Poi-icY RESEARCH WORKING PAPER    1942
Poverty  Correlates and                                                Social protection systems in
the transition economies have
Indicator-Based Targeting                                              been inadequate to meet the
in  Eastern  Europe and  the                                           challenges of transition, being
both costly and poorly
Former Soviet Union                                                    targeted. The largest group of
poor people is the working
poor - especially workers
Christiaan Grootaert                                                   with little education (primary
Jeanine Braithwaite                                                    education or less) or outdated
vocational or technical
education.
The World Bank
Social Development Department
and
Europe and Central Asia
Poverty Reduction and Economic Management Sector Unit
July 1998



| POLICY RESEARCH WORKING PAPER 1942
Summary findings
Grootaert and Braithwaite compare poverty in three           social transfers (other than pensions) or other nonearned
Eastern European countries (Bulgaria, Hungary, and           income. But through sheer mass, the largest group of
Poland) with poverty in three countries of the former        poor people is the working poor -  especially workers
Soviet Union (Estonia, Kyrgyz Republic, and Russia).         with little education (primary education or less) or
They find striking differences between the post-Soviet       outdated vocational or technical education. Only those
and Eastern European experiences with poverty and            with special skills or university education escape poverty
targeting. Among patterns detected:                          in great numbers, thanks to the demand for their skills
d Poverty in Eastern Europe is significantly lower than    from the newly emerging private sector.
in former Soviet Union countries.                               * The poverty gap is remarkably uniform in Eastern
* Rural poverty is greater than urban poverty.             European countries, especially Hungary and Poland,
* In Eastern Europe there is a strong correlation          suggesting that social safety nets have prevented the
between poverty incidence and the number of children in    emergence of deep pockets of poverty. This is much less
a household; in the former Soviet Union countries this is    true in the former Soviet Union, where those with the
less pronounced, except in Russia.                           highest poverty rate also have the largest poverty gap.
* There is a gender and age dimension to poverty in          In the short to medium term, creating employment in
some countries. In single-person households, especially      the informal sector will generate a larger payoff than
of elderly women, the poverty rate is very high (except in    creating jobs in the formal (still to be privatized) sector,
Poland) and poverty is more severe. The same is true in      so programs to help (prospective) entrepreneurs should
pensioner households (except in Poland). In Poland the       take center stage in poverty alleviation programs.
pension system has adequate reach.
* Poverty rates are highest among people who have
lost their connection with the labor market and live on
This paper is a joint product of the Social Development Department and Europe and Central Asia, Poverty Reduction and
Economic Management Sector Unit. The study was funded by the Bank's Research Support Budget under the research
project "Poverty and Targeting of Social Assistance in Eastern Europe and the Former Soviet Union" (RPO 680-33). Copies
of this paper are available free from the World Bank, 1818 H Street NW, Washington, DC 20433. Please contact Gracie
Ochieng, room MC5-158, telephone 202-473-1123, fax 202-522-3247, Internet address gochieng@worldbank.org. The
authors may be contacted at cgrootaert@worldbank.org or jbraithwaite@worldbank.org. July 1998. (112 pages)
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about
development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The
papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this
paper are entirely those of the authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the
countries they represent.
Produced bv the Policy Research Dissemination Center



Poverty Correlates and Indicator-Based
Targeting in Eastern Europe
and the Former Soviet Union
Christiaan Grootaert
Jeanine Braithwaite
The authors would like to thank RobertAckland, Branko Milanovic, and Gi-Taik Oh for comments
on an earlier draft, and Precy Lizarondo and Gracie Ochieng for document processing.



Table of Contents
Page
1.  INTRODUCTION ........................................................ 1
2. METHODOLOGICAL ISSUES IN THE MODELING OF POVERTY                   ..  3
3. POVERTY PROFLES   ..20
A. Eastern Europe .23
Location .25
Family Composition .26
Labor Force Participation .30
Gender and Age .36
B. Former Soviet Union .41
I. Pre-Transition Poverty and Macroeconomic Impact of Transition .42
II. Who are the Poor After Five Years of Transition? .44
Location .45
Family Composition .'46
Labor Force Participation .50
Gender and Children .54
4. MULTIVARIATE ANALYSIS OF WELFARE AND POVERTY..                            54
A. Eastern Europe                                                      56
Welfare Equations .56
Poverty Equations .62
Poverty Gap Equations .66
B. Former Soviet Union .70
Welfare Equations .70
Poverty Equations .76
Poverty Gap Equations .81
5. MEANS-TESTING AND INDICATOR-BASED TARGETING ............................... 85
A. Eastern Europe ................................................. 86
B. Former Soviet Union .................................................  91
i



6. S U             MMARY ....................  99
REFERENCES ..................                                             106
STATISTICAL ANNEXES .................. 111
ii



1.    Introduction
This paper undertakes a comparative analysis of poverty in three East European
countries (Bulgaria, Hungary, Poland,) and three countries of the Former Soviet Union
(FSU) (Estonia, Kyrgyz Republic, Russia). To that effect, we constructed a comparative
data set, whereby household survey data from the six countries were carefully checked,
cleaned and made comparable.  The resulting data set has been dubbed HEIDE
(Household Expenditure and Income Data for Transitional Economies) and its content and
construction method are described in detail in Ackland et al (1997).
Although our analysis of the HEIDE data found elements in common, the most
striking result is how different the post-Soviet experience with poverty and targeting is
from the East European one. Overcoming the Soviet legacy has not been as easy as the
generally positive East European prototypes would have suggested. Poverty correlates
for the FSU are not as sharp nor as well-defined as in Eastern Europe, yet poverty levels
are also higher in the FSU, presenting a larger challenge to governments as they try to
reduce poverty and improve targeting.
We have set ourselves three tasks in this paper. First, we construct a profile of the
incidence and depth of poverty in the six countries, using aggregate poverty indexes. The
aim is to find out what the common elements are in the profile of poverty in Eastern
Europe and the FSU, and which aspects of poverty are country-specific or bi-modal (e.g.
the immediate Soviet legacy of the FSU vs. the more diluted Soviet legacy of Eastern
1



Europe). If we find a large common element, it opens up the possibility of a region-wide
policy approach to poverty alleviation.
Second, we undertake a multivariate analysis of the determinants of poverty. This
overcomes the limitations of the one- or two-dimensional approach typically embodied in
a tabular presentation of a poverty profile. The econometric modeling work addresses
separately the incidence of poverty and the depth of poverty, using reduced-form
equations. Our objective is to find important correlates of poverty, and, where possible,
attribute causality to them. The results will also clarify whether the determinants of
welfare, such as the demographic characteristics of households and the returns to
household assets, differ between the poor and the non-poor.
Our third and most important task, is to derive a policy approach towards poverty
alleviation. Specifically, we wish to evaluate the role which means testing and indicator-
based targeting can play in channeling social transfers to the poor. In part because of the
socialist legacy, social transfers constitute a huge component of public expenditure in
Eastern Europe and the FSU, representing as much as one-fifth of gross domestic product
(GDP). The need to reduce these expenditures is pressing and the need for suitable
targeting devices is high. We will demonstrate the contribution which indicator-based
targeting can play.
Each of these three tasks is given a section in this paper (respectively, sections 3,
4, and 5). Before presenting empirical results though, we address in the next section the
relevant methodological issues.
2



2.      Methodological Issues in the Modeling of Poverty
In line with most recent work on poverty, the analysis in this paper is based on a
money-metric measure of utility and welfare. Total household expenditure is used as
measure of household welfare and as a basis to rank households and to define a poverty
line. Expenditure is preferred to income because it is usually better reported in household
budget surveys.'  Furthermore, there is the important theoretical consideration that
expenditure reflects better permanent income. This argument is particularly relevant in
transition economies where the volatility of current income is still quite high, due to the
lack of steady private sector employment and the resulting high rates of unemployment.
Arrears on the payment of wages and pensions, especially in FSU countries, further adds
to the unreliability of current income as a measure of welfare.
The analysis below takes into account differences in needs due to different
household size and composition and therefore uses household expenditure per equivalent
adult as the welfare measure. There is a wide choice of adult equivalency scales, and
different scales are used in different countries.  Our comparative analysis objectives
require the use of a single scale, and we have opted for the OECD scale, because of its
simplicity of use and wide familiarity. This scale is expressed as follows
This is only recently the case in East European countries. Prior to transition, income was usually better reported,
because most income sources were under the direct control of the state, and data collection agencies could verify
reported income at the source. This is why most pre-transition analysis of poverty has used income-based
measures. After transition, the emnergence of private sector income (especially self-employment income) has led
to a decline in reliability of reported income data, in line with the experience of western countries (see, for
example, Revesz, 1994 for the case of Hungarian income and expenditure data).
3



EPEQ =  (0.7)
where EXP is total household expenditure and n is household size.2 The OECD scale
reflects economies of scale due to household size but does not incorporate gender
differences.
Household expenditures were not deflated by a regional price index to take
potential differences in prices within the country into account. The reason is that, except
for Russia, the countries in the analysis are all fairly small and regional price differences
can be expected to be minor. For example, for Poland (the second largest country in the
set), regional price differences were found not to exceed 2 percent (Grootaert, 1995). For
Russia, informal calculations suggested that the effect on poverty estimates of corretting
for regional price differences was very small. During the period of analysis, several
countries experienced significant inflation and in these cases expenditures were deflated
with a month-by-month consumer price index. This yields real household expenditure per
equivalent adult as measure of household welfare.
A cut-off point needs to be selected to serve as poverty line across the distribution
of real household expenditure per equivalent adult. We rejected the use of an absolute
line, such as x dollars in PPP-terms, due to the wide variation in income levels across the
six countries. Indeed, it is not very meaningful to compare poverty profiles, when for one
country the profile pertains to less than 5 percent of the population and for another
2 For the household sizes typically found in Eastern Europe and the FSU, this fonnulation is a close equivalent of
the more conventional statement of the OECD scale whereby the first adult = 1, other adults = 0.7, and
children = 0.5. The exponential formulation however simplifies the calculations.
4



country to almost half the population. Hence, we opted for a relative poverty line, which
after some experimentation, was set at two thirds of mean household expenditure per
equivalent adult.3
Obviously, the exact position of the poverty line selected affects the results.
Individual country studies have shown that in certain ranges of the distribution, even fairly
small movements of the poverty line can have large effects on the estimated incidence of
poverty (see e.g. Grootaert, 1995 for Poland; Grootaert, 1997a for Hungary; World Bank,
1995b for Russia). -However, poverty profiles tend to be more robust than incidence
figures, and significant modifications do not tend to occur unless the poverty line is set in
the very lowest ranges of the distribution, especially in the lowest decile. Nevertheless, a
sensitivity analysis would be useful, and the earlier cited country studies contain analyses
with different poverty lines. The sheer bulk of tabular and regression results for a six-
country study make it impractical however to include a formal sensitivity analysis in this
paper. We refer the interested reader to the country studies.
Our selection of aggregate poverty index is the popular P-alpha class of poverty
measures introduced by Foster, Greer and Thorbecke (1984). This index is defined as
q I=I (
3 It is generally agreed that poverty measures should be calculated over individuals. Hence the relative poverty line
was defned over an individual distribution, under the assumption that each individual in the household has the
same welfare, equal to total household expenditure per equivalent adult.
5



where    n   =  number of people
q   =  number of poor people
z   =  poverty line
yi  =  expenditure of individual i
a   =  poverty aversion parameter
The poverty aversion parameter can take any positive value or zero. The higher
the value the more the index "weighs" the situation of the very poor, i.e., the people
farthest below the poverty line. Of specific interest are the cases where a = o and a = 1.
If a = o, the index becomes
p =q
n
which is the simple head count ratio of poverty, i.e. the number of poor people as a
percentage of the total population. While this is a useful first indicator, it fails to pay
attention to the depth of poverty. To do so, one also needs to look at the extent to which
the expenditures of poor people fall below the poverty line. This is customarily expressed
as the "income gap ratio" or "expenditure gap ratio" which expresses the average shortfall
as a fraction of the poverty line itself, i.e.,
z-y'
z
where y, is the average income or expenditure of the poor.
6



A useful index is obtained when the head count ratio of poverty is multiplied with
the income or expenditure gap ratio. This corresponds to
which reflects both the incidence and depth of poverty. This measure has a particularly
useful interpretation because it indicates what fraction of the poverty line would have to
be contributed by every individual to eradicate poverty through transfers, under the
assumption of perfect targeting. This can be considered as the minimum amount of
resources needed to eradicate poverty, given that perfect targeting is not likely to be
achieved in practice.
In the tables below we show the head-count ratio Po, and the ratio PI/Po, i.e. the
expenditure gap ratio.4 We prefer to call the latter "poverty gap" (PG) to highlight that it
is a measure of the average depth of poverty calculated over the poor only. In contrast, Po
and Pi are ratios which are calculated over the entire population (for a further discussion
of these measures, see Ravallion, 1993). In the tables below each of these measures has
been multiplied by 100 for easier interpretation.
The comparative poverty profile in the next section of this paper is based on one-
or two-dimensional disaggregations of the P-alpha index. While this yields a useful
4 In making this selection, we trade-off "ideal-ness" of the poverty measures for the sake of familiarity and ease of
interpretation. An ideal poverty measure must meet the monotonicity axiom (all other things equal, a reduction in
the income of a poor person must increase the poverty measure) and the transfer axiom (all other things equal, a
net transfer of income from a poorer to a richer person must increase the poverty measure). Neither PO nor Pi/Po
meet these axioms, but their product, Pi, meets the monotonicity axiom. In general, the P-alpha class of measures
meets the monotonicity axiom for a > O, and the transfer axiom for a > 1 (Foster, Greer and Thorbecke,
1984).
7



identification of important correlates of poverty, it cannot establish the relative importance
of each correlate (or determinant, if causality can be assumed). A multivariate model of
poverty is hence indicated. A basic model uses real household expenditure per equivalent
adult as dependent variable in a regression with exogenous household endowments and
characteristics as explanatory variables. Such welfare model is a reduced-form equation of
the various structural equations which express the income-earning and consumption
behavior of the household (see e.g. Glewwe, 1991). This model can explicitly recognize
the economic characteristics of the environment in which households operate. Consider
the following model:
E, =/3X, +82 W + s,    (1)
where Ei  =  real household expenditure per equivalent adult of household i
Xi =  a set of characteristics of households i
Wi  =  a set of characteristics of the economic environment of household i
A,2 =  model parameters
-aj  =  error term
While such model is not able to predict the effect of household characteristics on
specific income or consumption decisions (this would require structural equations), it
allows to observe the net effect of any given characteristic, holding all others constant, on
resulting household welfare. It is assumed at this point that there is no simultaneous effect
of household welfare on household characteristics so that no X, are endogenous. This
assumption is time-dependent, i.e., we assume this to be the case within some relevant
time period. (We revisit this issue below when discussing the specific variables to be
8



included in the model). With this assumption, simple OLS estimation of equation (1) is
appropriate.
From the point of view of understanding poverty, equation (1) is not necessarily
optimal. It unposes constant parameters over the entire distribution. It thus assumes that
the effect of a given household characteristic, e.g. education, is the same across the entire
welfare spectrum, and that the underlying structural equations do not differ for poor and
non-poor. One could say that-in this representation the poor are viewed merely as "rich
people with less money." This is arguably an incomplete representation. While one
should of course not see the poverty line as a barrier which divides the population into two
entirely different groups, it is certainly arguable that poor people face different (often more
severe) constraints, e.g. to obtain credit, to obtain labor market information, to set up
enterprises, etc. On the other hand, they may well be more adept at obtaining transfer
income. This calls for additional modeling of poverty.
There are several ways of addressing a situation whereby parameters can be
expected to differ across different segments of the distribution. One can estimate the
welfare regression separately for poor and non-poor, or introduce a set of interaction
variables (between a binary variable for poor/non-poor and the other right-hand side
(RHS) variables). Both methods are equivalent econometrically, but their estimation is
problematic. In the first method, each group (the poor and the non-poor) forms a
truncated section of the overall distribution, so that OLS estimation would lead to biased
estimates. The second method leads to the same result, because the binary interaction
variable is clearly endogenous-it is merely a binary representation of the dependent
9



variable. This endogeneity problem also rules out the use of a Heckman-type selection
model to, first, determine poverty status and, then, using the derived inverse Mills-ratio to
correct the welfare equations of the poor and non-poor groups. In practice, since the
poverty criterion is the same as the dependent variable in the welfare equations, it would
be very difficult to place an identifying restriction on the welfare equation.
A workable solution is at hand, however, if the situation can be seen as a censored
model, in which case Tobit estimation becomes possible. This requires the assumption
that equation (1) is the correct welfare model for the poor and that the same set of
explanatory variables determine whether one is poor or not. No assumptions are made
about the determinants of welfare of the non-poor (the process and the parameters could
or could not be the same). The model sets any expenditure level higher than the poverty
line equal to the poverty line, i.e. the data are censored at the poverty line.
E;'= E,    if   E c<z                               (2)
E = z      otherwise
where z = poverty line, and Ei is determined as in equation (1)
This model allows for the possibility of different parameters for the poor and non-
poor and can be estimated consistently if the error terms is assumed to be nornally
distributed (Maddala, 1983). Furthermore, a comparison of the estimated parameters of
(2) with those of (1) provides a test of whether the parameters of equation (1) do indeed
differ between the poor and the non-poor. This is especially relevant for the parameters of
asset variables, which measure the returns to these assets, and one can hence test whether,
10



for example, the returns to education differ between the poor and the non-poor (Appleton,
1995).
Conceptually, this model specification corresponds to modeling the poverty gap,
i.e. the poor's expenditure shortfall expressed as a ratio of the poverty line, i.e.
z-E,        for    Ei sz
z
Whereas this ratio is constrained between 0 and 1, the poverty gap itself is constrained
between 0 and z. In practice, it ranges between zero and z minus the lowest E, in the
sample, which is what equation (2) depicts.
When estimating poverty models on the basis of household survey data, it needs to
be recognized that such data are likely to contain a certain amount of measurement error.
If the error is limited to the dependent variable, it does not bias the estimated coefficients
(so long as the error is not correlated with any of the RHS variables), but it will affect the
variance-covariance matrix. A potential concern though is that the measurement error of
household expenditure may rise systematically with the level of expenditure.  This
increases the probability of correlation with RHS variables such as education, which is
positively correlated with the level of expenditure. This could lead to biases in the
estimation of equations (1) and (2).
The presence of measurement error has led severai authors to substitute limited-
dependent variable models for the continuous welfare equation. Gaiha (1988) used a
binary logit model to predict the probability that a rural household in India would be
11



poor.5 Diamond et al (1990) estimate a multinomial logit model on U.S. data to predict
the probability of belonging to an income quintile, conditional upon certain personal and
household characteristics. Diamond et al justify their approach, relative to a continuous
welfare regression, by arguing that the restrictions imposed by the functional form of a
levels regression (often linear or log-linear) may cause it to fit poorly on the actual
distribution, and demonstrate that this is the case for their U.S. data set. The multinomial
logit model allows for discontinuities in the underlying welfare model and thus also solves
the concern of imposing equal parameters over the entire distribution discussed earlier. In
the case of two groups (poor and non-poor) the approach collapses to a binary logit or
probit model, although then the underlying welfare model is again continuous (Ravallion,
1996).  There has been a recent debate in the literature on the merits of welfare
regressions versus binary poverty models.  Ravallion (1996) argues that the binary
response model is redundant, since the parameters measuring the effect of household
characteristics on the probability to be poor can be derived from the levels regression,
which is consistently estimable under weaker assumptions about the distribution of the
error. As argued in Grootaert (1997b), this argument applies if there is only random
measurement error and if a case can be made for imposing constant parameters over the
entire distribution.
As we discussed earlier, the latter issue has been dealt with in this analysis through
Tobit estimation of the expenditure of the "poor" segment of the distribution. The
possibility of systematic measurement error has led us to undertake also probit estimation
5 To our Inowledge, the first use of such model in the empirical poverty literature is by Bardhan (1984) in a study
of poverty in rural West Bengal.
12



of a poverty equation where the dependent variable is binary (poor/non-poor).
Explanatory variables are the same as in the welfare regression. It is clearly a judgment
call whether the loss of information embodied in the binary regression (collapsing the
entire distribution into two values) outweighs the risk of bias due to measurement error.
However, to the extent that results from a binary model confirm levels-regression results,
they can act as- a robustness test for the latter. In recent years, use of probit and logit
models (mainly the former) have become common practice in poverty analysis (see e.g.
Alderman and Ga7rcia, 1993; Lanjouw and Stem, 1991; World Bank, 1995d, 1996d;
Appleton, 1996; Grootaert, 1997b).
In summary, the determinants of poverty will be estimated in this paper on the
basis of three models:
(i)    OLS regression of welfare equation (1);
To account for differences in parameters between the poor and nonpoor, (the poor are not
rich with less money) without losing information from level-regressions:
(ii)   Tobit estimation of the welfare level of the poor, based on equation (2);
this is equivalent to modeling the poverty gap;
To solve the problem of non-random measurement error (especially mismeasurement as a
function of level of expenditure):
(iii)   Probit estimation of a binary poverty equation.
13



