96
Predicted values of logarithm of per capita household consumption were computed in the following stepwise regression
LNPCEX 5 Logarithm of per capita expenditures (cash expenditures plus in-kind consumption)PCINC_5 Per capita cash incomeAUT05 Automobile (Ono. I-,es)BWTV5 Black and white I'V (Ono. I=ves)1151111.ANI) Access to land (0)no. I =ves)FRIG5 Refrigerator (0)no. I=yes)111111_AGE5 Age of household headIIIIII_ID )U5 Years of education of household headIHII_CilGN5 Gender of household head (0=female. I =male)IIIIIISQ5 Age soared of household headIV5 Color'l V (0 -no. I-yes)VCR5 VCR (0-(no. I yes)WASH5 Washing machiie (0=no, I=vcs)NCH S Number of children in the householdNELDER 5 Number of elderly in the householdNUEM_5 Number of unemployed (those wN ho do not report any work. receis e neither pension nor disability henefit and would like to work)RURURB type of'settlement (O=urban incIliding Metropolises i.e. Moscom and St. Petersburg. I =rural)NEDUC_5 Nurnber of people in thc household %sho have undergraduate or graduLate degree (have diploma from Institute. U niversity etc
Or Graduate School i.e. those who answsered "yes" (I) for i5insuni or i5gradre)
ANNEX FIVE
Tide Welfare Mobility, Poverty and Inequality in Russia In 1994-1"996 -
Authors Elena GlinskayaCarolina Population Center, University of North Carolina at Chapel HillJeanine BraithwaitoWorld Bank
O)bjectives Examine households' one- and two- year trnsitions between quintiles of =comc andiexpenditure distributions.
Investigate relationship between changes in position within distributionsof xfeitarandicharacteristics of households.
Data Three waves (1994, 1995, 1996) of the Russia Longitudinal Monitoring Survey (RLMS)
Empirical model
,* - P,,, = 30j + 03
4,j ' + 2q/ X1,,, + P33. X2 is,j ,4qj li,- e
for P=1.2.3,4.5
-*w,ere:I, indicates initial quintile of distribution, at time to. j= {l..5)p.. indicate percentile in thc welfare distribution at time t.p,1., indicate percentile in the welfare distribution at time r-/,I binasy indicator. tak-es value i' 1" for the observation fiom 1995-1996, and 'O" for; t!94-1995:
XI,, vectc' of characteristics of the household head:a: and age squared of the household bcad,education and cducation squared of the household head,occupation of the household head,St .:der of the household head;
X2,, vectoi of characteristics of demographic composition of the household and indicators of presence ofhousehold members from specific groups:
IF rPportion of chiTdren aged I months 6 years in the household,pFsportion of cihildren aged 7 years - 17 years in the household,pr, portion of active females (18-60 years old) in the household,propbortion of retircd males (61+ years old) in the household,preportion of retired females (55+ years old) in the household,houschold sizc
in' iicator for presence of handicaped persons in the householdindicator for presence of persons on maternity leave in the households
ZL, vector cif indicators:eight regions of the Russian Federationrural
Oneyearcnge lo positionwithin distribution ofequivalent household expenditure by the occupationof the bhusehold head, conditional oan being in the specfic quintile in 1394 distribution, number ofpercentiles moved:
Occupation of the household head Bottom 20W 20-40% Top 20 %(one digit ISCO title) (lowest quintile) (second quintile) (Highest quintile)
Managers 13 34 _
Profcssionals 14 -24
Technicians and associate professionals 13 16 -20
Clerls 13 16 -22
Service workers and shop and mark-et sales 13 20 -26workcrs
Agricultural work-ers 13 3 -30
Craft and related trades workers 13 16
Pl ant and machine operators and assemblers 13 16 - 8
Elmenlen:y occupations 13 14
I Not working f 13 j 1 .Valuc for tL oher regressions sei 1he following: re-ion - Vulga Basm. No ?iandicapped or 'on Inacemirv ic ' persons auxpresent in tL.. houschuld. klouschold consists of a chlild 0-h. an active male and au urive female Aun active Uaho al ;e i. u,idi 12years of cduL ion is considered as a bouschold hlcad.
Among 1ic: poorcst:-There is no ditfference in patterns ofwelfare mobility bv tie occupation of the lhuuscnl_k4 hclid.
Among those who started bctwcezi 20 and 40 percentiles:Households headed by 'Managers" are the most upward mobile. These are followcd by the
households headed by "Service and Shop sales" work-ers. Households of agricultural workers are the mostlikcly to e.:hibit downward mobility. Households with non-working heads do not exhibit relative dowvnwardmobility.
4nmong the wealthiest:Houscholds of"Associatc professionals"and "Plant and machine operators and assembibers" are less
likely to Icavc thc top quintile of thc welfare distribution. Households of agricultural workers aic die oneswho are the most likely to leave the top quintile.
2
Otae yeardagtei positiOn witCl distribution of equivavth*oubold expenditure by the educs.tionof the household bhed, conditioal on being in the specific quitile In 1994 distribution, number ofercenties moved.
Education of the household head Boam 20A. 2040% Top 20 %(lowest quimile) (second qwnnle) (Higcst quindle)
7 years of education or less 13 16 6
8 years 13 16 -
9 year-s ~~~~~13 16 -24
10 ye= ~~~~~~13 16-2
3 6 . _ 1
12 yas 13 16 -22
13. years 13 16 -_
14 yeas 3 6 -20
5ye=.13 -'l
16 year.; 13 _ 1
I year. 13 22 1
+ _.24Value for W= OUter regmaions sel to hc following: tcgion - Volpa Basin. No hwidic.Vppd or 'on matomity lea .. Paso" amc
psctin Utl- iunscwod. Housnctold consists ota.child 0.6. an aaive male saud anf =cive tci?iac AA active maLn anj -40. %varking_a a aaf.s. .is ctnsidemd a ..ouschold hod.
