WFS Iqlo POLICY RESEARCH WORKING PAPER 1980 The Effect of Household While household wealth is strongly related to Wealth on Educational educational attainment of Attainm ent children nearly everywhere, the magnitude and pattern of the effect of wealth differs Demographic and Health Survey widely.Thegapinattainment Evidence of children of the poor and rich ranges from only one or two years in some countries Deon Filmer to nine or ten years in others. Lant Pritchett This attainment gap is the result of different patterns of enrollment and dropout: while in South America low attainment among the poor is almost entirely due to children who enroll then drop out early, in West Africa and South Asia many poor children never enroll. The World Bank Development Research Group Poverty and Human Resources September 1998 1II) R \io7 ;ilfi P 'i 1 98( Summary findings tI sitig hooseholdo( sur\e e data from 44 Demzograpliic and educational artain merit bet s% r ic ri . i: * I lCalth Sur\veys in 3,i c ounltrics, Filmer and Pritchett soile countries tile differeiic . rw\r en tilc :'a' p dociiilcit differenit paitternls in the enrollmenlt and in the mnedilan im hiber o, vear ( oo ( ,ni.ete i n attainment ot children fro(m riclh and poor housellolds. on]x a ear or t\\o; in itlier> t:te giip i i a' great .l - f.- t lhev find(i that: tenl year . Fninrollment profiles of the poor differ across * ihc attaililinellt pr tilCs ctll h( :Ied as JIagnlklit couintries but fall into distinictive regional patterns. In tools to CXamnIeIII iCSueS iII tIIe Cdiicati It I " >t(cI, soiime areas (including muich of South America) the poor including the cxtent to which enrollime-it ;s low b6eca.ise- reach nearlv univcrsal enrollnment in first grade but then of the physical uLinavailability ot schools. drop oUt in droves. In others (including much of South Filmer and Pritchctt ovecrcomile the Lack of data on Asia and West Africa), the poor never enroll. Both inconme and consumptioni expenditures in the sur\ eys >vh patterns lead to low attainment. constructing a proxy for longn-run houisehold \xealth. *'I'here are enormious differences across countries in using survey informationi on asSCt" an11d Using the the "wealth gap" - the difference in enrollment and statistical techniqUe of principal comnponents. -T his paper -a product of Poverty and Human Resources, Development Research G.roup - is part of a larger effort in the groUp to inforimi education policy. The study was funded by the Bank's Research Support Budget under the research project "Iducational Enrollment and Dropout" (RPO 682-1 1). Copies of this paper are available free from the World Bank, 1818 Hi Street NW, Washingtoni, DC 20433. Please contact Sheila Fallon, roomi MC3-638, telephionie '02-47/3-8009, fax 202- 822-1 153, Internet address sfallonriaworldbank.org. The authors man. he contacted at dfilmer ou orldbank.org or lpritchett(f worldbank.org. September 1998. (38 pages) I11e Polioc Rese.rc \Y orking IPaper Series disseminates the findings ot u owk in progrs tS r enicowaiige the exchange ot ideas a.sout dcle lopnment issues. Ani objectit e of the series is to get the findigs oiut quicklv. ci 'n i th pr 7tesen'ta 1 oins less tkan jUo/K polisoai. [½e papers .arr the iaoU Sf the authors and should be cited according/v. Tbe fIndings. iwiterpretations ii il .iJ cwlitlsionl s expressedl1 in this paper are e ntire1% trh(Sc of the authors. Thev do not necessarily represent thec 1ie it ot \t 1,1 Ia UikI !ts Exricutit i Directors, or the coulntlles tiex rcpreseet. Produced by the Policy Research D[isseminitiom (Cllntcl The Effect of Household Wealth on Educational Attainment Demographic and Health Survey Evidence Deon Filmer Lant Pritchett The Effect of Household Wealth on Educational Attainment Around the World: Demographic and Health Survey Evidence' Introduction ][n this paper we are interested not just in countries' average educational enrollment and attainment, for which there has been a great deal of examination both from official and academJ[c sources, but in how educational attainment differs by household wealth within countries.2 How much schooling are children from poor households India, Brazil, or Kenya receiving, both absolutely and relative to the rich in the same country? Answering this question, especially in a way that produces valid comparisons across countries is hampered by the limited availability, difficulty of use, and comparability of household survey data. The Demographic and Health Surveys (DHS), having applied essentially the same survey instrument in 35 countries potentially overcomes these problems. One potential limitation of the DHS is that it lacks questions on household income or consumption expenditures, which are conventionally used as indicators of households' economic status. However, in a separate methodological