Each of these three models has the same RHS and we turn now to the discussion
of which variables can be considered exogenous household characteristics. As we pointed
out earlier, this is mainly a function of the time horizon considered relevant. It has become
fashionable in econometrics to take a rather narrow view on this (i.e., to consider a long
time horizon) and to estimate welfare models with very parsimonious RHS (see e.g.
Glewwe and Hall, 1995). As Appleton (1995) has argued, reasons can be found why
almost every conceivable determinant of poverty is simultaneously determined with
welfare, and he cites a number of examples of such discussions in the literature. In the
end, little more than gender, age and a few parental characteristics end up as truly
6
exogenous.  Such econometric purity is problematic if the analysis is meant to guide
policy. Most policy and targeting variables at the household level become endogenous if
the time period is made long enough. All assets (education, physical capital, land) as well
as household size are to some degree a function of the household's welfare level and its
evolution over the life cycle. Location can change due to migration. Likewise, the
household head can change as a result of migration, or the splitting of one household into
several households (or the reverse process).
While we recognize the strict validity of these arguments, for this exercise we have
taken a pragmatic view, and used a fairly generous set of RHS variables. The objective is
to identify determinants of welfare and poverty which, in the short run, are valid policy
and targeting variables. As relevant time frame, we consider the reference period for the
data collection, i.e. a year or less. We include therefore on the RHS variables which the
typical household in the six transition economies in question cannot change in a one-year
6 In an inter-generational context, even parental characteristics can be endogenous.
14



period or only with great difficulty or cost. This takes the specific situation of these
economies into account, and explains e.g. why some labor market variables are included
on the RHS. In a filly functioning market economy, occupation and labor market status
must be viewed as endogenous, but this is not the case in many transition economies.
Unemployment is high and largely structural, retraining opportunities are limited, and in
some countries, the supply of housing is not yet sufficiently flexible to permit easy
migration to areas of growing labor demand.
On the other hand, among the asset variables, we have not included ownership of
durable goods on the RHS for estimating the three models listed above. This is actually
more of a judgment call than it may appear. Until a few years ago, in the countries in our
analysis, such goods were rationed. With the possible exceptions of Hungary and Estonia,
there is not yet a fully operating market for these goods, accessible to the entire
population. Markets for durable goods such as personal computers, VCRs, etc. often
exist only in cities, and due to very high relative prices (compared e.g. to Western Europe)
accumulation and decumulation of such goods is rare for all but the very rich. For many
households, the existing stock is still largely determined by the pre-transition allocation.
Nevertheless this situation is rapidly changing.
Generally speaking, asset variables have to be seen as endogenous with respect to
household welfare, because in an inter-temporal context, the household's welfare level will
determine the extent of education children receive   and will determine capital
accumulation. For one-period model estimation, based on cross-sectional household data,
the case for exogeneity is stronger, but not absolute. In principle, it would be desirable to
15



replace these variables with instruments such as parents' education, inherited wealth, etc.
Unfortunately, such variables are not available in the data sets and our regressions include
productive asset variables on the RHS. In interpreting the regression results, some
caution will thus be necessary, not to view the estimated coefficients as measuring strictly
one-way causality from assets to welfare or poverty.
Using a one-year time frame, we consider as exogenous the following sets of
variables:
*     household  assets:   education, physical capital (house, household
enterprises), land;
o     demographic household characteristics: household size and composition
and characteristics of the head of household;
*     labor market connections: unemployment, and share of wages in total
income;
*     economic environment: location.
The human capital of the household is embodied in its members and hence their
numbers (by sex and age group) are introduced as regressors.  Since it is likely that the
education of the head of household has a greater influence on welfare and poverty
outcomes than that of other members, the education level of the head was introduced as a
separate regressor, by means of a series of dummy variables reflecting the highest level of
education achieved (primary or less, secondary, vocational/technical, university). The
earlier cited country poverty studies have indeed found strong bivariate correlations
between poverty incidence and the level of education of the head of household. The data
at hand do not provide information on work experience, but this can be proxied by age.
16



The age of the household head is also a good indicator of the stage in the life cycle of the
household.
Information on physical capital is somewhat scant in the data sets. We know
whether the household owns a farm or small business but have no information on the value
of its assets. Nevertheless, information on ownership (or use, in countries where legal
ownership is still unclear) is bound to be very important, because the emergence of small
private enterprises is a key feature of transition, and poverty among such entrepreneurs is
likely to be below average.
Ownership of a house is important in the same sense. In many cases it provides
the location for a household enterprise, and for many households it constitutes the main
asset against which it can borrow and from which it derives rental income (actual or
imputed).' In most transition economies, the supply of housing is still quite rigid, and a
housing market is absent in many locations. Housing ownership is still frequently the
result of pre-transition allocations by the state.  hence, there is a strong case for
considering home ownership as exogenous to the process of determining welfare.
Similarly, ownership of land is in most transition settings not yet a full household choice
variable, and, especially in rural areas, it is a key determinant of cash income and
consumption of food.
7 Ackland et al (1997) discuss in detail the procedures used for the computation of the value of housing services.
17



The link with the labor market is captured in the model with two variables: the
share of wages in total household income, and the number of unemployed household
members (in some cases this was replaced by the employment status of the head of
household, if this variable yielded a better specification). The case for exogeneity of these
variables rests on the fact that in the transition context, many of the labor market status
outcomes are determined, or at least greatly influenced, by the labor market status that
obtained prior to transition and/or by the macro-economic changes. Of course, it must be
recognized that personal characteristics do contribute to unemployment, or make it more
or less likely than a person will successfully obtain self-employment income. Again,
instrumental variables would provide a solution if they were available.  (E.g. one
possibility would be to use regional rather than household-specific labor market variables).
We kept these variables in the equation mainly because of their importance for targeting,
but again recognize the need for caution in interpreting the coefficients.
The way in which the household utilizes its asset endowment is a function of
various demographic household characteristics.  The demographic structure of the
household has been shown to have a strong relation with poverty incidence. Beyond the
number of children and adults, it is useful to specify the age and sex of the household head
because those factors may be related to the household's ability to cope with a changing
economic environment.
Lastly, the incidence of poverty is affected by the economic environment in which
the household operates. This relates especially to income earning opportunities and the
level of social and economic infrastructure. In a transition context, the household's ability
18



to adjust to a new economic reality will depend very much upon whether it lives in an
urban or rural area, in a large or small city, in an old industrial region, etc. In this
research, we will capture this by categorical variables for type of locality (capital or other
city, village).
Annex 1 shows the means and standard deviations for the full set of variables used
in estimating the welfare and poverty equations.
Apart from laying out the set of determinants of welfare and poverty (the objective
of Section 4), these equations can also be used to investigate how feasible means-testing
and indicator-based targeting is. Almost all East European and FSU countries rely on
these techniques to allocate social assistance and sometimes other transfers as well. If an
effective, reliable and low-cost test for income were available, there would of course be no
need for indicator-based targeting. In practice, most social assistance authorities find it
very difficult to apply means tests and find that applicants on average underreport income,
especially self-employment income. We wanted to test how many poor people could be
correctly identified based on a simplified means test and relying on easily identifiable
indicators. To that effect we re-estimated equation (1) with an expanded set of variables,
adding wage-income and public-transfer income (the two "official" and most easily
verifiable income components for most households), and also a list of durable goods
owned by the household.
As we discussed earlier, these variables are likely to be endogenous to the level of
welfare, but our objective here is simply to predict outcomes. Hence, we do not interpret
the estimated coefficients of e.g. TV ownership as the "contribution" of this variable to
19



welfare, but merely as a partial correlation coefficient incorporating all feedback effects
from welfare to durable ownership. We estimated the expanded equation (1) with forward
step-wise OLS, so as to identify the strongest correlates and best predictors first.8 as to
identify the strongest correlates and best predictors first. The results of this exercise are
discussed in Section 5.
3.     Poverty Profiles
The changing nature of poverty in Eastern Europe and the Former Soviet Union has
paralleled the sharp changes in economic management and in government in the region
over the past two decades. Even before the collapse of the Berlin Wall and the break-up
of the Soviet Union, East European countries had been experimenting with economic
reforms which brought their systems closer to market economies. Two of the early
leaders in such reform efforts, Hungary and Poland, are case studies for this analysis.
Hungary was arguably the first country in Eastern Europe to embrace economic reforms,
with its market-oriented New Economic Mechanism, and Poland's Solidarity movement
was an early large-scale populist movement towards more democratic government and a
freer economic environment.
Along with economic reform in Eastern Europe quickly came the labor market
consequences of shutting down non-profitable state enterprises. Unlike in the FSU, where
adjustment was much later and fell almost exclusively on real wages, in Eastern Europe,
s It would also be possible to use a step-wise poverty probit equation for this objective. However, most social
assistance authorities are interested not just in classifying applicants as poor/non-poor but also in detennining the
extent of the welfare shortfall. Hence, the OLS welfare equation is a more useful basis for a predictive model.
20



open unemployment along with real wage declines was characteristic of phase changes in
government and the economy. One paradoxical result of this is that poverty is much more
clearly defined in Eastern Europe than in the FSU, and the poverty profiles of East
European countries identify poverty correlates more clearly. This makes improvements in
targeting in Eastem Europe much more realistic to posit than in the FSU, where the poor
are not so well-differentiated from the not-so-poor.
Although this conclusion might seem somewhat surprising, it is not especially new.
Even with far more inferior databases, Atkinson and Micklewright (1992) concluded that
poverty in Eastern Europe was more defined and less all-encompassing than poverty in the
Former Soviet Union. However, during the reference period for their work (1991 and
earlier) the FSU had not broken up; nor had there been the sharp changes in the
macroeconomic environment associated with the dissolution of the FSU, so it is not
surprising that the earlier time period and the use of official data led Atkinson and
Micklewright (1992) to conclude that overall, FSU poverty was not as severe as in many
East European countries, but further, that poverty within the FSU  was highly
heterogeneous (see also Braithwaite, 1991).
With the breakup of the FSU, there were severe disruptions in the old trading and
monetary regimes. The demise of the ruble zone, the political ramifications of the
declarations of independence, the build-up of arrears in the payment for energy imports,
the difficulties in macro management of the newly independent countries, and the
difficulties in finding alternative suppliers for intermediate inputs (which in many cases
were highly specialized), all combined to result in catastrophic declines in GDP. Whereas
21



the aggregate decline in GDP for the East European countries was 10 percent during the
period 1990-96, it was 45 percent for the FSU. Especially sharp declines were registered
in 1993 and 1994, which were run-away hyperinflation years in most FSU countries.
Under these circumstances, it is hardly surprising that open poverty has increased
drastically in the FSU. Poverty also increased in Eastern Europe, but Eastern Europe
managed to avoid most of the macroeconomic disruption associated with the break-up of
the FSU, or if problems such as hyperinflation and collapsing real wages were
encountered, they were encountered much earlier than during 1990-96. As a result,
poverty in Eastern Europe has become much more like poverty in Western
Europe-highly correlated with the situation in the formal labor market and the skills of
individuals. As the poverty profiles below indicate, in the FSU, poverty is not well-
correlated with the nature of labor market participation of household members, but neither
is it -well-correlated with the lack of formal labor market ties. Basically, in the FSU,
poverty is more pervasive than in Eastern Europe and not as well-defined. It is much
more difficult to differentiate a poor FSU household from a non-poor one based on
observable correlates.
These qualitative and quantitative differences in the experience of poverty in
Eastern Europe and the FSU are demonstrated in by the cross-tabulation of poverty
correlates, headcounts, and measures of severity of poverty presented below.
22



A.    Eastern Europe
In Eastern Europe, the start of rapid transition in the early 1990s accelerated the
existing trend towards increasing poverty9. The main contributing factors were the loss of
employment in a suddenly contracting state-sector, without coincident emergence of
private sector employment. Rapidly rising unemployment has in fact been one of the most
visible signs of the social costs of transition. A number of East European countries also
experienced significant inflation (although it did not reach the level of the hyper-inflation
experienced by some FSU countries). Adjustments in wages, pensions and other social
transfers lagged behind, and real incomes for many people fell. However, the emerging
evidence suggests that these effects have been fairly short-lived. The three East European
countries in this study experienced less GDP declines than the three FSU countries, and in
the 1994-1995 period, they have each returned to positive growth.
The figures in Table I indicate that poverty rates as well as poverty gaps are lower
in the East European countries than in the FSU countries. As a reminder, poverty rates
measure the incidence of poverty as the percentage of population below the poverty line
(two-thirds of mean household expenditure per equivalent adult). The poverty gap
measures the depth of poverty as the poor's average shortfall in expenditures from the
poverty line expressed as a percentage of the poverty line. (Both measures were discussed
in detail in Section 2). FSU poverty rates exceed 30 percent and poverty gaps exceed
20 percent. Russia has the worst situation with a poverty incidence of almost 40 percent
and an average poverty gap of 30 percent. While the poverty rate of Kyrgyz Republic is
9 See Milanovic (1990) for an analysis of pre-transition trends in poverty.
23



higher (42.5 percent), its poverty gap is lower (25 percent) than Russia's. Hungary and
Poland show the most favorable situation with respective poverty rates of 21 percent and
23 percent. The poverty gap is slightly higher in Hungary (14.1 percent) than in Poland
(13.3 percent). It thus appears that poverty in Eastern Europe is much more shallow than
in FSU, which is good news from the point of view of poverty alleviation in Eastern
Europe. It suggests that as economic growth resumes, rising incomes may rapidly lift
many people above the poverty line.
Table 1: Poverty and Locality
Locality             Bulgaria Hungary Poland  Estonia    Kyrgyz       Russia
Republic
Headcount (P0, in percent)
Capital                             17.5    20.3    10.1    20.6        22.9        18.2
Other Cities                        20.5    17.7    16.9    31.6        38.0        38.4
Urban Subtotal                      19.9    18.5    16.2    27.5        33.3        35.8
Rural                               39.2    24.0    33.8    38.7        47.2        49.6
Total                               26.1    20.6    23.0    30.5        42.5        39.4
Poverty Gap (in percent) 1/
Capital                             19.9    13.9    13.4    19.7        24.0        20.7
Other Cities                        17.8    13.5    12.7    19.2        26.2        28.7
Urban Subtotal                      18.1    13.6    12.7    19.4        25.7        28.1
Rural                               21.7    14.6    13.8    21.9        24.7        33.2
Total                               19.8    14.1    13.3    20.2        25.0        29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
Notes: .
1/ The poverty gap is the poor's average shortfall in expenditures from the poverty line, expressed as a
percentage of the poverty line (this measure is also known as the expenditure gap ratio).
24



Location
The strong causal role played by changes in employment in creating poverty during
transition in Eastern Europe make it likely that transition economies will show strong
geographic patterns of poverty and that urban and rural areas will be affected
differentially.
This is confirmed by Table 1 which shows that in all three East European countries
rural poverty is higher than urban poverty. In Bulgaria and Poland, rates of rural poverty
incidence are almost twice the urban rates. In Hungary, the urban-rural difference is small.
Within urban areas, the differences between the capital and other cities are not so
pronounced. (This is a marked difference with the situation in the FSU in which capital
cities are markedly less poor than other cities). In Bulgaria and Poland, poverty rates are
slightly lower in the capital than in other cities, but in Hungary the reverse is true.
The depth of poverty varies less than the incidence of poverty in Eastern Europe.
In general, poverty is slightly deeper in rural areas than in urban areas, but within the latter
poverty is deepest in the capital cities. So, while East European capitals have generally
less poverty than elsewhere, the poor in those capital cities do have a greater shortfall in
expenditure than elsewhere. This situation is distinct from the FSU, where both poverty
incidence and poverty gap are lowest in the capital cities.
25



Family Composition
Almost all empirical work on poverty in Eastern Europe and the FSU has identified
a strong correlation between household size and composition and poverty incidence. In
Eastern Europe, the correlation is strongest with number of children. In each of the three
countries analyzed here, households with three or more children have poverty incidence
about double the national rate (Table 2). It does not matter much whether this is a nuclear
household or an extended household with more than two adults. The exception is
Hungary where the poverty rate in extended households with three or more children is
more than triple the national rate. This is because in Hungary extended households often
arise as a result of poverty, which forces separate households to merge in order to benefit
from economies of scale in housing and other expenditures.
The implication is that in Eastern Europe, poverty among children is higher than
average and the presence of children needs to be considered as a strong candidate
indicator for targeting. We will revisit this proposition in the following sections when
reviewing the multivariate results. The finding of a strong correlation between poverty
and the presence of children also constitutes a call to reform entitlement programs such as
family allowances which provides fixed amounts of money to households with children.
These allowances are probably not needed by the richer households, and they are clearly
insufficient to prevent households with many children from falling into poverty. A
possible solution is to introduce means-testing and to increase the amounts given to large
26



Table 2: Poverty and Family Composition
Bulgaria Hungary Poland  Estonia   Kyrgyz     Russia
Republic
Family Composition
Headcount (P0, in percent)
One Male Adult, No Children        33.1    24.2    15.6    32.5      40.0        52.5
One Female Adult, No Children      45.0    27.8    13.5    37.0      51.8        47.8
One Adult, One or More Children    23.4    32.1    28.2    43.5      39.7        45.0
Two Adults, No Children            27.4    17.9    12.2    28.2      40.1        37.4
Two Adults, One Child              15.2    20.1    16.1    30.5      42.4        37.0
Two Adults, Two Children           19.4    19.9    24.7    29.6      39.9        38.7
Two Adults, Three or More Children  61.3    38.1    43.3    28.5     49.1        64.2
Three or More Adults, No Children  22.7    13.9    16.6    24.4      37.0        30.2
Three or More Adults. One Child    20.1    17.7    20.2    27.8      35.6        35.8
ThreeorMoreAdults.TwoChildren      35.8    29.5    36.2    31.6      43.3        51.6
Three or More Adults, Three or More    55.9    71.1    46.2    57.6  43.6        60.4
Children
All                                26.1    20.6    23.0    30.5      42.5        39.4
Poverty Gap (in percent) 11
One Male Adult, No Children        26.0    17.9    22.4    34.3      39.5        42.0
One Female Adult, No Children      28.9    18.6    17.3    27.5      47.4        44.2
One Adult, One or More Children    25.0    20.8    19.7    24.3      26.8        35.9
Two Adults, No Children            20.8    13.8    14.5    20.3      31.8        33.5
Two Adults, One Child              14.3    15.1    13.6    18.8      27.7        26.9
Two Adults, Two Children           18.1    13.1    12.8    17.4      26.7        27.0
Two Adults, Three or More Children  22.0    13.5    13.6    16.8     27.0        28.3
Three or More Adults, No Children  16.2    12.6    13.2    17.3      26.2        27.6
Three or More Adults, One Child    18.6    12.7    12.5    16.2      21.8        26.1
Three or More Adults, Two Children  19.3    12.8    12.8    16.6     23.0        25.0
Three or More Adults, Three or More    24.5    13.8    11.7    17.1  22.8        26.6
Children
All                                19.8    14.1    13.3    20.2      25.0        29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
Notes:
1/ The poverty gap is the poor's average shortfall in expenditures from the poverty line, expressed as a
percentage of the poverty line (this measure is also known as the expenditure gap ratio).
27



poor households.   Grootaert (1995, 1997a) contains simulation exercises which
demonstrate, in the cases of Poland and Hungary, that this can be achieved in a budget-
neutral fashion, and that it has the potential of significantly reducing poverty among
children. In part, the potential success from introducing means-testing results from the
fact that the poverty gap is not higher among households with many children. This means
that on a per capita basis, the resources needed to lift these households out of poverty is
not greater than for other kinds of households. In fact, the uniformnity of the poverty gap
across different types of households, displayed in Table 2, is quite a remarkable feature of
poverty in Eastern Europe.
Apart from large households, poverty incidence is also above average in
households with one adult. The situation is especially bad in Bulgaria among women
living alone, where poverty incidence is 45%. Most of these are pensioners. In Poland, in
contrast, households consisting of one man or one woman have below average poverty
rates, reflecting that pensions in Poland are higher than elsewhere. In Hungary and
Poland, one-adult households with children have higher poverty rates than those without
children, and there is some evidence that such households are more likely to fall through
the cracks of the family allowance system and to not receive these benefits (Grootaert,
1995, 1997a). Poor one-adult households also experience deeper poverty than other poor
households: in all three countries, they have larger poverty gaps than any other type of
households.
While Table 2 expresses the composition of the household in terms of the number
of adults and the number of children, Table 3 indicates that the number of elderly among
28



the adults is also correlated with poverty. Except in Poland, households consisting only of
elderly have the highest poverty incidence and poverty gap. We return to this later when
discussing the age-dimension of poverty.
Table 3: Poverty and Aggregate Family Composition
Bulgaria Hungary Poland  Estonia   Kyrgyz     Russia
Republic
Family Composition
Headcount (P0, in percent)
No Children, No Elderly             18.3    13.0    13.3    23.8      37.6       31.9
Child(ren), No Elderly             25.0    23.9    28.1    32.8       43.1       40.5
No Children, Elder(ly),            39.0    26.1    18.1    39.4       43.8       43.8
Child(ren), Elder(ly)              28.0    22.2    32.7    35.2       42.9       51.5
All                                26.1    20.6    23.0    30.5       42.5       39.4
Poverty Gap (in percent) V1
No Children, No Elderly             18.8    13.6    13.6    22.0      29.4       32.2
Child(ren), No Elderly              19.4    14.1    13.2    18.3      24.2       27.1
No Children, Elder(ly)             21.1    15.1    15.5    22.8       34.0       35.5
Child(ren), Elder(ly)              19.5    10.5    11.9    19.2       23.9       27.0
All                                 19.8    14.1    13.3    20.2      25.0       29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
Notes:
I/ The poverty gap is the poor's average shortfall in expenditures from the poverty line, expressed as a
percentage of the poverty line (this measure is also known as the expenditure gap ratio).
29