F-ducation of the houselhold head has no effect on probabilities of upwards mobility of the poorest,households.
lncrease in educ~ationof the household head nearly allows the household to escape fromr the second lowest.quizftile of expenditue distribution.
Hiouscboldi wtid college'-ucaica heads (15+ years of education) are more likely to stay at thic top of the&xpendiruzc distribution.
3
Simui on msuIts
One year change in position within distrioution of eq uivalent household expenditure by the householdcomposition,conditionalon being in the specific quintile in 1994 distribution, number of percentiles*moved:.
I child 0-6 years old 18 13 -25child 7-18male 19-60 years oldfemale 19-55 years old
2 cbild 0-6 years old 21 16 on-male 19-60 years oldfemale 19-55 years old
3 male 19-60 years old 23 13 -19female 19-55 years old
4 child 0-6 years old 25 19 -20female 19-55 years old
if fem i55+ years old 35 10 -24
6 male 60+ years old 37 13female 55+ years old
7i male 1 -60 years old 18 I1 -20
Value for the other regressions set to the following. region - Volga Basin. No handicapped or 'on matcmity Ica-mc persons arepresent in the household. In cascs 1,2.3 an active male nged 40. with 12 years of education working as craftsmeo is wonsidered asa household hcad. In case 4 an active female with the saene characteristics as the above is considered a household head. In cases4 (6) a non-wvorking 60 ycars old female (male) with 12 years of education is considered a household head.
Among the poorest:Households with retired members arc thc most likefy to escape from the bottom quintile ofexpenditure distributioti.
'Households with children and households of men living alone exhibit the lowest upward mobility.
Among those who started between 20 and 40 percentiles:'One parent one child" families and "two parents one child" families tend to improvc their relativepositions the most.
Among the wcalthiest 20 percent of the households:Households of men and women withoutcchildren or other dependents tend to be least likely to drop
of from the top of welfare distribution.
4
Estimation method OLS with cluster correction for the household spesific autogregressive errors
Findings
There is substantial mobility within the distributions of household income and expenditurcs.Between 1994 and 1995, the probability of staying in the lowest quintile of the per-capita expendituredistribution was 0.46, and between 1994 and 1996 (the two year probability) it was 0.39. Persistence ofremaining in the top quintile of the per capita distributionsis higher than the persistence of remaining in thelowest quintile.
a Accounting for mobility within the distributions of the housceiolds' welfare decreasei measurmdinequality.
- The pattem of mobility between quintiles of income distributions is similar to the patzern ofnobility between quintilesof cxpendituredistributions. This might indicate that households find it diftcultto smooth their current consumption following fluctuations in current income.
* Demographic composition of the household is a significant determinant of degree and directionof mobility at all points of the initial welfare distribution.
* Occupation of the household head is a significant dcterminant of the households proh-ability ofmaovin" fiv the households in the central and higih percentiles (i.e. all, except bottom 20 perce-viles) ot theinitial weitire distribution.
* Y. .rs of education of the houschold head are positively related to upward moveineiitb ni theincome an; expenditure ladders for the households from higher (top 40) percentilesof the initial ,;istribution.
* Th^re are no significant ditferences in the extent and direction of mobility by the age f thehousehold head. Among households in dte central part of the initial distribution (40-80 percentiles) femaleheaded households show higher upward mobility than male headed households. There is no e6idence thatpatterns of households'welfare mobility differ by the presence of handicapped members and members onmaternity lcave.
* Patterns of households' welfare mobility are significantly different among the regions of theRussian Federation. At all points of the initial distribution. households residing in the metrovolitan iteas,Moscow and St. Peterburg) tend to irnprove their reiative positions. i nese is some evide,nce that ruralrcsidt:t.s are *nore likely to exhibit downward mobillity.
5
Oneyearchangeh positionwithindistnrbutionof equivalent householdexpenditure by the occupationof the household head, conditional on being in the specific quintUle in 1994 distribution, number ofpercentiles moved:
Occupation of the household head Bottom 20°/e 2040% Top 20 %(one digit ISCO title) (lowest quittile) (second quintile) (Highcst quintile)
Managers 13 34 4
Professionals 13 14 -24
Technicians and associate professionals 13 16 -20
Clerkus 1;3 16
Service work-ers and shop and market sales 13 20 -26workcrs
Agricultural workers 13 -8 -30
Craft and related trades workers 1 3 1
Plant and machine operators and assemblers .13 16 18
Elemen= :y occupations 13 14
Not workLing 13 12Value for tW other regressions set to the following: region - Vulga Baia6n. No handicapped or 'on macemi le .- persons armpris.2nt in tr.. household. Household consists of a child 0-. an active niale and ani active female Ani activc male a1 41). %vith 12years of edL;L aion is conisidered as a household hcad.
Among t-n; poorcst:Th1ere is no difference in patterns of welfare mobility by thie occupation of the houscti.d. hcad.
Among those who started bctwcen 20 and 40 percentiles:Households headed by "Managers" are the most upward mobile. These are followed by the
houselioldsheaded by "Serviceand Shop sales"workers. Households of agricultural workers are the mostlikely to e~.hibit downward mobility. Households with non-working heads do not exhibit relative dowvnwardmobiliry.
Among the wealthiest:Households of `Associate professionals"and "Plant and machine operators and assem biers" are less
likely to icavc thc top quintile of thc welfare distribution. Households of agricultural work-ers a;c the oneswho are the most likely to leave the top quintile.