paper (Filmer and Pritchett, 1998a) we shoNv that an index constructed from the questions asked in the DHS about household assets and housing characteristics (e.g. construction materials, drinking water and toilet facilities) works as well, and arguably better, than consumption expenditures as a proxy for household long-run wealth. This finding allows us to use a comparable method, principal components, in ' This work is the result of research developed jointly with Jee-Peng Tan, and it has greatly benefited from her input. We would also like to thank Emiliana Vegas and seminar participants for helpful comments and suggestions. This research was funded in part through a World Bank Research support grant (RPO 682-1 1). 2 Several recent estimates of the stocks of schooling years in many countries have been produced based on the UNESCO Yearbook series enrollment rates and labor free and census surveys (Nehru, Swanson and Dubey, 1993; Barro and Lee, 1993; Dubey and King, 1994; Ahuja and Filmer, 1996). 2 constructing a ranking of households within each country. The "poor" are simply defined as the bottom 40 percent in each country, so while levels of poverty are not comparable across countries, the rankings are constructed using a similar method. An analysis of this data on education and wealth reveals three key findings. First, very low primary attainment by the poor is driven by two distinct patterns of enrollment and drop- out. There is a South Asian and Western/Central African pattern in which many of the poor never enroll in school. In these countries more than 40 percent of poor children never complete even grade 1 and typically only one in four complete grade 5. In contrast there is a Latin American pattern in which enrollment in grade 1 is (nearly) universal but drop-out is the key problem. In South American countries less than 10 percent of the poor never enroll, but drop-out is so high that median years of school completed is only between 4 and 6 years. Even though 92 percent of the poor in Brazil complete grade one, only 50 percent of those complete grade 5. The result is that median attainment of poor children in South America is less than that of poor children in Ghana, Kenya, or Zimbabwe. Second, the wealth gaps vary enormously across countries and in most instances raising the enrollment of the poor will be the key to achieving universal basic education. The difference in median grade attainment between the poor and rich is very high in South Asia (10 years in India, 9 in Pakistan), high in Latin America and Western/Central Africa (4 to 6 years) and low in Eastern/Southern Africa (1 to 3 years). Where the wealth gap is large, increasing the educational attainment of the poor will play the key role in universalizing primary or basic education. In Colombia and Peru over 70 percent of the shortfall from primary completion is due to children from the bottom 40 percent of households. 3 Third, these data cast some doubt on the notion that physical availability of school facilities at the primary or secondary level is the key issue in many countries. In South America typically over 90 percent of the shortfall from primary completion is from children that complete grade 1 (hence likely could attend a school) but fail to complete primary school. In South Asia and Western/Central Africa a larger fraction is due to children that never enroll, but in those countries the wealth gap suggests that even poor children had physical access to schools. A companion paper examining differences within Indian states has estimates of school effects, which are quite small relative to household wealth impacts (Filmer and Pritchett, 1998b). This suggests that in many cases issues of the access to quality schooling and maintaining household demand are as important as the number of schools. At the secondary level the smooth patterns of attainment do not suggest that high drop- out across the transition from primary to secondary is a major issue except in a small number of cases (e. g. Turkey, Indonesia, Tanzania). In many ways this analysis confirms findings of previous studies. There are many country specific studies which look at the enrollment rates by wealth groups. In the context of benefit incidence analysis there is even some cross national compilation of those results (Castro-Leal, Dayton, Demery, and Mehra, 1997). The main value-added of this paper is the direct comparability across countries of educational data, the focus on not just enrollment but the entire attainment pattern (showing the importance of drop-out within levels), and a comparable methodology for documenting attainment differences due to household wealth. 