Labor Force Participation
It is not surprising that labor force status is strongly correlated with poverty
outcomes in Eastern Europe. In all countries, wage earners and the self-employed have
the lowest poverty incidence and poverty gap (Table 4). Which of these two groups does
best depends on the stage of transition. In Hungary, with perhaps the best developed
private sector, and the earliest initiation of transition, the self-employed have the lowest
poverty incidence-slightly more than half the national rate. Elsewhere though, wage-
work still provides the better alternative.
Table 4 also shows though that being a pensioner sharply increases the odds of
being poor, except in Poland, and in all countries pensioners have above average poverty
gaps. The favorable situation of pensioners in Poland is due to the generosity of the Polish
pension system. Of all East European countries, Poland increased spending on pensions
the most: between 1988 and 1993, pension spending rose from 6.9 percent to 14.7 percent
of GDP (Perraudin and Pujol, 1994). One reason for this was the sudden swelling of the
ranks of pensioners by 1.5 million early retirees in the period 1989-1992. Furthermore, in
1992-93, the average pension in Poland was 64 percent of the average wage-the highest
ratio in Eastern Europe. Polish pensions were at that time also fully indexed (Milanovic,
1995). 10
I0 The pension systen in Poland is discussed in detail in World Bank (1993) and Peffaudin and Pujol (1994). For a
more general discussion of pension systems in transition economies, see World Bank (1994).
30



Table 4: Poverty and Socio-Economic Status
Socio-Economic Group of     Bulgaria Hungary Poland  Estonia   Kyrgyz      Russia
Household Head                                             Republic
Headcount (Po, in percent)
Wage Earner                         16.4    15.7    18.7    23.1      38.8        32.5
Self-Employed                      24.3    12.7    26.8    26.7       40.3        31.5
Pensioner                           44.3    27.4    19.4    47.7      57.0        52.6
Other Transfer Recipient            63.7    57.1    64.1    54.3      61.9        68.7
Other                               46.5    49.4    33.5    31.0      42.2        45.2
All                                26.1    20.6    23.0    30.5       42.5        39.4
Poverty Gap (in percent) 1/
WageEarner                          14.4    12.1    11.7    17.0      23.4        26.9
Self-Employed                       15.3    11.1    14.4    24.6      24.8        27.9
Pensioner                           23.7    15.9    14.4    23.1      28.6        35.8
Other Transfer Recipient            30.0    21.1    18.6    26.1      26.8        33.1
Other                              28.8    16.5    11.5    17.9       25.1        27.8
All                                 19.8    14.1    13.3    20.2      25.0        29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
Notes:
1/ he poverty gap is the poor's average shortfall in expenditures from the poverty line, expressed as a
percentage of the poverty linc (this measure is also known as the expenditure gap ratio).
31



While the self-employed are a new socioeconomic category in countries in
transition, representing people who have succeeded in adapting economically to transition,
there is also another socioeconomic category emerging of people who have fallen victim
to transition: those who have severed ties to the labor market, and who are unemployed
or irregularly employed, and for whom as a result transfer income (other than pensions)
has become the main source of income. This category of people has poverty rates that are
around 60 percent, and they also have poverty gaps which are above average. However,
except for this category of households, Table 4 again confirms the remarkable evenness of
the poverty gap across society. We already pointed at the uniformity of the poverty gap
across demographic types of households (Table 2) and the same uniformity is seen across
socioeconomic categories.
The specific effect of being unemployed is illustrated in Table 5 which shows the
poverty measures by the number of unemployed household members. In Hungary,
households without unemployed members have a poverty incidence of 16.9 percent. If
one household member is unemployed, the figure jumps to 30.5 percent, and it rises
further to 68.7 percent if three or more members are unemployed. In Poland, poverty
incidence is 19.7 percent in households without an unemployed member but 50.7 percent
in households with two unemployed members. Again though, the poverty gap is not
systematically related to the number of unemployed household members, indicating that
the social safety net does what it is supposed to do, namely preventing the emergence of
pockets of deep poverty. (Of course, this finding does not consider overall cost or
efficiency in achieving this result).
32



Table 5: Poverty and Unemployment
Number of Unemployed       Bulgaria Hungary Poland  Estonia    Kyrgyz      Russia
Members in the Household                                        Republic
Headcount (P0, in percent)
0                                     ...   16.9    19.7    28.4      42.0         37.6
1                                     ..   30.5    35.7    42.6       41.9         53.2
2                                     ...   39.2    50.7    53.1      54.4         73.7
3 or more                            ...   68.7    46.5    73.2       40.6         66.7
All                                 26.1    20.6    23.0    30.5      42.5         39.4
Poverty Gap (in percent) V/
0                                     ...   13.1    13.0    19.5      24.4         30.1
1                                    ...   16.0    13.6    22.2       26.5         28.5
2                                     ...   13.6    15.1    27.3      24.4         25.6
3 or more                            ...   17.9    17.5    30.8       33.9         39.7
All                                 19.8    14.1    13.3    20.2      25.0         29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
Notes:
I/ The poverty gap is the poor's average shortfall in expenditures from the poverty line, expressed as a
percentage of the poverty line (this measure is also known as the expenditure gap ratio).
The role of education in this process is made clear in Table 6. There is a distinct
difference between the East European and the FSU countries. In Eastern Europe, the link
between lower poverty and higher education is extremely pronounced, but in the FSU this
link is much weaker, to being almost non-existent in Kyrgyz Republic. In Hungary, e.g.,
the poverty incidence among households where the head has primary education or less is
33.9 percent, while in households where the head has university education it is
3.3 percent, i.e. ten times less.  The equivalent figures for Kyrgyz Republic are
43.2 percent and 37.6 percent. The other countries are somewhere in-between these
extremes.
33



Table 6: Poverty and Education
Education of           Bulgaria Hungary Poland  Estonia    Kyrgyz      Russia
Household Head                                              Republic
Headcount (P0, in percent)
Primary or Less                     41.1    33.9    33.0    41.5       43.2        46.2
Secondary                           15.6    10.4    13.1    30.0       49.2        40.3
Vocationa/Technical 1/              15.0    18.7    26.3       ...    41.3         39.5
University or Above                  8.9     3.3      3.8    12.7      37.6        25.9
All                                 26.1    20.6    23.0    30.5       42.5        39.4
Poverty Gap (in percent) 2/
Primary or Less                     21.6    15.6    14.7    22.7       35.4        35.4
Secondary                           16.1    11.8    12.8    19.4       29.6        29.6
VocationalTechnical 1/              12.6    11.9    12.3       ...    28.2         28.2
Universitv or Above                 14.6     9.8      8.1    14.8      24.0        24.0
All                                 19.8    14.1    13.3    20.2       29.8        29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
NVotes:
I/ For Estonia, secondary education and vocational-technical education are combined and shown in the
category labeled "Secondary." Definitional problems in the Estonian dataset precluded a separation of
these two kinds of education.
2J/ The poverty gap is the poor's average shortfall in expenditures from the poverty line, expressed as a
percentage of the poverty line (this measure is also known as the expenditure gap ratio).
34



This difference in the impact of education is clearly related to the stage of
transition. The further advanced transition is, the more a private sector emerges which
needs well-educated workers, with general education backgrounds which makes them
flexible and adaptable to the newly emerging skill requirements. Pre-transition vocational
and technical education, often geared towards traditional industrial occupations, is no
longer in demand. Similarly, low-skill jobs, of the type held by workers with primary
education or less, have disappeared in great numbers. The more advanced transition
countries such as Hungary and Poland have already experienced skill-shortages in fields
like engineering, computer science and the like, and this will further push up wages
received by workers with university education, and increase the wage-gap across skill-
levels. This is one of the main reasons why the distribution of wages has increased in
many transition economies (Milanovic, 1995, 1997).
Education is also the only dimension where the wage gap is not uniform across
categories in Eastern Europe. Workers with primary or less education have not only
poverty rates well above average, but the poverty gap is also significantly higher than for
other groups. Households where the head has a university education have the lowest
poverty gap of any category, along any dimension, displayed in the poverty profile. It may
be surprising that the poverty gap varies so much with education level, while it varies very
little with the number of unemployed in the household. In part, the reason is that
education is not used as a targeting variable for any transfer program (although our results
suggest that perhaps it should become a targeting variable for Eastern Europe-see
Section 5).  Although clearly low  education is in itself a contributing factor to
35



unemployment, many people with low education still hold full-time jobs (let us not forget
that they are a very large category:  in Poland and Hungary, about two-thirds of
households have heads with primary or vocational/technical education). Their wages are
low, and as our results indicate, often insufficient to keep them above the poverty line.
Still, as full-time workers, they do not qualify for any transfers (other than general
entitlements) to supplement their income. There is no immediate solution to this situation.
In the medium to long term, retraining and a general upgrading of schooling curricula will
reduce the number of people with low education. Also, people with low education are
older than average, and many of them will become absorbed in the pension system in the
near term.  Whether this will alleviate their poverty, depends partly upon policies
pertaining to minimum pensions.
Gender and Age
-     We already noted the correlation between household composition and poverty
outcomes, especially the association between the presence of three or more children and
high poverty incidence. Since demographic household characteristics are easily observable
and potentially useful targeting variables, it is worthwhile to look in more detail at the age
and gender dimensions of poverty in Eastern Europe.
Table 7 shows that female-headed households have systematically higher poverty
incidence and poverty gaps than male-headed households. The difference is slight in
Poland, but more pronounced in Hungary and Bulgaria. The multivariate analysis in the
next section will confirm that such a gender-effect remains even after controlling for the
36



characteristics of female-headed households that are strongly correlated with poverty such
as low education.
Table 7: Poverty and Gender of Household Head
Gender ofHousehold Head      Bulgaria Hungary Poland  Estonia    Kyrgyz      Russia
Republic
Headcount (P0, in percent)
Male                                24.0    19.1    22.7    27.9       41.6         37.8
Female                              40.5    25.6    23.7    39.1       50.5         46.0
All                                 26.1    20.6    23.0    30.5       42.5         39.4
Poverty Gap (in percent) 11
Male                                18.6    13.3    13.0    18.7       24.7         28.5
Female                              24.5    16.0    14.2    23.9       27.5         34.5
All                                 19.8    14.1    13.3    20.2       25.0         29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
Notes:
I/ he poverty gap is the poor's average shortfall in expenditures from the povertv line, expressed as a
percentage of the poverty line (this measure is also known as the expenditure gap ratio).
The age distribution of poverty in Table 8 highlights the extent to which poverty in
Eastern Europe is concentrated among the very young and the very old. The average
poverty incidence in Poland is 23 percent, but among children under ten it exceeds 30
percent. The numbers for Hungary show a similar pattern. In Bulgaria, the relative
concentration of poverty among children is actually least. This is not a contradiction with
the earlier finding that in Bulgaria poverty rates among households with three or more
children are very high, because such households are quite rare in Bulgaria (much rarer than
in the other two countries). Hence, in Bulgaria most children live in households with one
or two children where poverty rates are lower.
37



Table 8: Poverty and Age
Age Bracket              Bulgaria Hungary  Poland  Estonia       Kyrgyz       Russia
Republic
Headcount (Po, in percent)
0-4                                    29.0      30.0     35.3     33.3       46.4          47.9
5-9                                    28.2      26.0     31.6     32.0       46.0          42.9
10-14                                  24.2     20.9      27.6     34.1      41.1          40.5
15-24                                  24.1      19.7     23.6     26.4      41.8           36.6
25-34                                  23.5      21.7     26.2     27.6       43.3          41.6
35-44                                   18.8     17.1     21.3     28.6       38.2          34.7
45-54                                   20.2     13.7     16.0     24.1       35.2          29.7
55-64                                  27.6      15.6     14.5     31.6       42.6          41.7
65-74                                   35.0     23.6     18.3     37.0       47.6          45.0
75 & Over                              47.5      37.7     22.1     47.9       41.4          45.9
All                                    26.1      20.6     23.0     30.5       42.5          39.4
Poverty Gap (in percent) j/
0-4                                    21.2      14.8     14.2     20.4       24.9          29.6
5-9                                     18.7     13.6     12.9     17.6       24.8          27.5
10-14                                  19.5      13.2     12.4     18.6      23.6           25.9
15-24                                  18.7      14.1     12.8     18.5      24.1           28.4
25-34                                  20.1      13.6     13.5     20.3       25.8          28.0
35-44                                   17.7     13.5     12.6     18.7       24.1          27.6
45-54                                   17.0     12.9     13.6     20.8       24.8          30.0
55-64                                   19.2     13.9     14.5     21.4       26.1          32.5
65-74                                  20.6      14.4     14.2     22.0       28.5          35.2
75 & Over                              26.1      17.4     15.4     26.0       34.2          37.7
All                                     19.8     14.1     13.3     20.2       25.0          29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
Notes:
1/ The poverty gap is the poor's average shortfall in expenditures from the poverty line, expressed as a
percentage of the poverty line (this measure is also known as the expenditure gap ratio).
38



Poverty incidence in Eastern Europe decreases with age, and reaches a minimum at
ages 35-44 in Bulgaria, ages 45-54 in Hungary, and ages 55-64 in Poland. After those
ages, the increase in poverty incidence is quite rapid and severe, except in Poland (as we
noted earlier, this is due to the generous pension system in Poland). In Bulgaria and
Hungary, poverty rates among people over 75 are close to twice the national average.
The vast majority of people in that age group are women, and their poverty rates are
higher than for the men in that group. Actually, the gender-breakdown of Table 8 (not
shown here) reveals that poverty rates among the elderly are higher for women in general
than for men. At lower ages though, the gender gap is not very pronounced, and for some
ages poverty is lower among women than men.
Gender is thus a relevant poverty dimension in Eastern Europe primarily for the
elderly, especially at very high ages, and for female-headed households. For many women,
the labor market changes of transition have had major implications. Prior to transition,
women were expected to work full-time but the state provided day care for their children.
Transition has led to a drop in female labor force participation (not all of it voluntarily) but
it has also led to a reduced supply of affordable day care centers (World Bank, 1996e)."
Both factors may well affect female-headed households disproportionately.
The effect oftransition on women is discussed further by Einhorn (1993), Funk and Mueller (1993), Chase (1995)
and Fong (1996).
39



Nevertheless, the poverty figures suggest that in general the age-effect outweighs
the gender-effect. This is clear also from Table 9 which classifies households by the
number of elderly people (over 65) in the household.  In Bulgaria and Hungary,
households without elderly members have below average poverty rates and those with
elderly members have above average poverty incidence. The latter increases with the
number of elderly. Poland is again the exception, where age proves to be an irrelevant
dimension of poverty. As discussed previously, Poland's pension system needs to be
credited with this result.
Table 9: Poverty and the Elderly
Number of ElderlyMembers    Bulgaria Hungary  Poland  Estonia    Kyrgyz        Russia
(Over Age 65) in the Household                                     Republic
Headcount (Po, in percent)
0-                                21.9     19.5    22.8    28.5         42.1        37.3
1                                 33.9    23.7    24.1    39.1          45.0        47.4
2                                 38.2    27.6    22.3    36.4          39.4        42.4
3 or more                          ...    56.8    24.7        ...         ...       50.0
All                               26.1     20.6    23.0    30.5         42.5        39.4
Poverty Gap (in percent) j/
0                                 19.1     13.9    13.3      19.6       25.0        28.8
1                                 22.2     15.2    13.5    23.6         24.3        32.7
2                                 18.5    13.7    13.2    17.5          29.2        32.1
3 or more                          ...      7.4     13.9      ...        ...        58.4
All                               19.8    14.1      13.3    20.2        25.0        29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
Notes:
V The poverty gap is the poor's average shortfall in expenditures from the poverty line, expressed as a
percentage of the poverty line (this measure is also known as the expenditure gap ratio).
40



The poverty gap shows little variation by age, although it is above average among
people aged over 65. It does not however increase systematically with the number of
elderly in a household. In fact, in Hungary, it is the reverse-the poverty gap falls
significantly in households with two or three elderly members. Many such households are
poor, but they are not very far below the poverty line.
a&    Former Soviet Union\
Poverty is generally considered to have sharply increased in countries undergoing
transition, partly because incomes are perceived to have become extremely unequally
distributed, and mostly as a result of drastic declines in GDP. Indeed, a static comparison
of the poverty rates for the FSU suggests that poverty is a serious problem in Russia and
Estonia, and a nearly overwhelming one in the Kyrgyz Republic.
After five years of economic contraction, the poor in the FSU appear to be
primarily the working poor, and especially the working poor with children. The working
poor are testimony to the adjustment in wages, rather than in open unemployment, which
has occurred.  The myth of the pensioner--he idea that pensioners are especially
vulnerable to poverty-is belied by several studies, including this one (World Bank,
various poverty assessments in transition economies), although it is true that the extremely
elderly (aged 75 and over) are more vulnerable to poverty. This sketch of the poverty
profile seems relatively robust across equivalence scales-most of the poverty assessments
cited used per capita measures, while in this study, the OECD equivalence scale is used.
41



Regardless of equivalence assumptions about economies of scale in consumption, family
composition appears to have a major influence on the household's poverty status.
L     Pre-Transition Poverty and Macroeconomic Impact of Transition
Poverty in the FSU was hidden and unacknowledged, but it was a fact of life for
approximately 6-10 percent of the population before the breakup of the country and the
embarkation on transitions to the market economy by the FSU countries (Braithwaite
1990, 1991, 1995). Five years after gaining or regaining independence, poverty has
become much more overt and has increased in scope. The large increases in measured
poverty are due to three major causes:  impact of severe macroeconomic declines
including hyperinflation; sharp increases in income inequality; and measurement error,
especially regarding the actual distribution of real consumption in the pre-transition
period.
Any comparison to the pre-transition period is fraught with methodological pitfalls.
Even the extent of macroeconomic declines is difficult to assess (Koen and Gavrilenko,
1994), although virtually no one would disagree that the FSU countries experienced
especially sharp contractions in output starting in 1992. The hyperinflations experienced
in 1993-94 by most FSU countries meant that the real value of wages, pensions, and other
cash transfers plummeted abruptly. The hyperinflations and the breakup of the ruble zone
led to macroeconomic disruption and a breakdown of the external trading relationships
among the FSU. Without reasonable trading regimes and owing to the necessity for a
complete realignment of production, real output declined precipitously.
42



At the same time that the size of the pie shrunk, its distribution became markedly
more unequal. The only statistics available are based on income and on the family budget
surveys, which are characterized by marked methodological shortcomings. Nonetheless,
even a casual comparison suggests that the extent of open income inequality has become
quite large during the transition (Milanovic, Forthcoming; Klugman and Braithwaite,
1998; Koen 1996, Commander, Tolstopiatenko, and Yemstov 1997).
A further complication stems from the prevalence of arrears in wages and pensions
in the FSU countries, and the irregular nature of even formal state-sector employment due
to forced administrative leave without pay and reduced working hours. Since the breakup
of the FSU, wage and cash transfer arrears have become common as countries have
grappled with the introduction of stabilization programs and fiscal austerity. Budgetary
sequestration in Russia, Armenia, and Kyrgyz Republic resulted in long delays for wages
in the "budgetary sphere" (health, education, government/administration, the Armed
Forces, law enforcement, etc.) and for cash transfers, including pensions and child
allowances.
It is difficult to accurately measure "official" or "registered" income given the
prevalence of arrears, but it is practically impossible to quantify the informal sector in
transition economies. In the FSU countries, households have been extremely reluctant to
report most of their income, even to survey researchers, much less to the tax authorities.
(Koen and Gavrilenko, 1994). Most studies suggest that the informal sector is around 40
percent of official GDP (KaufTmann and Kaliberda, 1996). In this study, total reported
household income as a percent of total household expenditures varied from less than 50
43



percent in the Kyrgyz Republic to more than 95 percent in Estonia. However, the
difference between reported income and household expenditures is large enough to mean
that conclusions about the distribution of consumption inferred from the distribution of
income are problematic. This is one of the reasons why the poverty profile in this study is
based on household expenditures.
It is also futile to compare exactly the distribution of consumption before and after
transition. During the pre-transition period, wages and prices were controlled and food
and other consumer goods were allocated by queuing, rationing, and favoritism. Neither
money income nor money expenditure reflected adequately the household's real
consumption, since much of that real consumption was allocated to the household through
non-market, non-money means.- For example, senior workers at larger enterprises
received better housing than junior workers at smaller enterprises, and they had shorter
queues for purchasing automobiles, etc. Without ever considering their money income, it
is clear that senior workers were better off. Unfortunately, there is no reliable way to
reconstruct the real consumption of the poor and non-poor prior to transition, due to the
absence of reliable, non-biased household data sets.
IEL    Who are the Poor After Five Years of Transition?
The working poor predominate in the poverty profile for the FSU countries. By
and large, in Estonia, Russia, and the Kyrgyz Republic, the head of household of poor
families is employed, most often in the state sector. Results indicate that poverty rates in
rural areas are much higher than in urban areas.
44



Location
In most countries, where a family lives has a significant correlation with poverty.
In the FSU and particularly Russia, there is a strong regional component to poverty,
relating to the legacy of the planned economy. There are many one-company towns in the
FSU which produced military-industrial goods for which demand has either disappeared or
sharply declined. Russia has its "rust belt" where textile production has been displaced by
competition from cheaper imports and service or other industry has not developed to fill in
this gap. Additionally, the quality, irrigation, and altitude of agricultural land varies, which
means that rural poverty is not homogenous.
In all the countries compared in this paper, urban poverty is markedly lower than
rural poverty, and the poverty rate in the capital city is the lowest (except in Hungary,
Table 1). For the East European countries, this is a conventional finding. In many
countries of the world, the rural poverty rate is higher than the urban poverty rate, and the
higher living standards of urban regions were portrayed as the major explanation for rural-
urban migration many years ago (Harris and Todaro, 1970). However conventional this
finding may be for Eastern Europe, it is not standard for the FSU. In almost every World
Bank poverty assessment completed to date for FSU countries, rural poverty has been
found to be somewhat less or markedly less severe than urban poverty both in terms of the
headcount and in terms of various measures of severity. 12
12 World Bank, 1995 a-c, 1996 a-d.
45



The HEIDE data set may lead to different conclusions for two reasons. First,
conditions in the transition countries may have changed from the time period referenced
by the Poverty Assessments and that covered by the HEIDE data base. Second, as
described in the methodological section, the HEIDE data base uses an equivalent adult
approach (which was not usually followed in Poverty Assessments), a relative poverty
line, and relied on self-reporting for the value of food produced by the household for its
own consumption. In most of the World Bank's poverty assessments for the FSU, the
value of food produced on private plots was imputed, usually based on the purchase prices
reported by all the households in the sample. Imputing the value of food this way tends to
lead to a higher consumption aggregate than asking respondents to assess the market
value of their food production.
In Estonia, there is little difference between the depth of poverty in Talinn, the
capital, in other cities, or in rural areas as measured by the poverty gap. In Kyrgyz
Republic, the average poverty gap seems to be most affected by the higher poverty gap in
other cities, since there is little difference between the poverty rate in the capital and the
countryside. In Russia, the poverty gap is highest in rural areas and noticeably lower in
the capital.
Family Composition
It is a truism of poverty studies that family composition is one of the most
significant correlates of poverty, since the number of earners and dependents has a critical
impact on the family's consumption needs and ability to fulfill those needs. In the FSU,
however, family composition does not correlate as strongly with poverty as it does in
46



Eastern Europe or in many other countries of the world (Tables 2 and 3). For example,
consider the issue of children. In this study, children were defined to be under the age of
15, which corresponds to the statistical definition of labor activity previously used in the
FSU, where those aged 0-14 were assumed to be outside of the "available labor
resources" of the country.13
In most poor countries, families with children are worse off than families without
children, and families with more children are worse off than families with few children.
The first part of this generalization seems to apply only weakly to the FSU countries in
termns of poverty rates, while the latter part seems to be clearly demonstrated in Estonia
and Russia, but only to a lesser extent in Kyrgyz Republic. Considering aggregated family
composition, in all three countries, families without children and without elderly members
are less likely (Kyrgyz Republic) or much less likely (Estonia, Russia) to be poor than
families with either. In Russia and Estonia, the highest poverty rates were experienced by
families with both children and elderly members (52 and 35 percent, respectively). The
Kyrgyz Republic showed the lowest degree of variation of poverty rates according to
family composition.
These findings are influenced by the overall demographic characteristics of the
populations compared. The Russian and Estonian populations are much more aged than
13 This definition of "children" as under the age of 15 is arbitrary, as are all definitions that do not correspond to the
age of legal majority, which in most of the countries in this study was 18. However, children aged 15-18 have
significant economic potential, and in the FSU, could usually drop out of school around the age of 14 during the
period wben the surveys were conducted.
47



Kyrgyz Republic, and the birthrate in Estonia is even lower than that in Russia, which is
itself very much lower than the birthrate in Kyrgyz Republic. The relative youth of
Kyrgyz Republic population and the widespread prevalence of children means that very
few Kyrgyz Republic households are without at least one young dependent (14.4 percent),
while nearly half of families in Estonia (49.5) and Russia (44.5) do not have a child.
In Estonia, families with three adults and three children, single-parent families with
one or more children, and single female adults have the highest poverty rates, but poverty
is most severe for single persons living alone. In Estonia, the dependency burden is more
associated with care of the-elderly than for children. Families with children comprised
only 53 percent of the poor, and families with two adults and any number of children had a
lower poverty rate than average.- In contrast, families with one or two elderly members
were poorer than average. About 18 percent of the poor are aged 65 or above, and given
the differential male-female survival, approximately three-quarters of the elderly poor are
female (Table 8).
In Russia, out of all poor individuals, approximately 60 percent live in families with
children, while slightly more than 40 percent live in childless homes. However, the
poverty rates and gaps are higher for single person households than in single-parent
households and significantly higher than two-adult households with one or two children.
This is an unusual finding, and the fact that one of the highest poverty rates recorded was
for a single male adult family type is even more surprising. One would expect that a single
male adult would have no dependents and presumably would have a reasonable earnings
potential. Age only partly explains this finding, since two-thirds of these single male
48



adults are younger than 65. For single Russian females, the poverty rate was high but so
too was the share (60 percent) aged 65 or older.
Other findings for Russia are more conventional. Families with three or more
children have the highest poverty rate in the sample (but not the highest poverty gap).
Families without an elderly member (aged 65 or above) have a much lower rate than
families with one or two elderly members. In Russia, more of the poor have children than
are responsible for an elderly member. However, those families with elderly members
have a higher poverty gap than families with children.
In Kyrgyz Republic, a single female living alone was the household/family with the
highest poverty rate and gap, followed by a many-child household with only two adult
members. This is probably related to the situation of the elderly who do not comprise as
high as share of the poor or total population, due to the younger age-structure of the
population. Additionally, in Kyrgyz Republic, having a child is fairly universal-only 14
percent of all families or of poor families do not have at least one child.
Even though most families with two or three children are not poor, there are so
many of such families in Kyrgyz Republic that they constitute the clear majority of the
poor-families with three or more children comprise 53 percent of the poor, while families
with two or more children are 72 percent of the poor. Such families are less likely to be
severely poor, though, as their poverty gaps are lower than those of single females or
males, or interestingly enough, of two adults without children.
49



Labor Force Participation
Aside from the truism that the more income-earners in a household, the better off
the household is, there are some unexpected differences about the relationship between
labor force participation and poverty in the FSU transition economies relative to other
countries. The first difference is that in the FSU, participating in the labor force does not
always mean that the participant is paid anything at all. Due to the pervasive wage and
cash transfer arrears (notably for pensions, but also for child allowances), the notion
"working poor" takes on a. whole different meaning. Indeed, there are many who are
working poor but would not be poor if their salaries were paid, and there are pensioners
that would not be poor if their transfer payments were received on time.
In 1997, Russia announced a commitment to clear pension arrears by the end of
the year. Previously, and during the survey period studied here, pension arrears were
averaging anywhere between three and nine months, with some more remote areas having
much longer lags in payment than in the well-off areas such as Moscow and St.
Petersburg. On the other end of the spectrum, Estonia initiated a pension reform in 1993
which reduced differentiated pensions, eliminated pension payment arrears, and provided
for gradual increases in the retirement age.
In addition to arrears, the phenomenon of retaining workers by forcing them to
work reduced hours (short-time) or to be on unpaid administrative leave (forced leave)
was widespread in Russia and Kyrgyz Republic.
50



A second difference between the transition economies of the FSU and other
developing economies is that the stigma of reporting that one is out of work is arguably
greater, while the entitlement attitude that one deserves a pension or allowance is perhaps
larger than in other country contexts. This is due to the legacy of the previous system, in
which labor was perceived as the right and obligation of anyone who was able-bodied, but
that those who had contributed previously to the labor market would be protected in old-
age or during periods of temporary "disability" (e.g. pregnancy or illness). These attitudes
are likely to evoke positive answers to survey questions such as "do you work?" or "do
you have a job?" in conditions which might receive negative answers in other country
contexts. Nonetheless, unemployment rates calculated from the HEIDE data are higher
than both registered unemployment in the FSU and the rate of unemployment benefit
receipts reported by HEIDE respondents, reflecting the extreme difficulty of qualifying for
an unemployment benefit and its short duration, especially in Russia and Kyrgyz Republic.
A third particular aspect of the labor market in the transition economies of the
FSU is that it is extremely in flux, as the private sector emerges, and the informal labor
market offers as many opportunities at the top end of the scale as it does at the bottom.
Given the historic legacy in Russia and the other FSU countries, where entrepreneurial
behavior has always been regarded with extreme distrust, it is truly difficult to determine
the extent of entrepreneurial business and earnings. As a result, it is quite possible that the
number of respondents who report that they work in the private sector might be
understated. It mnight be preferable for a person who has a state sector job "in name only"
to maintain that legal affiliation while in essence running a full-time business on the side.
51



With these caveats in mind, the conclusions about poverty and labor market
participation for the FSU countries are rather conventional (Tables 4 and 5). Households
with employed heads have lower poverty rates than those with unemployed heads, while
the addition of one or two household members who are unemployed sharply increases the
poverty rate. In Estonia, a third unemployed member increases the poverty rate further,
but this is not the case in Russia and in Kyrgyz Republic, perhaps for some of the reasons
detailed above. However, all three countries demonstrated an increasing poverty gap with
each additional unemployed household member.
Some occupations or-socio-economic groups are associated with a lower poverty
rate than others. In all three countries, self-employed household heads live in households
with-a lower poverty rate than average, as do heads who describe themselves as wage-
earners. However, those self-employed heads who are poor are poorer than average in
Estonia. In all three countries, wage-earners live in households with a lower poverty gap.
Households with pensioners and other transfer recipients as household heads have sharply
higher poverty rates than average, but their average shortfall in expenditures is not
markedly different from average.
Households with access to land-a private plot-had lower poverty rates than
households without land, and the poverty gap was smaller, in Estonia and Kyrgyz
Republic (Table 10). Unlike other studies of Russia (Klugman 1997, World Bank 1995b,
Kolev 1996), this study found that Russian households with a plot were poorer than
households without. As noted above, the change in methodology from imputing the value
52



of private plot produce to relying on self-valuation may explain partially why this result
was obtained.
Table 10: Poverty and Ownership of a Private Plot
Whether the household has a   Bulgaria Hungary  Poland  Estonia    Kyrgyz     Russia
private plot                                               Republic
No plot                              26.6       .    18.5    34.4       48.2        38.1
Has plot                            25.6       ...   26.5    27.7       39.1        43.5
All                                 26.1    20.6    23.0    30.5        42.5        39.4
Poverty Gap (in percent)_1/
No plot                              21.6      ...    13.6    21.4      28.1        29.1
Has plot                             17.2      ...    13.2    19.2      22.2        31.8
All                                  19.8    14.1    13.3    20.2       25.0        29.8
Source: Household Expenditure and Income Date for Transition Economies Data Set (HEIDE).
Motes:
I/ The povertv gap is the poor's average shortfall in expenditures from the poverty line, expressed as a
percentage of the poverty line (this measure is also known as the expenditure gap ratio).
Education of the household head has a strong influence on the household's poverty
status, with the lowest headcounts achieved by those with university education (Table 6).
In all three countries, households headed by those with primary education or less than
complete primary education had poverty rates higher than average, and higher than those
with secondary education. The depth of poverty paralleled the poverty rates, with the
highest poverty gaps for those with primary or less education, followed by secondary and
vocational-technical education. The poverty gaps for those poor with higher education
were below average.
53



Gender and Children
Unlike most of the World Bank's poverty assessments, this study suggests that
gender is a significant dimension of poverty (Table 7) in the FSU. In the three FSU
countries, the poverty rate was sharply higher in households headed by women as opposed
to men, with this difference being largest in Estonia and smallest in Kyrgyz Republic.
Additionally, female-headed households were poorer than comparable male-headed
households as their poverty gaps were larger, although this difference was not as
pronounced as the difference in poverty rates.
Given the differential survival rates of women and men, for all three countries,
women comprise an increasing share of those who are poor as age increases. Half or even
less than half of poor children are girls, but by age 65 and over, women are approximately
70-75 percent of the poor. The elderly aged 65 and above are also poorer than average,
as measured by the poverty gap. This means that elderly female poverty is more pervasive
as well as deeper than male poverty in the Former Soviet Union.
4.    Multivariate Analysis of Welfare and Poverty
The goal of the multivariate analysis of welfare and poverty is to assess the relative
importance of various correlates of poverty and if possible, to attribute causality to these
correlates. Additionally, determinants of welfare such as the demographic characteristics
of households and the return to household assets, may differ between the poor and the
non-poor, and the multivariate analysis will help to elucidate these differences. As was the
case with the two-dimensional examination of poverty in the poverty cross-tabulation
54



tables, by and large, these goals were better met by results for the East European
countries. Almost all the variables included in the models for the East European countries
have estimated parameters significantly different from zero, and the pattern of results is
very consistent across the three East European countries. Thus, while there remain
important unidentified welfare determinants (e.g. personal ability), the model does point at
a set of significant factors which affect welfare outcomes and which can be identified and
affected in the context of policy intervention to alleviate poverty in Eastern Europe.
Unfortunately, the results are not so clear for the FSU. Overall, the explanatory
power of the welfare regressions is low, and it is difficult to find as many clear poverty
correlates as for the East European countries. In several ways, it can be argued that such
a finding is not surprising, and relates to the different degrees of the transition process. In
particular, the FSU still lacked much open unemployment during the HEIDE survey
periods, although there was a strong correlation between actual unemployment and
poverty. Further, in the FSU the labor market and especially the private sector are not
well-defined, and it is difficult to determine a priori who are likely to be the losers and
winners, since many of the traits associated with winning in the new regimes
(entrepreneurial skills, political connections) are extremely difficult to measure reliably by
a household survey. However, those aspects which could be measured in the household
surveys (access to a household business or private plot) were strongly associated with the
ability of a household to avoid poverty in the FSU.
55



A.    Eastern Europe
Welfare Equations
Several general observations emerge from  the estimation results in Table 11.
While moderate, the overall goodness-of-fit is in line with typical results for this type of
equation (l2 ranges from 0.267 to 0.301). The reported results are for the log-linear
functional form. These were compared against the results from the linear form, using the
test developed fof that purpose by Davidson and MacKinnon (1981). In each case the test
results pointed at the superiority of the log-linear specification. This implies that effects of
household characteristics on welfare are proportional rather than linear. For example, the
effect of education is to increase expenditure per equivalent adult in a fixed proportion,
rather than with a fixed amount (i.e. the absolute returns are lower for the poor).
56



Table 11: Welfare Equations (OLS): East European Countries
Dependent Variable = In (household expenditure per equivalent adult)
Bulgaria                   Hungary                     Poland
Parameter     Standard     Parameter     Standard     Parameter    Standard
Estimate       Error       Estimate       Error       Estimate       Error
Intercept                                    8.558*        0.122          9.424*     0.042           7.556*     0.036
Number of children                           -0.064*       0.013         -0.075*     0.005          -0.091*      0.003
Number of male adults                       -0.065*        0.015         -0.012       0.007         -0.022*      0.005
Number of female adults                     -0.039*        0.014         -0.011      0.007          -0.017*      0.005
Number of elderly                           -0.084*        0.020         -0.039*      0.010         -0.025*      0.007
Educationofhead: primary                     -0.235*       0.024         -0.228*     0.011          -0.195*      0.008
Education of head: vocational/technical      0.066*        0.037         -0.109*     0.011          -0.098*      0.008
Educationofhead: university                  0.117*        0.029          0.135*      0.014          0.207*      0.011
Age of head                                   0.021*       0.005          0.017*      0.002          0.011*      0.001
Age of head squared                          -0.000*       0.000         -0.000*      0.000         -0.000*      0.000
Female head                                 -0.117*        0.025         -0.063*      0.009         -0.058*      0.007
Householdownsenterprise                       0.321*       0.043          0.163*      0.015          0.229*      0.014
Household owns land                           0.178*       0.022         -           -               0.033*      0.007
Household is renter                          -0.326*       0.038         -0.173*      0.010         -0.047*      0.007
Shareofwagesinhouseholdincome                0.318*        0.036          0.172*      0.018          0.141*      0.012
Number of unemployed in household           -             -              -0.102*      0.009         -0.120*      0.007
Head is unemployed                                                       -           -              -           -
Head is inactive                            -             -              -0.035*      0.017         -0.007       0.011
Location: non-capital city                   -0.140*       0.027          0.017       0.010         -0.095*      0.011
Location: village                           -0.246*        0.033         -0.007       0.011         -0.198*      0.013
Number of observations                    2465                         8104                      16,050
R2 (adjusted)                                 0.288                       0.301                      0.267
F-statistic                                 63.39*                      205.56*                    325.47*
Note: * Indicates that estimated paraneters are significantly different from zero at the 90 percent confidence level.
57



The general pattern of findings is that education and the ownership of a household
enterprise have the largest effects on welfare outcomes, followed by the nature of the
household's link with the labor market. Demographic characteristics are a distant third.
Some countries (Bulgaria) show strong location effects, while others (Hungary) show
almost none.
In Bulgaria, ownership of a household enterprise, owning one's home, and
deriving all household income from wages each imply increases of household welfare in
excess of 30 percent. No single variable has such high welfare premium attached to it in
Hungary or Poland. In Hungary, the strongest effect comes from primary education
(negative 23 percent relative to the reference category of secondary education).
Enterprise and home ownership, and a wages-only income each add 16-17 percent to
household welfare. In Poland, the strongest welfare determinant is also a home enterprise
(23 percent), but wage-income and home ownership have smaller effects (14 percent and
5 percent, respectively).
The results clearly indicate the key role played by education in transition
economies. In Bulgaria, households where the head did not achieve more than primary
education, have a welfare level 23.5 percent below that of the reference category (a
household where the head has secondary education). This welfare "penalty" for low
education is similar in the other two countries. Vocational and technical education is
associated with a small welfare gain in Bulgaria, relative to secondary education, but with
a welfare loss in Hungary and Poland. The likely explanation is that Bulgaria is not yet as
far advanced in its transition as the other two countries, and still has many unconverted



state industries where the pre-transition vocational and technical education continues to
have a high pay-off. The conversion process in the other countries has put a premium on
job flexibility, and the more general secondary education and especially university
education have proven to lend themselves better to the needed adaptation. This is
reflected in the higher coefficients for university education in Hungary and Poland relative
to Bulgaria.
These results underscore the crucial importance of general education (especially
post-secondary education) for a successful long-term strategy in coping with transition.
The huge gaps in the return to education between the primary and higher levels point at
the unequalizing effect on the distribution of household welfare which is likely to result
from transition. It is not practically possible to "upgrade" people's education in the short
run-particularly since almost 2/3 of heads of households in the East European countries
have primary  or vocational/technical education  levels.   These  households will
progressively fall behind as transition proceeds, unless they can be re-schooled or re-
trained. There is evidence that following transition the distribution of wages has become
more unequal in Eastern Europe (Milanovic, 1995), and our results indicate that this effect
from education extends to the overall distribution of household welfare as well.
The poverty profiles earlier in this paper, as well as other analyses of poverty in
Eastern Europe (Grootaert, 1995, 1997a) have highlighted the strong correlation between
poverty and open unemployment. Unemployment is perhaps the most visible aspect of the
social cost of transition and it has severe distributional implications. The results in
Table 11 indicate that over and above other household attributes (some of which, such as
59



iow education, increase in themselves the probability to be unemployed), the presence of
an unemployed household member reduces household welfare by 10-12 percent.'4  A
significant number of households in Eastern Europe have more than one unemployed
member.
As one can expect, household size is negatively related to household welfare (since
we defined the latter as household expenditure per equivalent adult). However, what is of
interest is the role of household composition, as reflected in the magnitude of the
coefficients for each type of household member. Except for Bulgaria, the strongest
negative coefficient is found for the number of children.   The implication is that
households do not succeed in maintaining their welfare levels when the number of children
increases-in spite of the generous social transfers in Eastern Europe and the presence of
general entitlement programs such as the family allowances which are targeted on
children. Given the positive correlation between household size and poverty, it may well
be needed that child-oriented transfer programs move away from being general
entitlements to being more poverty-targeted by paying larger amounts to poor families
with children.
The pattern of the coefficients of the other demographic variables is country
specific. With respect to age of the head of household, each of the three countries
indicates an inverted-U life-cycle pattern with welfare levels rising over most of the adult
4 One can assume that the effect is the most severe if the head of household is unemployed. Grootaert (1997a)
contains some evidence to that effect for Hungary. Since the household head was defined as the main earner in
the Poland and Hungary household surveys used for the HEIDE database, few heads of household are classified as
unemployed for those countries.
60



age-range, and then falling in the elderly years. The turning points are 49 years of age for
Bulgaria, 53 years for Hungary, and 68 years for Poland. This may suggest differential
effectiveness of the pension system  to maintain welfare levels.  We discussed the
generosity of the Polish pension system previously, but other factors likely play a role,
including private transfers and the ability (and willingness) of retired people to earn
secondary incomes.
In each of the three countries, female-headed households have a lower welfare
than male-headed households who are similar in all other characteristics. The shortfall
ranges from 5.8 percent in Poland to 11.7 percent in Bulgaria. There appears to be a
coincidence of demographic factors. In Poland, old age clearly matters least in terms of its
impact on welfare levels (the age turning point is highest, and the coefficient of "number
of elderly" in the household composition variables is lowest) and this is also the case for
gender effects. In contrast, Bulgaria has the strongest age and gender effects, two to three
times larger than those observed in Poland.
Lastly, we need to point at the country-specific location effects. In Hungary, the
large welfare differences across locations are fully explained by the distribution of
demographic and economic characteristics of households, and residual location effects are
not statistically different from zero. In Bulgaria and Poland, in contrast, large location
effects remain. Relative to the capital city, households living in other cities have a 10-14
percent lower welfare level, and those in villages are 20-25 percent lower, even after
controlling for all household characteristics included in the model. This suggests that
economic and social infrastructure, as well as other supply factors of economic activity,
61



have important locational inequalities. Indeed, it has been a characteristic of much of the
transition in Eastern Europe that certain regions, such as those with traditional heavy
industry, have suffered the most from transition due to the impossibility to convert such
industries to privately-owned competitive firms.  Similarly, the conversion of state-
controlled agriculture to private farns has not happened without loss of income to many
farmers (Milanovic, 1995).
Poverty Equations
As we discussed in Section 2, we are concerned about the effect of possible
measurement error of household expenditure which could be correlated with some of the
explanatory variables in the model (e.g. educated people report household expenditures
more accurately; older people have more difficulty with reporting; households with self-
employment income try to hide income and expenditure for fear of taxation). This could
bias the coefficients of the welfare equation estimated by OLS. There is also a concern
about the extent to which a given functional form fits the distribution. For both reasons,
we estimated poverty equations with a binary dependent variable (poor/non-poor) using
probit techniques. The consistency, or lack thereof, of probit results with the welfare
equation results serves as a test for the presence of measurement-error or functional-form-
fit problems.
The results in Table 12 suggest that the binary model provides a good fit. The
model correctly classifies 77 percent to 82 percent of households as poor or non-poor, and
as was the case with the OLS-model, almost all of the included variables have estimated
coefficients significantly different from zero at the 90 percent confidence level. Table 12
62



Table 12: Poverty Equations (Probit): East European Countries
Bulgaria                   Hungary                    Poland
Probability   Standard     Probability    Standard    Probability    Standard
Derivatives     Error      Derivatives     Error      Derivatives     Error
Number of children                           0.039*       0.013          0.056*      0.006          0.062*     0.003
Number of male adults                        0.052*       0.015          0.012       0.008          0.013*     0.004
Numberoffemaleadults                         0.018        0.015         -0.006       0.008          0.004      0.005
Number of elderly                            0.055*       0.020          0.020*      0.010          0.015*     0.007
Educationofhead: primary                     0.171*       0.024          0.188*      0.014          0.161*     0.010
Education of head: vocational/technical     -0.056        0.037          0.077*      0.016          0.076*     0.009
Education of head: university               -0.085*       0.029         -0.09l*      0.015         -0.094*     0.010
Age of head                                 -0.014*       0.004         -0.016*      0.002         -0.008      0.001
Age of head squared                          0.000*       0.000          0.000*      0.000          0.000*     0.000
Female head                                  0.113*       0.027          0.031*      0.011          0.027*     0.007
Household owns enterprise                   -0.169*       0.028         -0.078*      0.014         -0.102*     0.008
Household owns land                         -0.131*       0.021         -           -              -0.029*     0.008
Household is renter                          0.244*       0.047          0.172*      0.015          0.020*     0.007
Share of wages in household income          -0.272*       0.037         -O.136*      0.021         -0.122*     0.012
Number of unemployed in household                        -               0.075*      0.010          0.078*     0.007
Head is unemployed                                                      -           -              -          -
Head is inactive                            -            -              -0.001       0.018         -0.022*     0.010
Location: non-capital city                   0.029        0.029         -0.037*      0.012          0.018      0.014
Location: village                            0.128*       0.037         -0.002       0.013          0.092*     0.017
Log-likelihood                           -1199.3                     -3418.7                    -6677.6
Chi-squared                                503.95                     1381.3                     2291.7
Prob > chi-squared                           0.0000                      0.0000                     0.0000
percent correct predictions                 76.6                        81.22                     81.63
Note: * Indicates significance of the underlying coefficient at 90 percent level probability. Derivatives are taken at the mean values of continuous
variables or for discrete change of dummy variables from 0 to 1.
63



does not report the probit coefficients, but the probability derivatives at the mean of each
continuous explanatory variable and for a change from zero to one in the case of dummy
variables. The estimation used non-poor as the base category, hence the derivatives
pertain to the probability to be poor.
Substantively, the pattern of determinants of poverty is entirely consistent with the pattern
of determinants of welfare that was revealed by the welfare regression. All factors which
are correlated with an increase/decrease in welfare are correlated with a decrease/increase
in the probability to be poor. Hence, qualitatively the poverty regression adds nothing to
the findings from  the welfare regression.  However, there are some quantitative
differences, in terms of the relative magnitude of the effects. This is to be expected of
course, since the poverty regression uses different information than the welfare regression.
A case in point is the effect of education in Poland. In the welfare regression, university
education was associated with a 21 percent welfare premium relative to secondary
education, while primary education was associated with a welfare reduction of 20 percent,
i.e. the two levels of education had symmetrical welfare effects around the reference
category.  In contrast, primary education increases the probability to be poor by
16 percentage points relative to secondary education, but university education reduces it
by only 9 percentage points. In other words, university education is an important
determinant of where on the welfare distribution a household will end up, and it has large
absolute returns, but it has a lesser role as a determinant of poverty.
There are several similar patterns in the shift of relative roles of variables between
the welfare regression and the poverty regression. These shifts have implications for the
64



targeting and design of poverty alleviation interventions.  Foremost is the role of
household enterprises. In Bulgaria and Poland, ownership of a household enterprise
makes the largest or second largest positive contribution to household welfare-a clear
reflection of the post-transition emergence of the small-scale private sector.  These
enterprises do reduce the probability that the household is poor, but the estimated effects
are smaller than several other variables such as education or the share of wages. The
contribution of household enterprises is hence more important in the upper part of the
distribution, and one characteristic of the poor in transition economies is that they have
not yet successfully got involved in the private sector as entrepreneurs.
Important differences between the welfare and poverty regressions also occur in
the demographic and location variables. While both models underscore the correlation
between number of children and low welfare or poverty, they fail to do so for other
categories of household members. E.g. in Bulgaria and Poland, additional female adults in
the household are associated with lower household welfare, but this does not increase the
probability to be poor. In Bulgaria, the coefficient for male adults stands out as much
higher than in other countries, and could indicate the difficulties in that country for men in
their prime earning years to find adequately paid employment.
The results for location also deserve highlighting. The degree to which regional or
locational targeting of poverty interventions is desirable and useful is frequently a major
issue. The answer is different for each country but it is important to underline that the
answer given by the welfare regressions is not the same as that given by the poverty
regression. This is simply saying that the geographic distribution of welfare is not the
65



same as the geographic distribution of poverty. 15 Specifically, in Bulgaria and Poland, the
probability to be poor does not differ between the capital city and other urban areas (after
controlling for all other variables), even though the latter areas have significantly lower
welfare levels. In the rural areas of these two countries, however, there is a higher
probability to be poor, which is consistent with a negative welfare effect. Hungary is
unique in that, certeris paribus, the probability to be poor is less outside the capital city.
Poverty Gap Equations
Table 13 presents Tobit estimation results for the right-censored subsample of
poor households. As we discussed in Section 2, this model is conceptually equivalent to
estimating the determinants of the poverty gap (i.e. the depth of poverty).   The
coefficients reported in Table 13 are directly comparable to the OLS coefficients of the
welfare model which was estimated over the full sample (see Table 11). This comparison
provides a test whether constant parameters apply for the entire welfare distribution. It is
clear that for the majority of variables this hypothesis is rejected: returns to assets and
contributions to welfare from other household characteristics are not the same for the
poor and the non-poor. In most cases, the coefficients are higher for the poor than for the
full sample.
5 This result is clearly sensitive to where exactly the poverty line is set.
66



Table 13: Poverty Gap Equations (Tobit): East European Countries
Bulgaria                   Hungary                     Poland
Dependent Variable: In (Erpenditure per  Parameter     Standard     Parameter     Standard     Parameter     Standard
Equivalent Adult) Right-Censored at Poverty  Estimates    Error       Estimates      Error       Estimates      Error
Line
Intercept                                    8.559*        0.184          9.273*      0.060          7.417*      0.055
Number of children                          -0.085*        0.020         -0.078*      0.007         -0.093*      0.005
Number of male adults                       -0.098*        0.023         -0.009       0.011         -0.027*      0.007
Number of female adults                      -0.056*       0.023          0.003       0.011         -0.004       0.007
Number of elderly                           -0.077*        0.031         -0.018       0.014         -0.02l*      0.011
Education of head: primary                  -0.285*        0.039         -0.252*      0.018         -0.225*      0.014
Education of head: vocational/technical      0.142*        0.071         -0.095*      0.020         -0.106*      0.013
Educationofhead: university                  0.171*        0.057          0.164*      0.032          0.207*      0.028
Age of head                                  0.024*        0.007          0.022*      0.002          0.012*      0.002
Age of head squared                          -0.000*       0.000         -0.000*      0.000         -0.000*      0.000
Female head                                  -0.173*       0.038         -0.038*      0.014         -0.042*      0.010
Household owns enterprise                     0.355*       0.090          0.132*      0.028          0.236*      0.026
Household owns land                          0.280*        0.035         -           -               0.055*      0.013
Household is renter                          -0.346*       0.055         -0.207*      0.015         -0.025*      0.011
Share of wages in household income           0.528*        0.060          0.199*      0.029          0.220*      0.018
Number of unemployed in household           -             -              -0.115*      0.012         -0.114*      0.010
Head is unemployed                                                       -           -              -           -
Head is inactive                            -             -               0,007       0.024          0.045*      0.016
Location: non-capital city                   -0.035        0.046          0.055*      0.016         -0.031       0.022
Location: village                            -0.231*       0.053         -0.005       0.017         -0.142*      0.024
Log-likelihood                            -1178.5                     -2580.1                    -5345.9
Chi-squared                                595.31                      1501.56                    2368.6
Prob > chi-squared                            0.0000                      0.0000                     0.0000
SER                                           0.527                       0.351                      0.390
Note: * Indicates that coefficient is significantly different from zero at 90 percent confidence level.
67



With respect to education, the main difference is that the welfare penalty
associated with having completed only primary education, or conversely, the welfare
benefit of secondary over primary education, is larger for the poor than for the population
at large.  It ranges from 22.5 percent in Poland to 28.5 percent in Bulgaria (the
corresponding figures for the full sample were 19.5 percent and 23.5 percent). In
Bulgaria, the welfare gain for the poor from vocational/technical education is 14.2
percent, against only 6.6 percent for the whole population. In Hungary and Poland,
vocational and technical education lead to lower welfare levels relative to secondary
education, and here the differences between the poor and the non-poor are small. The
returns to university education are not different for the poor and non-poor in Poland, but
in the other two countries the returns are much higher for the poor. All this suggests that
re-schooling and re-training could have potentially high pay-offs in the context of a
poverty alleviation program.
The returns to land ownership are also higher for the poor, especially in Bulgaria.
The returns from ownership of a household enterprise-already the single most important
welfare determinant for the population at large-are higher still for the poor in Bulgaria
-and Poland (but lower in Hungary). Clearly, the ability to participate successfully in the
informal private sector is the key factor to reduce the depth of poverty in the East
European countries considered here, and programs to promote private entrepreneurship
are probably the most important ingredient in active labor market policies, from the point
of view of poverty reduction. Of course, the results also suggest that obtaining a wage job
is an equally or even more successful road towards reducing the poverty gap-the wage-
68



share variable has much higher coefficients for the poor than for the entire population. In
Bulgaria, each increase in the share of wages in total income of 10 percentage points is
linked with a rise in household expenditure per equivalent adult of 5.3 percent. In
Hungary and Poland, the corresponding figures are 2.0 percent and 2.2 percent.
As far as demographic characteristics are concerned, the lower welfare
experienced by female-headed households amounts to 17.3 percent for poor households in
Bulgaria, against 11.7 percent for the population at large. In Hungary and Poland though,
the welfare gap between female-headed and male-headed households is less for the poor
than for the population.  The welfare burden stemming from  large households is
significantly greater for the poor than for the non-poor in Bulgaria, but the evidence is
mixed in other countries. Lastly, the location effects are smaller for the poor in Bulgaria
and Poland, but larger in Hungary. This could reflect a greater effectiveness of the
Hungarian social safety net in reaching the poor living outside the capital city.
In summary, the general finding from the poverty gap equations, in comparison
with the welfare equations estimated over the full population, is that returns to human and
physical capital are often higher for the poor than for the non-poor and that the promotion
of access to such capital and upgrading of capital owned by households are sensible
components of poverty reduction strategies in Eastern Europe.  The role of other
variables, especially household composition and location, also differs between the poor
and non-poor, but the pattern of differences is country-specific.
On the methodological front, these results call for a certain amount of caution in
using welfare regressions estimated over the full sample as a basis for poverty analysis.
69



Our results suggest that one of the basic assumptions of this practice, the constancy of
parameters over the entire distribution, may not hold for a number of key variables,
especially household assets.
B.    Former Soviet Union
Welfare Equations
For consistency and also owing to a lack of specifications which perfonned better,
the same specification for the OLS model used for Eastern Europe was used for the FSU.
In general, the overall goodness-of-fit for the FSU countries is much lower than for the
Eastern European countries, as shown by the R2 measures reported in Table 14. The R2
for Estonia is the only one close to the lower boundary for the East European countries,
while the explanatory power of the equation is quite low for both Russia and Kyrgyz
Republic. Most but not all of the determinants of welfare included in the specification had
estimated parameters significantly different from zero, but there was no discernible pattern
to these differences. For all the FSU countries, the number of children, female-headed
household, household ownership of an enterprise, the share of wages in household income,
university education of the head, and the location dummy variables were significant in
determining expenditures per equivalent adult. These significant factors are in most cases,
easy to measure, and can serve as the basis for policy interventions.
70



Table 14: Welfare Equations (OLS): FSU Countries
Dependent Variable = In (household expenditure per equivalent adult)
Estonia                Kyrgyz Republic               Russia
Parameter     Standard     Parameter     Standard    Parameter    Standard
Estimate       Error       Estimate      Error       Estimate       Error
Intercept                                   7.301        0.089          9.627*      0.191          1.068*     0.120
Number of children                          -0.080*      0.012         -0.060*      0.013         -0.123*     0.015
Number of male adults                       -0.057*      0.021         -0.029       0.023         -0.030      0.022
Number of female adults                     0.011        0.017          0.025      -0.022          0.019      0.020
Number of elderly                          -0.047        0.022         -0.061       0.041         -0.021      0.025
Education of head: primary                 -0.076*       0.023          0.045       0.060         -0.084*     0.036
Education of head: vocational/technical      ...          ...           0.102*      0.059          0.034      0.030
Education of head: university               0.214*       0.028          0.158*      0.060          0.147*     0.037
Age of head                                 -0.003        0.004         0.017*      0.009          0.010*     0.004
Age of head squared                         0.000         0.000        -0.000*      0.000         -0.000      0.000
Female head                                -0.103*        0.028        -0.191*      0.056         -0.141*     0.032
Household owns enterprise                    0.174*       0.022         0.160*      0.042          0.235*     0.041
Household owns land                         0.160*        0.021         0.249*      0.040         -0.043      0.029
Household is renter                         -0.122*       0.019         0.064       0.070          0.024      0.026
Share of wages in household income          0.329*        0.032         0.205*      0.066          0.297*     0.040
Number of unemployed in household           -0.189*       0.026          ...        ...           -0.240*     0.038
Head is unemployed   .                        ...         ...          -0.148*      0.071           ...
Head is inactive                            -0.119*       0.026         -0.056      0.061         -0.149*     0.037
Location: non-capital city                  -0.196*       0.022         -0.338*     0.059         -0.365*     0.037
Location: village                           -0.311*       0.028         -0.504*     0.059         -0.428*     0.043
Number of observations                    2817                       1929                       5147
R2 (adjusted)                                0.256                      0.103                      0.115
F-statistic                                57.95*                       13.26*                    38.08*
Note: * Indicates that estimated parameters are significantly different from zero at the 90% confidence level.
71



As was the case for Eastern Europe, testing the log-linear specification against the
linear form (Davidson and MacKinnon, 1981) demonstrated that the log-linear form was
preferred. This means, for example, that the effect of adding an additional child is to
decrease household welfare (expenditure per equivalent adult) in a fixed proportion,
implying that the absolute costs of adding a child are lower for the poor.
In the FSU, locational factors have the strongest effect on household welfare,
followed by the share of wages in household income, whether the household has a
household enterprise, and higher education. These general findings are discussed in detail
below.
In the FSU, the strongest effects on welfare were related to household location,
with the sharpest change in household welfare (increasing it by 50 percent) implied by
moving from a rural area to the capital in Kyrgyz Republic. Even moving from an urban
area to the capital, Bishkek, would increase household welfare by one-third. The location
effects are nearly as strong in Russia (43 percent rural-to-capital, 37 percent urban-to-
capital) and not inconsiderable in Estonia (with rural-to-capital shifts increasing welfare by
nearly one-third).
The dominant role of location, especially location in the capital city, has been
documented in other FSU countries such as Armenia and Ukraine (World Bank 1996a,
1995c). In many ways, this finding demonstrates the slowness of transition and of
business-encouraging reforms and private sector development, as well as questions of
scale in many small FSU countries. Aside from Russia (and possibly Ukraine) most FSU
countries are quite small in terms of population and GDP, so most private sector
72



development has been concentrated in the capital cities (which are often the only cities of
any appreciable size).
The next most significant factors for increasing household welfare in the FSU were
the household's entrepreneurial activity, either through owning an enterprise or farming a
private plot of land, and the household's link to the labor market, as proxied by the share
of wages in total household income. However, the ranking of these factors was country-
specific. In Estonia, land ownership and owning an enterprise was significant but most
important was the share of wages in household income, which can increase welfare by
one-third. In Russia, land ownership was not significant,'6 but the share of wages in
household income and ownership of household businesses were important, raising
household welfare 30 and 24 percent respectively. In Kyrgyz Republic, after location,
ownership of a private plot had the largest effect on welfare, increasing it by 25 percent,
while-household welfare increased 20 percent from increasing participation in the official
economy (as captured by an increase in the share of wages in total household income).
In all three countries, the presence of an unemployed household head (or
household member) was found to significantly decrease household welfare. In Estonia and
Russia, specifications based on the number of unemployed demonstrated that adding an
unemployed household member reduced household welfare per equivalent adult between
16 Ths is possibly due to recordinWmeasurement errors for this variable which is carried in the data set as hectares
of land held by the household. When the raw data were examined, there were several improbable outliers.
Unfortunately, removing these outliers did not significantly improve perfornance, nor did the subsitutation of a
dummy variable for land ownership. The dummy variable did perform better in Russia than the number of
hectares, so it was retained. Sonewhat similar measurement problems also plagued the Kyrgyz Republic data
(the two surveys were conducted by the same consulting group), but the dununy variable was significant for
Kyrgyz Republic. In Estomia, a different survey and methodology recorded only whether the household had access
to land, not the amount of hectares.
73



15 and 20 percent, respectively. In Kyrgyz Republic, the dummy variable for household
head performed better (partly because of the lower number of "unemployed" in Kyrgyz
Republic, where many family members work on the same private plot and are thus
automatically not counted as unemployed) and resulted in reductions of welfare in the
order of those in Russia. Additionally, in Russia and in Estonia, households headed by
individuals not in the labor force (inactive heads) were associated with declines in
household welfare of 15 and 12 percent, respectively.
The final significant factor was whether the household head had university
education. In all three countries, welfare gains were approximately 15-20 percent. In
Estonia and Russia, primary education of the head was also significant and in the expected
(negative) direction, reducing household welfare about 8 percent in both cases. In Kyrgyz
Republic,  vocational-technical  training   was   also   associated   with  improved
welfare-having a household head with it would raise household welfare 10 percent
relative to a household head with secondary education. In Russia, however, vocational-
technical did not have a significantly different return from secondary education. 7
Demographic factors, except for the number of children and female headship, were
generally not very important for household welfare in the three FSU countries. In each
case, adding children meant reducing household welfare, from a low of a 6 percent
reduction in Kyrgyz Republic to a 12 percent reduction in Russia. In Russia, there was a
17 For Estonia, data on household heads with vocational-technical education were combined with households with
general secondary education. Although the base data set for Estoma did have a very few individual household
heads with vocational-education, they were few in number and their welfare level improbably high. It is possible
that these households should have been classified elsewhere, but it was not possible to discern the exact
definition differences that led to the non-comparability.
74



system of generalized child allowances but there were significant payment arrears and the
take-up rate was low. Of families with children under 18, only 60 percent reported receipt
of a child allowance (World Bank, 1995b). In Kyrgyz Republic, budgetary sequestration
resulted in a withdrawal of the child allowance in 1994, and the substitution of a new
benefit, the common monthly subsidy. In Estonia, fiscal austerity resulted in flat-rate
pensions in 1993 and a withdrawal of social assistance benefits inherited from the Soviet
Union.
In Russia and Kyrgyz Republic, increases in the age of the household head were
associated with small increases in household welfare, but only in Kyrgyz Republic was
there a discernible U-shaped life-cycle pattern. Other demographic variables were not
significant, except for the number of adult males in Estonia.
Overall, these results underscore the critical role of the labor market in determining
household welfare in the FSU. Those households, who have been able to most effectively
capture the returns from their own labor and effort in owing family businesses or private
plots, have been able to stay out of poverty. Individuals with university education seem to
have been best suited to capture the new possibilities as domestic markets have opened up
and economic control and regulation relaxed.
The importance of location, household participation in the labor market, and
education in affecting household welfare presents formidable challenges to country
authorities seeking to reduce poverty. and to keep poor households from becoming poorer.
Whether a household has a private plot or university-educated head are not easy factors to
change in the short run, while the extreme locational disparities are so large as to be
75



unlikely to be rectified in even the longer term.  However, the importance of
entrepreneurial activity for increasing household welfare may very well be fostered by
steps in the short- and medium-term, perhaps including public works and micro credit
programs. In Russia, where non-governmental organizations have been especially active
(Nizhnyy Novgorod, Yekaterinburg), credit unions and small business incubators have
been set up.
Poverty Equations
Given the difficulties of conducting household surveys in the FSU (high refusal
rates, extremely high reluctance to reveal sensitive information on income and alcohol
consumption), it is not surprising that measurement error is a significant concern for the
data sets on Estonia, Russia, and Kyrgyz Republic. Additionally, the lower levels of
goodness-of-fit for the FSU countries noted above suggest that a binary dependent
-variable (poor/non-poor) estimated by probit techniques might perform better and would
provide valuable information about some of the variables included. In this sense, lack of
consistency between the probit and OLS results would demonstrate problems either with
measurement error or in specification.
The probit results in Table 15 suggest an acceptable fit, but not a particularly good
one. The model serves to predict correctly the poverty status of about three-quarters of
Estonian households, but only 60-65 percent of Russian and Kyrgyz Republic households.
As was the case with the OLS results, not all coefficients were found to be significantly
different from zero at a 90 percent confidence level, and a small subset of the coefficients
which were significant in the OLS were not significant in the probit (Estonia: primary
76



education and inactive head; Russia: primary education; and Kyrgyz Republic: two
education variables and unemployed head). In Russia, the tenancy status of the household
was significant for the probit but not for the OLS.
Results of the probit are reported in Table 15 as the probability derivatives at the
mean levels of continuous variables, and for a discrete 0 to 1 change for the dummy
variables. The estimation used non-poor as the base category, so derivatives with a
positive sign indicate an increased probability of being poor and derivatives with a
negative sign pertain to reducing the chance of being poor. Other than the four cases
reported above where a variable was significant in OLS but not in the probit (or vice-
versa), the results suggest that the determinants of poverty identified in the probit are
essentially the same as the determinants of household welfare discussed in the preceding
section.
77



Table 15: Poverty Equations (Probit): FSU Countries
Estonia                Kyrgyz Republic               Russia
Probability   Standard     Probability    Standard    Probability    Standard
Derivatives     Error      Derivatives    Error      Derivatives     Error
Number of children                          0.056*       0.012          0.020*     0.008          0.065*      0.010
Number of male adults                       0.050*       0.021          0.008      0.014          0.011       0.014
Numberoffemaleadults                        0.013        0.018         -0.010      0.014          -0.006      0.013
Number of elderly                           0.037        0.022          0.003      0.026          -0.005      0.016
Education of head: primary                  0.045        0.023          0.001      0.037          0.032       0.023
Education of head: vocational/technical      ...          ...          -0.027      0.036          -0.028      0.019
Education of head: university              -0.149*       0.029         -0.059      0.036          -0.094*     0.023
Age of head                                 0.003        0.003         -0.010*     0.005          -0.006*     0.003
Age of head squared                         -0.000       0.000          0.000*     0.000          0.000       0.000
Female head                                 0.077*       0.028          0.119*     0.036          0.066*      0.020
Household owns enterprise                   -0.135*      0.022         -0.094*     0.025          -0.127*     0.023
Household owns land                        -0.159*       0.022         -0.092*     0.025          0.010       0.018
Household is renter                         0.088*       0.019         -0.032      0.043          -0.046*     0.016
Share of wages in household income         -0.277*       0.033         -0.104*     0.041          -0.184*     0.025
Number of unemployed in household           0.115*       0.025          ...         ...           0.137*      0.024
Head is unemployed                           ...          ...           0.036      0.044           ...        ...
Head is inactive                            0.046        0.030          0.014      0.039          0.074*      0.024
Location: non-capital city                  0.154*       0.024          0.195*     0.039          0.187*      0.025
Location: village                           0.273*       0.032          0.266*     0.035          0.222*      0.030
Log-likelihood                          -1504.8                     -1241.5                   -3201.8
Chi-squared                               494.53*                     136.75*                   478.38
Prob > chi-squared                          0.000                       0.000                     0.000
% correct predictions                      72.6                        61.5                      65.0
Note: * Indicates significance of the underlying coefficient at 90% level probability. Derivatives are taken at the mean values of continuous
variables or for discrete change of dummy variables from 0 to I.
78



There are discrete quantitative differences in the coefficients estimated by the two
procedures, which is as expected, since the probit equation is based on different
infornation than the levels regression, although the rank ordering is much the same
(except in Kyrgyz Republic). For example, in Estonia, share of wages in household
income (increase of 33 percent) and location in a village (decline of 31 percent) had the
largest effects on household welfare. In the probit, these two variables had the largest
effect on the probability of the household's being poor, with changes in probability of 28
percent for both, with the appropriate signs. The main exception to the consistency of the
findings was for primary education of the household head, which was found to reduce
household welfare by 7 percent in the welfare regression but was not significant in the
probit regression. Access to land was in fifth place for determining poverty but in eighth
place for welfare. Aside from this minor reordering, there were no other significant
changes. The first four variables which were most significant for poverty were also the
most significant for household welfare: share of wages in household income, location in a
village, university education of head, and non-capital city location.
In Russia, two factors (renting one's home status and age of head) were associated
with poverty in the probit regressions but were not identified as contributing significantly
to household welfare, while one factor which did reduce household welfare (primary
education for household head) was not significant for poverty. However, these three
variables were all relatively unimportant-renting reduced the risk of poverty less than 5
percentage points and age of head by less than one percentage point (the two lowest of the
significant variables for poverty), while primary education reduced welfare by about 8
79



percentage points (the lowest of all significant variables for welfare). More significantly,
the ranking of the first five variables remained the same for both poverty and welfare:
rural, non-capital city, share of wages in household income, number of unemployed in
household, and enterprise ownership.
In Kyrgyz Republic, the picture is less clear. The locational factors which were
most significant for household welfare were also for poverty (rural and non-capital city
respectively), but the probit found that female headship was in third place (increasing the
risk of poverty by 12 percentage points while in the OLS, it was in fifth place (reducing
welfare by 19 percent). The correspondence unraveled further for other factors. In the
OLS, land access was third most important, raising household welfare by nearly 25
percent, but in the probit, land access was in sixth place and only reduced the chance of
poverty by approximately 9 percentage points. The education variables which were found
to be significant for Kyrgyz household welfare were insignificant in the poverty
regressions, and the probit also did not identify an unemployed head as a risk factor for
poverty, although it was found to reduce household welfare by nearly 15 percent.
Aside from some of the rerankings in the Kyrgyz Republic, the overall pattern of
poverty determinants is consistent with the pattern demonstrated by the welfare
regressions. The findings represent a challenge for social assistance authorities in the FSU
countries, since the factors most significant for household welfare and poverty are ones
that are extremely difficult to change in the short-run: location and share of wages in
household income. Demographic characteristics (number of children, female-headed
households) were generally statistically significant, but only increased the risk of poverty
80



slightly (usually under 10 percentage points) and less than reducing welfare
(10-20 percent).
Poverty Gap Equations
Tobit estimates for the right-censored subsample of poor households are presented
in Table 16. As discussed earlier, this formulation is essentially equivalent to estimating
the determinants of the poverty gap, and the Tobit coefficients can be compared to the
welfare regression coefficients estimated by OLS over the full sample (Table 14). This
comparison is an informal test of whether the parameters apply to the entire welfare
distribution, and is clearly rejected for many of the variables, which have Tobit coefficients
sharply higher than in the OLS full sample estimate.
The welfare loss from living outside of the capital is dramatically larger for the
poor than overall in Russia and Kyrgyz Republic, suggesting that rural poverty is more
much severe and other urban (non-capital) poverty a significant problem. Rural poverty is
also associated with a higher welfare loss to the poor in Estonia, although the welfare
differential of location in a non-capital city is virtually the same for the poor as for the
general population. The reduction in welfare of the poor to living in a rural area is greater
than 60 percent in Russia and Kyrgyz Republic, and reaches nearly 40 percent in Estonia.
In the non-capital urban areas, the reduction of welfare of the poor is 52 percent in Russia
and 44 percent in Kyrgyz Republic, but only 19 percent in Estonia. These findings suggest
81



Table 16: Poverty Gap Equations (Tobit): FSU Countries
Estonia                Kyrgyz Republic               Russia
Dependent Variable: In (Expenditure per  Parameter   Standard     Parameter    Standard    Parameter    Standard
Equivalent Adult) Right-Censored at Poverty    Estimates  Error    Estimates      Error      Estimates      Error
Line
Intercept                                   7.287*       0.122         -9.721*    0.249          1.124*      0.176
Number of children                         -0.075*       0.016         -0.058*    0.017          -0.145*     0.021
Number of male adults                      -0.060*       0.029         -0.034      0.030         -0.036      0.033
Number of female adults                     0.004        0.024         0.032       0.029         0.038       0.030
Number of elderly                          -0.031        0.030         -0.028      0.054         -0.000      0.036
Education of head: primary                 -0.079*       0.030         0.028      0.078          -0.113*     0.052
Education of head: vocationaltechnical       ...         ...           0.086      0.077          0.083*      0.044
Education of head: university               0.265*       0.048         0.171*     0.079          0.256*      0.058
Age of head                                -0.003        0.005         0.019      0.011          0.017*      0.007
Age of head squared                         0.000        0.000         -0.000     0.000 °0.000*              0.000
Female head                                -0.090*       0.038        -0.229*     0.073          -0.182*     0.046
Household owns enterprise                   0.173*       0.033         0.205*     0.056          0.316*      0.064
Household owns land                         0.228*       0.030         0.304*     0.053          -0.058*     0.042
Household is renter                        -0.152*       0.027         0.115      0.095          0.052       0.039
Share of wages in household income          0.367*       0.046         0.289*     0.088          0.407*      0.058
Number of unemployed in household          -0.203*       0.032          ...        ...          -0.305*      0.053
Head is unemployed                          ...          ...          -0.165*     0.093           ...        ...
Head is inactive                           -0.112        0.040         -0.067     0.080          -0.193*     0.053
Location: non-capital city                 -0.188*       0.033         -0.440*    0.085          -0.523*     0.065
Location: village                          -0.366*       0.042        -0.615*     0.085          -0.637*     0.072
Log-likelihood                          -1466.8                    -1709.5                   -4444.4
Chi-squared                               542.6                      172.66                    555.9
Prob > chi-squared                          0.000                      0.000                     0.000
SER                                         0.506                      0.918                     0.968
Note: * Indicates that coefficient is significantly different from zero at 90% confidence level.
82



the importance of developing a rural development & poverty reduction strategy, perhaps
centered around land privatization which has moved very slowly in most of the FSU.
Preliminary findings for Armenia strongly suggested the importance of the 1992 land
privatization program in preventing rural poverty (World Bank, 1996a), which was
significantly lower than urban poverty in Armenia during 1994,1s
In Russia, the household link to the labor market in terms of the share of wages in
household income followed locational factors as the third-most significant determinant of
welfare of the poor, as for the general population, but the welfare premium of maintaining
the wage-labor market link was more significant for the poor (increasing welfare
40 percent) than for the population at large (30 percent). In Estonia, the link to the labor
market was identified as the largest determinant of overall welfare (increasing it by one-
third), and this was also true for the poorer half of the distribution.
In Kyrgyz Republic, the share of wages in household income and access to land
both had returns to the poor which were larger (29 and 30 percent) than to the general
population (20 and 25 percent). This suggests that active participation in farming or the
official labor market is an even more effective vehicle for reducing poverty than for
increasing welfare overall. Access to land also was more important for reducing poverty
'g At the time of this writing, of the FSU countries, only Armenia had instituted a full-scale land privatization
program (1992) by which former state and collective farms had been broken up and all land holdings passed to
individuals. In the other countries, households retained access to land through their private plots. Under Soviet
law, rural and even urban households were entitled to a very small (less than 0.06 and 0.02 hectares respectively)
plot of land. These land plots were passed from generation to generation, comprised 3 percent of arable Soviet
land, and produced up to 25 percent of the gross output of (non-wheat) agricultural production. (Gregory and
Sta, 1990).
83



(raised the welfare of the poor 23 percent) than for raising welfare generally (increased
welfare for the full sample about 16 percent) in Estonia. Land access was not found to be
especially important for the poor in Russia, although it was significant in the Tobit
regressions, unlike the OLS findings, but further interpretation is futile considering the
measurement problems associated with this variable for Russia.
Demographic characteristics tended to demonstrate approximately the same effect
on the welfare of the poor as on the welfare of the general population. Female-headship
led to a larger reduction of welfare among the poor vis-a-vis the general population in
Russia and Kyrgyz Republic (18 and 23 versus 14 and 19 percent, respectively), but this
was not the case in Estonia where the level was essentially the same (poor 9, general 10
percent).
Lastly, education effects were found to be distinctly larger for the poor than the
non-poor, suggesting that retraining and supplemental educational programs could have a
distinct impact on poverty reduction, although the orders of magnitude of increasing
welfare of the poor through education (10-30 percent) are much lower than the impact of
rural location (60 percent).
Overall, the poverty gap equations demonstrate that although the returns to human
and household capital are higher for the poor than for the general population, the
overwhelming effect of location presents a significant challenge for poverty reduction.
Without some sort of rural development strategy, and a corollary development strategy for
non-capital urban areas in Russia and Kyrgyz Republic, it is difficult to postulate that
conventional human capital development strategies will have a sufficient effect on poverty
84



reduction. At the same time that these countries are faced with the daunting challenge of
rural and regional development, improved targeting of scarce social protection resources
is imperative. We turn to this issue in the next section.
5.    Means-Testing and Indicator-Based Targeting
In the previous section we explored the determinants of welfare and poverty, and
of the depth of poverty. This has provided a number of useful findings, especially
regarding the role of household assets and the link to the labor market, which can be used
in the design of poverty reduction programs-either for the targeting of transfers, or in
active employment creation policies. In this section, we try to answer the question
whether indicator targeting is a feasible modus operandi in such policies. The indicators
considered are the economnic and demographic household characteristics which we used as
regressors in the models. Indicator-targeting is usefil in situations where an overall
means-test is difficult to administer because it is costly and/or unreliable.
Indicator-based targeting is commonly used in East European and FSU countries
for certain components of the social safety net19. Family allowances are allocated on the
basis of the number of children.  Eligibility for social assistance often relies on a
combination of indicators pertaining to household size, ownership of durable goods (e.g.
car or house), and employment status. It is generally not known how efficient such
targeting mechanisms are in correctly identifying the poor.
9 For a discussion of indicator-based targeting in other regions, see e.g. Grosh and Baker (1995) for Latin America
and Subbarao et al. (1997) for other regions.
85



This can be checked empirically using the welfare or poverty equations we
estimated in the previous section, by comparing predicted with actual values of the
dependent variable and calculating the percentage of correct predictions. Below we report
the results of one such exercise, based on an expanded welfare regression. As we
discussed in Section 2, the expansion consists of adding variables for "official" income
(wages and social transfers) and for ownership of household durables. Due to the
endogeneity of these variables, no causal interpretation should be given to the coefficients.
The purpose is simply to test their predictive ability. The model was estimated with
forward stepwise regression.
The set of regressors-household durables, official income, demographic
household characteristics, location, employment status-are all fairly easily identifiable
indicators, of the sort that social workers could observe or ask about easily in the course
of a visit to a household to determine eligibility for a transfer program. How well do they
identify the poor?
A.    Eastern Europe
Table 17 shows, for the three East European countries, the five and ten best
predictors and the results in terms of identifying correctly poor and non-poor households.
Overall, the results are impressive: the set of 25-30 indicators included in the model
correctly predicts poverty status in about 80 percent of the cases. However, the model is
clearly much better in identifying the non-poor, with an accuracy of 90 percent for
Bulgaria and 97 percent in the cases of Hungary and Poland. For the poor, the results are
86



much worse, with no model reaching even 50 percent accuracy. Clearly, this is inadequate
for real-life application.
Table 17: Stepwise Targeting Regressions (All Observations)
East European Countries
Bulgaria                 Hungary                   Poland
Best Five Predictors
Color TV       --       Car                       Washing machine
Education: primary      Wage income               Number of children
Refrigerator            Color TV                  Wage income
Car                     Number of children        Car
Wage income -           Education: primary        Number of male adults
% Correct Predictions
Poor                 47.3                     13.1                    16.7
Non-poor             86.8                     98.7                    97.5
All                  75.9                     81.1                    82.1
Second Best Five Predictors
Number of children      Renter                    Social transfers
VCR                     Number of unemployed      Household enterprise
Renter                  Washing machine           Education: university
Household enterprise    Education: university     Number of unemployed
Number of male adults    Household enterprise     Number of female adults
% Correct Predictions
Poor                 35.5                     24.7                    22.7
-Non-poor             92.3                     97.2                    96.7
All                  76.6                     82.4                    82.6
All Variables - % Correct Predictions
Poor                 45.4                     30.7                    24.8
Non-poor              90.8                    96.6                    97.0
All                  78.3                     83.1                    83.2
Note: Dependent variable is the log of per equivalent adult expenditure. The regressors are the same as in the
welfare and poverty regressions with the addition of wage and transfer income and consumer durables.
87



The results also show a remarkable robustness to the numbers of indicators used.
In the case of Poland, the five best predictors do almost as good a job at identifying
poverty status as the full set. In the case of Hungary, the best five variables identify the
non-poor almost perfectly (98.7 percent) but correct identification of the poor improves
significantly, from 13 percent to 31 percent, as more indicators are added.
It is noteworthy that several household durables keep coming back across
countries as good predictors-car, color TV, washing machine top the list. These are
clearly durables that identify the rich. Other critical predictors are number of children,
ownership of a household enterprise, level of wage income, renting one's home,
education, and number of unemployed in the household.
Given the apparent success of the model in identifying the rich, we undertook a
second simulation. Suppose that the variable list identified in Table 17 was used to
correctly identify the upper half of the distribution, how well would the set of indicators
do to distinguish poor from non-poor households within the group of household below
median welfare level?
The results in Table 18 show that the indicators are now quite able to correctly
identify poor households: success rates range from a low of 60 percent in Poland to a
high of 87 percent in Bulgaria (actual poverty rates in the below-median sample are
respectively 38 percent and 55 percent). This is a very respectable performance and
suggests that such approach is worth considering for real-life application. Moreover, in
88



Table 18: Stepwise Targeting Regressions(Observations Below Median)
East European Countries
Bulgaria                 Hungary                   Poland
Best Five Predictors
Refrigerator             Color TV                 Washing machine
Color TV                 Renter                   Number of children
Education-primary        Car                      Wage income
Land ownership           Number of children       Social transfer income
Wage income              Wage income              Number of male adults
% Correct Predictions
Poor                  87.8                     69.1                   53.7
Non-poor              27.3                     62.0                   76.1
All                  60.6                      64.8                   67.6
Second Best Five Predictors
Number of male adults    Refrigerator             Car
Number of children       Number of unemployed    Color TV
Age of head of household   Education - primary    Number of unemployed
Car                      Social transfer income   Number of elderly
Social transfer income   Sewing machine           Number of female adults
% Correct Predictions
Poor                 87.3                      65.6                   57.8
Non-poor              32.5                     70.9                   74.8
All                  62.7                      68.8                   68.3
All Variables - % Correct Predictions
Poor                  86.7                     67.8                   60.4
Non-poor             35.9                      70.9                   75.0
All                  63.9                      69.7                   69.5
Note Dependent variable is the log of per equivalent adult expenditure. The regressors are the same as in the
welfare and poverty regressions with the addition of wage and transfer income and consumer durables.
89



Hungary and Bulgaria, the same level of correct identification of poor households was
achieved with the five best predictors alone.20 In Poland, going from five predictors to the
full set yields an improvement from 54 percent to 60 percent correct predictions for poor
households.
Interestingly, the set of predictors which emerges as the best is not that different
from those which came out of the estimation over the full sample. Among durables, car
and color TV  are still the best identifiers.  Among the other variables, household
composition and official income are now more to the fore.
In summary, this exercise illustrates that a fairly simple set of observable indicators
at the household level can be used to correctly identify 90 percent or better of non-poor
households. This could serve as a first-step screen to eliminate better-off households from
consideration in poverty-oriented programs. In a second step, the same indicators can be
used to identify the poor from the non-poor in the remaining bottom part of the
distribution. Success rates in this exercise were in the 60-87 percent range which is far
better than what current social assistance systems in Eastern Europe achieve. Grootaert
(1995, 1997a) has documented leakage rates of 47 percent of households in Poland and
almost 90 percent in Hungary in the case of social assistance.
20 Note that adding variables only ensures a better overal prediction rate. Predictions for the poor or non-poor
separately may actually go down. This is the case e.g. for Bulgaria, where the best five predictors correctly
identify 87.8 percent of poor households, but the full model identifies only 86.7 percent of poor households
correctly. Overall prediction rate however rose fom 60.6 percent to 63.9 percent.
90



There is thus significant scope to improve the targeting of social assistance and
other poverty-oriented programs, and the simulation reported here indicates that indicator-
based targeting can make a significant contribution. Our results suggest also that the list
of indicators will need to be country-specific. While our results are indicative of the
potential of indicator-based targeting, they are not a blue print for practical application.
Specifically, results could undoubtedly be improved by testing alternative combinations of
variables and by modifying scoring procedures. For example, the best predictors can be
given a greater weight than what the regression implicitly gives them. This would improve
results, and our findings must therefore be seen as a low-end estimate of the effectiveness
of indicator-based targeting to identify poor households.
a     Former Soviet Union
The regression findings for FSU clearly suggested a link between welfare and
poverty and such easily identified household attributes as location, the number of children
and elderly members, and whether the household is female-headed. Certain- traits, such as
the link to the formal labor market and a household enterprise, were associated with
higher levels of welfare.
Under the previous Soviet social welfare system, all benefits were categorical ones.
For example, all males aged 60 and over received some sort of pension (regardless of
whether they continued to work), which was also the case for all females aged 55 and
above. Starting in 1992, all children under the age of 16 (or 18 if they were full-time
students) were eligible for a general child allowance. Certain categories of people,
91



particularly the disabled received diverse benefits, such as free or reduced-price utilities
and transportation services.
Since most of those who received such categorical benefits were demonstrated to
actually be the non-poor (see various World Bank poverty assessments), categorical
targeting received significant and warranted criticism from external and internal advisors
and policy makers. However, the problem with categorical targeting may have been in the
poor choice of categories more so than the idea of using an indicator or combination of
indicators (a proxy means test) to identify the poor. The choice of categories was dictated
by political considerations-not by a carefil study of who was poor and what determined
poverty.
In this section, we try to determine whether a combination of indicators can
identify the poor, which in turn would provide the necessary information for effective
targeting of cash or in-kind benefits, or for active labor market policies. The indicators
used here are the same economic and demographic variables which were used as
regressors in our previous models, with some additional variables. In practice, in the FSU,
and particularly in Russia and Ukraine, increasingly benefits are being awarded to
applicants who meet a categorical filter and an income-test. Typically, this means-test is
based only on official income. As noted in the poverty profile section, in the FSU, official
income is a particularly poor predictor of household welfare, due to the pervasive informal
sector and the general unwillingness of households to disclose such sensitive information.
Preliminary evidence from the housing allowance subsidy programs in Ukraine and
Russia, which are based on official income (wages plus transfer income) suggests that an
92



official income-test has a very high error of exclusion (those who are actually poor are not
receiving the benefit).  Partly this originates from  the very different goal of these
prograns, which is to promote housing privatization, and partly it may originate from a
lack of consideration of other factors related to poverty which are not captured in official
income.
In order to improve means-testing where it currently exists, and to revise and
update the categorical approach overall, we estimate an expanded welfare equation with
variables added for official income (wages and social transfers) and for ownership of
household durables. The data in Table 19 shows that the proxy means test was able to
identify correctly the poverty/non-poverty status of approximately 65-75 percent of the
population, with all three countries having better predictions for the non-poor than for the
poor. Only about 60 percent (57-62) of the poor were identified correctly, but this still
represent a significant improvement over the previous single-indicator/categorical
approach used to allocate benefits such as old-age pensions and student stipends.2'
At first glance, the five best predictors for the FSU countries seem to be more
related to the non-poor side of the spectrum (wage income, car, color TV, household
business, university education, land ownership) as to the poor (transfer income). The
addition of the next five (best ten total) predictors shows a mixture of factors associated
21 Analysis of individual countries (Russia, Kyrgyz Republic) in World Bank poverty assessments and comparative
analyses (Knumn, Milanovic, and Walton 1994) found that in general, only child allowances were well-targeted
transfers in FSU countries. All other transfers were regressive or highly regressive.
93



Table 19: Stepwise Targeting Regressions (All Observations)
Former Soviet Union
Estonia              Kyrgyz Republic               Russia
Best Five Predictors
Wage income             Wage income               Wage income
Car                     Car                       Transfer income
Color TV                Washing machine           Color TV
Higher education        Color TV                  Refiigerator
Transfer income         Land ownership            Household enterprise
% Correct Predictions
Poor                 53.3                    57.4                       56.4
Non-poor             75.7                    67.0                       76.2
All                  70.3                    63.6                       68.9
Second Best Five Predictors
Stereo                  Number of children        Inactive head
Household enterprise    Renter                    Car
Number of unemployed    Household enterprise      Location: other urban
Inactive head           Location: rural           Location: rural
Number of children      Location: other urban     Sewing machine
% Correct Predictions
Poor                 58.3                    56.7                       57.1
Non-poor             76.8                    68.1                       77.0
All                  72.5                   63.7                        69.5
All Variables - % Correct Predictions
Poor                 61.9                    57.1                       56.9
Non-poor             77.1                    68.6                       75.5
All                  74.5                   64.0                        68.9
Note: Dependent variable is the log of per equivalent adult expenditure. The regressors are the same as in the
welfare and poverty regressions with the addition of wage and transfer income and consumer durables.
94



with higher welfare (stereo, car, household enterprise) as with low welfare (number of
children, transfer income, rural location, other urban location, number of unemployed,
inactive head). This addition does little to improve the fit, raising the overall correct
prediction rate only slightly (64-73 percent) and the rate for the poor a bit more (57-58
percent) than was observed by using only the five best predictors. The fiull model shows a
barely greater prediction accuracy.
Given the presence of-so many variables associated with the higher end of the
welfare distribution and the higher identification rates for the non-poor, we undertook a
second simulation similar to-what was done for the East European countries, but we found
vastly different results. If there was some way to screen out the upper portion of the
distribution, how well would the proxy means test distinguish among the poor and non-
poor in the lower half of the distribution? For the Eastern European simulation, we
assumed that the screen would correctly identify the upper half of the distribution, since
the identification rates for the non-poor were all above 90 percent. Although this was a
reasonable assumption for Eastern Europe, in the original expanded regression for the
FSU countries, only 70-80 percent of the non-poor were correctly identified, thus making
this assumption a bit more questionable. However, for consistency, we simply re-ran the
expanded welfare regression via forward stepwise regression on the half of the FSU
samples with welfare below the median.
The results in Table 20 demonstrate that such an assumed screen would somewhat
improve the identification of the poor in Estonia (from 62 percent to 66 percent correctly
identified) but would improve the identification of the poor much more in Russia and
95



Kyrgyz Republic, increasing to 80 and 83 percent respectively. Of course, there is a cost
to this-the few non-poor which remained in the below-median sample were either poorly
identified (Estonia), extremely poorly identified (Russia), or virtually unidentified (Kyrgyz
Republic). This suggests that a proxy means test system could perform rather well in
Russia and Kyrgyz Republic, and acceptably well in Estonia, provided that an effective
mechanism could be found to screen out the upper portion of the welfare distribution. In
all three cases even without the screen, the proxy means test would represent a significant
improvement over the old categorical approach.
Further, in all three countries, the five best predictors alone did as good a job in
identifying the poor (Kyrgyz Republic, Russia) or almost as well (Estonia) as did the full
model, implying that only a few key data would be required for collection. As in Eastern
Europe, the set of predictors which emerges as the best for identifying the poor (given that
the upper 50 percent of the distribution was screened out of consideration) is more or less
the same as which resulted from estimation over the full sample. Interestingly enough, for
Kyrgyz Republic, using the below-median observations resulted in only six variables
meeting the entry criteria for the forward stepwise regression: land ownership, wage
income, car, motorcycle, renter status, and washing machine. For Russia and Estonia,
more than 10 variables entered into the forward stepwise specification.22
22 Restricting the observations to those below the median significantly improved identification of the poor in Kyrgyz
Republic and Russia but worsened the identification of the non-poor, which thus prompted an additional
experiment with other regressors, in an ultimately futile attempt to improve the predictions of household
consumption. Adding "kitchen sink" variables like housing amenities (hot water, central heating, etc.) and an
additional dunmny variable for self-employed household head, resulted in error rates which were virtually
identical to those for the original specification for Estonia and Kyrgyz Republic and which were only marginally
better (2-3 percent) for Russia. This specification was therefore not further considered.
96



Table 20: Stepwise Targeting Regressions (Observations Below Median)
Former Soviet Union
Estonia              Kyrgyz Republic'               Russia
Best Five Predictors
Wage income               Land ownership           Wage income
Transfer income           Wage income              Color TV
Color TV                  Car                      Transfer income
Inactive head             Motorcycle               Education: primary
Number of unemployed      Renter                   Refrigerator
% Correct Predictions
Poor                  64.2                     83.0                       79.6
Non-poor              54.3                      0.0                       22.4
All                   63.7                    82.0                        73.1
Second Best Five Predictors
Land ownership            Washing machine'         Education: higher
Location: rural                                    Land ownership
Education: voc.-tech                               Household enterprise
Car                                                Renter
Washing machine                                    Number of elderly
% Correct Predictions
Poor                  65.4                     83.1                       79.6
Non-poor              63.1                      9.5                       22.2
All                   65.2                    81.5                        73.0
All Variables - % Correct Predictions
Poor                  65.5                     83.1                       79.5
Non-poor              61.1                      8.7                       21.8
All                   65.1                     81.3                       72.9
Note: Dependent variable is the log of per equivalent adult expenditure. The regressors are the same as in the
welfare and poverty regressions with the addition of wage and transfer income and consumer durables.
Only six variables met the entry criteria.
97



Overall, the acceptability of the proxy means test for the FSU countries depends
on the reasonability of the assumed screening device.  Unlike in Eastern Europe,
90 percent or more of the non-poor can not be assumed to be removed from consideration
through an inventory of their consumer durables and other factors. Only about 70-80
percent of the non-poor could be removed at best in the FSU. Once the non-poor are
removed from consideration, virtually the same information collected could be used to
further refine the identification of the poor and non-poor in the remaining portion of the
welfare distribution, resulting in identification rates of 65-82 percent. Although the
potential of proxy-means testing in FSU is not quite as impressive as in Eastern Europe, it
could still be a significant improvement over the existing system of categorical indicators,
which is plagued by very large leakage to the non-poor.
98



6.    Summary and Conclusions
Poverty has emerged as a significant problem in the transition economies.
Although more widespread in the Former Soviet Union, much "transitional" poverty has
proved to be difficult to eradicate even in Eastem Europe. The social protection systems
of the transition countries have been inadequate to meet the challenges of transition, being
both poorly targeted and costly. Although the open incidence of poverty increased
everywhere during the transition period, distinctly different patterns of poverty emerged
from the East European experience than from the Former Soviet Union. In general,
poverty correlates are more- sharply defined in Eastern Europe than in FSU, holding out
the potential for better targeting in Eastern Europe.
In this paper, we undertook a comparative analysis of poverty in three East
European countries and three FSU countries. We used the HEIDE data set, specially
constructed for that purpose. The analysis consisted of three tasks: a profile of the
incidence and depth of poverty using aggregate poverty indexes (Section 3); a multi-
variate analysis of the determinants of poverty (Section 4); and an empirical evaluation of
the role of means testing and indicator-based targeting in poverty alleviation programs
(Section 5).
We also raised a number of methodological issues. For the poverty profile, we
used the well-known P-alpha class of poverty indexes, disaggregated along relevant
socioeconomic and demographic dimensions. We opted, however, for a relative poverty
line, rather than the more customary approach of absolute lines in cross-country research.
In doing so, we put the comparability of the poverty profile ahead of the comparability of
99



the headcount. When countries have significantly different levels of GDP, the same
absolute line would cut-off from very small to very large proportions of the population,
which are difficult to disaggregate and compare in a meaningful way.
For the multivariate analysis we took note of the current debate over welfare
regressions and binary poverty regressions as the main analytic tool for poverty research,
but we argued that in transition economies both are needed as they serve different
purposes. While welfare regresssions utilize the maximum available statistical information
on the dependent variable, they ignore measurement errors of the type typically present in
transition economy databases. In our results, the two models provided qualitatively
consistent answers on the significant determinants of poverty and welfare, but underlined
that some variables, such as education and productive assets, play different roles in escape
from poverty as opposed to determining position on the nonpoor segment of the welfare
distribution.
Likewise, our Tobit-based estimation of the poverty gap indicated that parameters
measuring the impact of household characteristics on welfare are often not the same for
the poor and nonpoor. Many human and physical capital assets had higher returns for the
poor. Locational disadvantages were also often larger for the poor than the population at
large. Our results call for caution when relying on one single multivariate model to study
the correlates of poverty. We would argue that the three models used here (OLS Welfare
Model, Probit Poverty Model, and Tobit Poverty Gap Model) constitute a useful. minimal
set to investigate the determinants of poverty.
100



Using the criterion of two-thirds of household expenditure per equivalent adult
(OECD equivalence scale), poverty in Eastern Europe was found to be significantly lower
than in FSU countries. Hungary and Poland have the lowest poverty incidence (21-23
percent) and a poverty gap less than 15 percent of the poverty line. Bulgaria is slightly
worse-off with a poverty rate of 26 percent and a poverty gap of 20 percent. Each of the
FSU countries exceeds those statistics by far. Estonia has a poverty incidence of 30
percent and a poverty gap of 20 percent. In Russia and Kyrgyz Republic, the poverty rate
is around 40 percent and the poverty gap is in the 25-30 percent range.
The profile of poverty shows some common aspects for Eastern Europe and the
Former Soviet Union as well as pronounced differences:
_ rural poverty is higher than urban poverty; within urban areas, the capital city
has the lowest poverty (except in Hungary); however, in the East European
capitals, the poverty gap was higher than elsewhere in the country.
- in Eastern Europe, there is a very strong correlation between poverty incidence
and the number of children in the household; in the FSU this is less
pronounced, except in Russia;
> single person household, especially elderly females, have very high poverty
rates (except in Poland); their poverty is also more severe;
= consistent with this, pensioner households have above average poverty
incidence and gap (except in Poland);
101



=> however, the highest poverty rates are found among people who have lost an
active or regular connection with the labor market and live on social transfers
(other than pensions) or other non-earned income as prime source of revenue;
their poverty rates can be as much as three times higher than the national
average, especially if two or more household members are unemployed; these
households often absorbed the highest social cost of transition by failing to
obtain a regular source of earnings;
the poverty gap is remarkably uniform in East European countries, especially in
Poland and Hungary, indicating that social safety nets have prevented the
emergence of deep pockets of poverty. In the FSU, this is much less the case,
and frequently those with the highest poverty rate also have the highest poverty
gap;
= the key role of labor market connections should not lead to the conclusion that
there is no poverty among the working class. Many have low education
(primary education or less) or outdated vocational/technical education, and
while poverty rates are low for workers, their sheer mass in the total
population means that the working poor constitute the largest group of poor.
the connection between education and poverty suggests that only those with
special skills or university education succeed in escaping poverty in great
numbers, thanks to demand for their skills from the newly emerging private
sector.
102



=> Last, the profile shows that there is a gender dimension to poverty in each
country. Female-headed households have higher poverty incidence and gap in
each of the six countries.
Without wishing to downplay differences across countries, the common aspects in
this profile of poverty suggests that there is a case to be made for a poverty alleviation
policy for the- East Europe/FSU region as a whole. Lessons learned in one country are
likely to have applicability irrothers. Priorities in targeting (children, elderly, low-
educated workers, female-headed households) are similar across countries and the design
of targeted interventions can benefit from region-wide experiences. Needless to say,
social and cultural differences are important, and must be given their due weight, even if
economic behavior and responses-are similar.
The multivariate analysis has corroborated the univariate observations from the
poverty profile, and made it possible to compare net effects, controlling for all other
factors, and identify the highest pay-off actions. Although there is more variation across
countries, some common factors have emerged:
> education plays a key role for welfare improvements. There is always a
significant welfare penalty to having achieved only primary education, and in
some countries (those most advanced in the transition process) the penalty
extends to vocational and technical education as well. Since it is not practical
to quickly upgrade education and/or retrain huge segments of the population,
this aspect of poverty will remain a long-term challenge.
103



> ownership of a household enterprise has very high payoffs in several countries,
often increasing household welfare by 20-30 percent. Returns are even higher
for the poor.  Unquestionably, programs of information, micro-credit,
marketing, small business incubators, etc. to help entrepreneurs and
prospective entrepreneurs must take center stage in poverty alleviation in
Eastem Europe and FSU. There is every reason to believe that in the short- to
medium-term, employment creation will be much higher in the informal sector
than in the formal (often still-to-be privatized) sector.
- age and gender effects are of concern in some countries (Bulgaria, Kyrgyz
Republic) but not in others like Poland where the pension system has adequate
reach. This is primarily an issue of re-targeting pensions and other social
transfers, or of increasing the level of the minimum pension (paid primarily to
women who lack adequate work-tenure to receive higher old-age pensions), at
the expense of compressing the rest of the pension distribution.
The multivariate analysis has confirmned the importance of household composition,
especially the number of children: households do not succeed in maintaining their welfare
levels when the number of children increases. This calls for child-oriented transfer
programs to move away from general entitlements to means-tested or proxy means-tested
programs.
In the final section of this paper, we undertook an assessment of the potential of
proxy means-testing and indicator-targeting for poverty alleviation programs such as
social assistance. We used a set of easily identifiable household characteristics, including
104



demographic composition, education, employment status, location, household durables
and official income to estimate a forward stepwise regression to identify the best
predictors. The results for Eastern Europe were more promising than for FSU. In a first
run, we could successfully identify more than 90 percent of nonpoor households, using a
set of 25-30 indicators. A surprising but potentially very important result was that this
accuracy was only marginally reduced when the best 5 or 10 predictors were used. In a
second run, limited to the lowest half of the distribution (assuming that the first run had
successfully identified households in the top half), we correctly identified 60-87 percent of
the poor. Again, the loss in accuracy was small when using only the best 5 or 10
predictors. These are very respectable results and suggest that such an approach is worth
considering for real-life application.
In the FSU countries, the first-stage correct identification of non-poor households
achieved only 69-77 percent, and in the second stage the poor were correctly predicted
65-83 percent of the time. These results are still far better than the systems currently in
place in the FSU countries. In Russia, experiments are currently being undertaken with
indicator-targeting to see how alternative formulas and sets of indicators can improve
correct identification of needy social assistance recipients.
105



References
Ackland, R., J. Braithwaite, M. Foley, T. Garner, C. Grootaert, B. Milanovic, G-T. Oh,
S. Sipos, S. Tsirounian, Y. Ying (1997). "Household Expenditure and Income
Data for Transition Economies (HEIDE)" mimeo, Policy Research Group,
Washington, DC: World Bank.
Aldennan, H. and M. Garcia (1993). "Poverty, Household Food Security, and Nutrition
in Rural Pakistan". International Food Policy Research Institute Research Report
No. 96. Washington, DC: International Food Policy Research Institute.
Appleton, S. (1995). "The Rich are Just Like Us, Only Richer - Poverty Functions or
Consumption Functions?" Centre for the Study of African Economies Working
Paper No. 95-4. University of Oxford
. (1996).  "Women-headed Households and Household Welfare: An
Empirical Deconstruction for Uganda". World Development. Vol. 24. No. 12.
pp. 1811-1827
Atkinson, A. and J. Micklewright (1992). Economic Transformation in Eastern Europe
and the Distribution of Income. Cambridge, UK: Cambridge University Press.
Bardhan, P. K. (1984). Land, Labor and Rural Poverty. Delhi: Oxford University Press.
Braithwaite, J. (1990). "Income Distribution and Poverty in the Soviet Republics."
Journal of Soviet Nationalities. Vol. 1. No. 3.
. (1991). "The Social Safety Net in the Sovereign Republics." Conference
paper presented at the American Association for the Advancement of Slavic
Studies National Convention, November 1991.
. (1995). The Old and New Poor in Russia: Trends in Poverty.
Background paper for Russian Poverty Assessment. Education & Social Policy
Department Working Paper. Washington, DC: World Bank.
Chase, R. (1995). Women's Labor Force Participation During and After Communism: A
Case Study of the Czech Republic and Slovakia. Economic Growth Center
Discussion Paper No. 768. New Haven: Yale University.
Commander, S., A. Tolskopietenko, and R.   Yemslov (1997).  "Channels of
Redistribution:  Inequality and Poverty in the Russian Transition."  Paper
presented at the European Bank for Reconstruction and Development "Inequality
and Poverty in Transition Economies" Conference, May.
106



Davidson, R. and J. Mackinnon. (1981). "Several Tests for Model Specification in the
Presence of'Multiple Alternatives". Econometrica. Vol. 49. pp. 781-793
Diamond, C. A., C. J. Simon and J. T. Warner. (1990). "A Multinomial Probability
Model of Size Income Distribution". Journal of Econometrics. 43
Einhorn, B. (1993). Cinderella Goes to Market: Citizenship, Gender and Women's
Movements in East Central Europe. London: Verso.
Fong, M. (1996). "Gender Barriers in the Transition to a Market Economy". Poverty and
Social Policy Department Discussion Paper No. 87. Washington, DC: World
Bank.
Foster, J., J. Greer and E. Thorbecke (1994). "A Class of Decomposable Poverty
Measures." Econometrica. Vol.52. pp. 571-576.
Funk, N. and M. Mueller (1993). Gender Politics and Post-Communism: Reflections
from Eastern Europe and the Former Soviet Union. New York: Routledge.
Gaiha, R. (1988). "On Measuring the Risk of Poverty in Rural India". Chapter 7 in T. N.
Srinivasan and P. Bardhan (eds.), Rural Poverty in South Asia. New York:
Columbia University Press.
Glewwe, P. (1991). "Investigating the Determinants of Household Welfare in the Cote
dIvoire". Journal ofDevelopment Economics, Vol. 35.
GIewwe P. and G. Hall. (1995). "Who is Most Vulnerable to Macroeconomic Shocks?
Hypotheses Tests using Panel Data from Peru". Living Standards Measurement
Study Working Paper No. 117. Washington DC: World Bank.
Gregory, P. and R. Stuart (1990). Soviet Economic Structure and Performance. Fourth
Edition. New York: Harper & Row.
Grootaert, C. (1995). "Poverty and Social Transfers in Poland". Policy Research
Working Paper No. 1440. Washington, DC: World Bank.
. (1997a). "Poverty and Social Transfers in Hungary". Policy Research
Working Paper No. 1770. Washington, DC: World Bank.
_. (1997b). "The Determinants of Poverty in Cote d'Ivoire in the 1980s".
Journal of African Economies. 6(2) pp. 169-196
Grosh, M. and J. Baker (1995). Proxy Means Tests for Targeting Social Programs-
Simulation and Speculation. Living Standards Measurement Study Working
Paper No. 118. Washington, D. C.: World Bank.
107



Harris, J. and M. Todaro (1970). "Migration, Unemployment, and Development: A
Two-Sector Analysis." American Economic Review. Vol. 60. pp. 126-142.
Kaufinann, D. and A. Kaliberda (1996). "Integrating the Unofficial Economy into the
Dynamics of Post-Socialist Economies: A Framework of Analysis and Evidence."
In Kaminski, B. (Ed.) Economic Transition in Russia and the New States of
Eurasia. Armonk, NY: Sharpe. pp. 81-122.
Klugman and Braithwaite (1998). "Poverty in Russia During the Transition: An
Overview." WorldBankResearch Observer. Forthcoming, February.
Klugman (1997). (Ed.) Poverty in Russia: Public Policy and Private Responses.
Washington, DC: Economic Development Institute, The World Bank.
Koen, V. and E. Gavrilenko (1994). "How Large Was the Output Collapse in Russia?:
Alternative Estimates and Welfare Implications." International Monetary Fund
Working Paper. WP/94/6. Washington, DC: International Monetary Fund.
Koen, V. -(1996). "Russian Macroeconomic Data: Existence, Access, Interpretation."
Communist Economies andEconomic Transformation. Vol. 8. pp. 321-333.
Kolev, A. (1996). "Poverty Analysis in Russia: What Can We Learn from the RLMS
Round 6?" Florence, Italy: European University Institute.
Krumni, K., B. Milanovic and M. Walton (1994). "Transfers and the Transition from
Socialism: The Key Tradeoffs." Policy Research Working Papers Series.
- No. 1380. Washington, DC: World Bank.
Lanjouw, P. and N. Stern. (1991). "Poverty in Palanpur". World Bank Economic
Review. 5(1).
Maddala, (1983). Limited-Dependent and Qualitative Variables in Econometrics.
Cambridge: Cambridge University Press.
Milanovic, B. (1990). "Poverty in Poland, Hungary, and Yugoslavia in the Years of Crisis
1978-87". Policy Research Working Paper No. 507. Washington, DC: World
Bank.
. (1995). "Poverty, Inequality and Social Policy in Transition Economies".
Transition Economies Research Paper Series No. 9. Washington, DC: World
Bank.
. (1997). "Explaining the Growth in Inequality during the Transition".
Mimeo. Policy Research Department. Washington, DC: World Bank.
(Forthcoming). Income, Inequality, and Poverty During the Transition.
108



Perraudin, W. and T. Pujol. (1994). "Framework for the Analysis of Pension and
Unemployment Benefits Reform in Poland". International Monetary Fund Staff
Papers. 41(4), pp. 643-674
Ravallion, M. (1993). "Poverty Comparisons: A Guide to Concepts and Methods."
Living Standards Measurement Study Working Paper. No. 88. Washington, DC:
World Bank.
. (1996). " Issues in Measuring and Modeling Poverty". The Economic
Journal. 106, pp. 1328-1343.
Revesz, T. (1994). An Analysis of the Representativity of the Hungarian Household
Budget Survey Samples. Mimeo. Department of Applied Econornics. University
of Cambridge, United Kingdom.
Subbarao, K. et al. Safety Net Programs and Poverty Reduction: Lessons from Cross-
Country Experience. Washington, DC: World Bank.
World Bank. (1993). Poland: Income Support and the Social Safety Net During the
Transition. Washington, DC.: World Bank.
. (1994a). Averting the Old-Age Crisis. Washington, DC.: World Bank.
. (1994b).  Poland:  Poverty In Poland   Report No. 13051-POL.
Washington, DC: World Bank.
_  (1995a). Kyrgyz Republic: Poverty Assessment and Strategy. Report
No. 14380-KG. Washington, DC: World Bank.
(1995b).   Russia: Poverty in Russia: An Assessment. Report
No. 14110-RU. Washington, DC: World Bank.
(1995c). Ukraine: Poverty in Ukraine.   Report No. 15602-UA.
Washington, DC: World Bank.
_  (1995d). Understanding Poverty in Poland. Washington, DC: World
Bank.
. (1996a). Armenia: Confronting Poverty Issues. Report No. 15693-AM.
Washington, DC.: World Bank.
. (1996b). Belarus: An Assessment of Poverty and Prospects for Improved
Living Standards. Report No. 15380-BY. Washington, DC: World Bank.
. (1996c).  Estonia: Living Standards During the Transition. Report
No. 15637-EE. Washington, DC.: World Bank.
109



. (1996d). Hungary: Poverty and Social Transfers. Report No. 14658-HU.
Washington, DC: World Bank.
(1996e). World Development Report 1996.  From Plan to Market.
Washington, DC: World Bank.
110



Statistical Annexes
Annex 1: Means and Standard Deviations of Variables (East European Countries)
Bulgaria                 Hungary                    Poland
Mean       Standard      Mean      -Standard        Mean        Standard
Deviation'              'Deviation                   Deviation
Household size                                 2.92         1.56        2.78        1.33          3.13           1.61
Number of children                             0.47         0.80        0.58        0.90           0.75          1.06
Number of male adults                          0.98         0.84        0,90        0.74           1.00          0.78
Number of female adults                        1.02         0.79        0.97        0.67           1.09          0.69
Number of elderly                              0.45         0.66        0.33        0.59          0.28           0.56
Ageofheadofhousehold                          55.04        15.31       49.2        16.3          48.7           14.8
Age of head squared                         3264.2       1694.4.    2682.1       1703.9        2594.3         1537.5
Female head of household                       0.21         0.41        0.31        0.46          0.35           0.48
Head with primary education                    0.47         0.50        0.43        0.49           0.33          0.47
Head with secondary education                  0.31         0.46        0.22        0.41          0.26           0.44
Head with vocational/technical education       0.07         0.26        0.24        0.43           0.32          0.46
Headwithuniversityeducation                    0.15         0.35        0.11        0.32          0.10           0.30
Tenancy status: renter                         0.07         0.26        0.17        0.38           0.45          0.50
Household enterprise ownership                 0.05         0.21        0.08        0.27           0.06          0.23
Land ownership                                 0.40         0.49       -           -               0.50          0.50
Share of wages in total household income       0.29         0.32        0.40        0.36           0.38          0.39
Number of unemployed household members        -            -            0.21        0.49           0.16          0.42
Unemployed head of household                  -            -            0.05        0.22           0.02          0.15
Inactive head of household                    -            -            0.36        0.48           0.35          0.48
Capital city                                   0.15         0.36        0.22        0.41           0.07          0.26
Other city                                     0.52         0.50        0.41        0.49           0.60          0.49
Rural areas                                    0.33         0.47        0.37        0.48           0.33          0.47
111



Annex 2: Means and Standard Deviations of Variables (FSU Countries)
Estonia                   Russia                      Kyrgyz
Mean       Standard       Mean       Standard        Mean          Standard
Deviation                Deviation                     Deviation
Household size                                  2.41          1.34        2.75         1.39           4.93          2.76
Number of children                              0.51          0.85        0.59         0.86           1.82           1.70
Number of male adults                           0.70          0.68        0.82         0.72           1.39           1.07
Number of female adults                         0.88          0.65        0.98         0.68           1.48           1.01
Number of elderly                               0.32          0.58        0.35         0.60           0.24          0.50
Ageofheadofhousehold                           48.10         16.41       48.4         15.75          41.16          14.00
Age of head squared                          2582.5        1656.8      2590.7       1620.4         1889.6        1304.5
Female head of household                        0.36          0.48        0.29         0.46           0.18          0.38
Head with primary education                     0.28          0.45        0.28         0.45           0.33          0.47
Head with secondary education 1/                0.58          0.49        0.34         0.47           0.24          0.43
Head with vocationaUtechnical education 1/       ...           ...        0.16         0.37           0.25          0.43
Head with university education                  0.13          0.34        0.25         0.43           0.18          0.38
Tenancy status: renter                          0.50          0.50        0.55         0.50           0.08          0.27
Household enterprise ownership                  0.22          0.41        0.08         0.28           0.32          0.47
Land ownership                                  0.53          0.50        0.23         0.42           0.58          0.49
Share of wages in total household income        0.45          0.39        0.43         0.39           0.27          0.33
Number of unemployed household members          0.11          0.37        0.08         0.30            ..            ..
Unemployed head of household                     ...           ...         ...          ...           0.08          0.27
Inactive head of household                      0.32          0.47        0.34         0.47           0.22          0.42
Capital city                                    0.47          0.50        0.61         0.49           0.26          0.44
Other city                                      0.26          0.44        0.28         0.45           0.59          0.49
Rural areas                                     0.27          0.44        0.10         0.31           0.17          0.37
Notes:
1/ It was not possible to separate household heads with vocational-technical education from heads with general secondary education in Estonia, due to lack of
comparability between definitions used in the Estonian survey and those used in other countries.
112



Policy Research Working Paper Series
Contact
Title                            Author                  Date              for paper
WPS1925 Half a Century of Development     Jean Waelbroeck          May 1998           J. Sweeney
Economics: A Review Based on                                                31021
the Handbook of Development
Economics
WPS1926 Do Budgets Really Matter?          Emmanuel Ablo           June 1998          K. Rivera
Evidence from Public Spending    Ritva Reinikka                             34141
on Education and Health in Uganda
WPS1927 Revenue-productive Income Tax      Fareed M. A. Hassan     June 1998          A. Panton
Structures and Tax Reforms in                                               85433
Emerging Market Economies:
Evidence from Bulgaria
WPS1 928 Combining Census and Survey Data Jesko Hentschel          June 1998          P. Lanjouw
to Study Spatial Dimensions      Jean Olson Lanjouw                         34529
of Poverty                       Peter Lanjouw
Javier Poggi
WPS1929 A Database of World Infrastructure   David Canning         June 1998          A. Abuzid
Stocks, 1950-95                                                             33348
WPS1930 The Main Determinants of Inflation in llker Domac          June 1998          F. Lewis
Albania                          Carlos Elbrit                              82979
WPS1931 The Cost and Performance of Paid   Ariel Dinar             June 1998          F. Toppin
Agricultural Extenion Services: The   Gabriel Keynan                        30450
Case of Agricultural Technology
Transfer in Nicaragua
WPS1 932 Air Pollution and Health Effects:  Bart D. Ostro          June 1998          C Bernardo
A Study of Respiratory Illness   Gunnar S. Eskeland                         31148
Among Children in Santiago, Chile   Tarhan Feyzioglu
Jose Miguel Sanchez
WPS1933 The 1997 Pension Reform in Mexico  Gloria Grandolini       June 1998          C. Zappala
Luis Cerda                                 87945
WPS1934 WTO Accession for Countries        Constantine Michalopoulos June 1998        L. Tabada
in Transition                                                               36896
WPS1935 Explaining the Increase in Inequality  Branko Milanovic    June 1998          G. Evans
during the Transition                                                       85734
WPS1936 Determinants of Transient and      Jyotsna Jalan           June 1998          P. Sader
Chronic Poverty: Evidence from   Martin Ravallion                           33902
Rural China
WPS1937 Aid, the Incentive Regime, and     Craig Burnside          June 1998          E. Khine
Poverty Reduction                David Dollar                               37471



Policy Research Working Paper Series
Contact
Title                             Author                   Date              for paper
WPS1938 What Explains the Success           David Dollar            June 1998           E. Khine
or Failure of Structural Adjustment   Jakob Svensson                          37471
Programs?
WPS1939 Second Thoughts on Second           Arturo J. Galindo       June 1998           M. Cervantes
Moments: Panel Evidence on        William F. Maloney                          37794
Asset-Based Models of Currency
Crises
\fVPS1940 The Structure of Labor Markets in    William F. Maloney    June 1998          M. Cervantes
Developing Countries: Time Series                                             37794
E vidence on Competing Views
WPS1941 Are Labor Markets in Developing     William F. Maloney      June 1998           M. Cervantes
Countries Dualistic?                                                          37794