4 I) Data and Methods A) The Demographic and Health Surveys The Demographic and Health Surveys (DHS) are large nationally representative household surveys.3 The surveys have been carried out using a nearly identical survey instrument in over fifty developing countries.4 While the main purpose of the surveys is to inquire about family planning and child and maternal health, the surveys also contain an educational history of all household members from a chosen respondent. The education variables we analyze are based on four questions: * Has [name] ever been to school? * If attended school: what is the highest level of school [name] attended? what is the highest grade/years [name] completed at that level? * If attended school: Is [name] still in school? These questions are used to construct an " attainment" history for a recent cohort, those aged 15 to 19 inclusive. This attainment profile is the proportion of the cohort who have completed any given grade or higher. The analysis so far has covered 35 countries. Countries have been grouped into six regions. The groups, ranked from lowest to highest median attainment of the bottom 40 'Table 1 shows that the samples of individuals in the 15-19 age range are usually above 2,000, but vary from 1,355 in Kazakhstan to over 50,000 in India. 4 There are three main designs of the survey instrument. DHS I surveys were carried out between 1985 and 1989, DHS II between 1990 and 1993, and DHS III are those that have been carried out since 1994. 5 percent are: Westem/Central Africa, South Asia, Central America and Caribbean, South America, Eastern/Southern Africa, East Asia, and Central Asia / North Africa / Europe. B) Constructing an "asset index" The DHS do not ask about household income or consumption expenditures, but the DHS [E and III survey instruments do include two sets of questions related to the economic status of the household. First, households are asked about their ownership of various assets, such as whether any member owns a radio, a television, a refrigerator, a bicycle, a motorcycle, or a car. Second, they are asked about characteristics of their housing, namely whether electricity is used, the source of drinking water, the type of toilet facilities, how many rooms there are for sleeping, and the type of materials are used in the construction of the dwelling. There is substantial overlap in the questions asked in each country, but the precise list varies. The number of variables constructed from these questions is usually 15 or 16 but varies from 12 to 21 (last column of Table 1). In order to use these variables to rank households by their economic status, they need to be aggregated into an index and of course the main problem in constructing such an index is choosing appropriate weights.5 We use the statistical technique of principal components to derive weights. Principal components is a technique for summarizing the information contained in a set of variables to a smaller number by creating a set of mutually orthogonal components of the data. Intuitively, the first principal component is that linear index of the underlying variables that captures the most common variation among them. If these assets were only to be used to used examine the impact of some other factor (e.g., maternal education) as a "control" for wealth in a multivariate regression we would not need to aggregate the variables (Montgomery, Burke, Paredes, and Zaidi, 1997) 6 Table 1: Summary information Country Year Number of Proportion of Value of Ist Difference Number of households variance eigen value between Ist and assets explained by Ist 2nd eigen PC values Westem and Central Africa Benin 1993 4499 0.268 4.293 2.722 16 Burkina Faso 1992-93 5143 0.276 4.005 2.270 1 5 Cameroon 1991 3358 0.247 3.809 2.032 15 C.A.R. 1994-95 5551 0.240 3.845 1.961 16 Cote d'lvoire 1994 5935 0.223 3.341 1.670 15 Ghana 1993 5822 0.211 3.166 1.618 15 Mali 1995-96 8716 0.230 3.448 1.430 15 Niger 1992 5242 0.265 4.234 2.553 16 Nigeria 1990 8999 . . . 0 Senegal 1992-93 3528 0.231 3.554 2.043 15 South Asia Bangladesh 1993-94 9174 0.285 3.987 2.334 14 Bangladesh 1996-97 8682 0.309 4.018 2.460 13 India 1992-93 87175 0.256 5.368 3.713 21 Nepal 1996 8082 0.219 2.622 0.898 12 Pakistan 1990-91 7193 0.283 4.237 2.704 15 Central America Dominican Republic 1991 7144 0.249 4.227 2.676 17 Dominican Republic 1996 8831 0.241 3.848 2.372 16 Guatemala 1995 11297 0.264 3.958 2.534 15 Haiti 1994-95 4818 0.266 3.987 2.230 15 South America Bolivia 1993-94 9114 0.311 3.732 2.347 12 Northeast Brazil 1991 6064 0.263 4.204 2.860 16 Brazil 1996 13283 0.226 3.163 1.261 14 Colombia 1990 7412 0.216 3.246 1.970 15 Colombia 1995 10112 0.240 3.606 2.325 15 Paraguay 1990 6348 . . . 0 Peru 1991-92 13479 0.283 4.238 2.878 15 Peru 1996 28122 0.267 4.001 2.540 15 Eastern and Southern Africa Comoros 1996 2252 0.230 3.453 1.738 15 Kenya 1993 7950 0.264 3.961 2.362 15 Malawi 1992 5323 0.186 2.598 1.071 14 Namibia 1992 4101 0.300 4.499 3.051 15 Rwanda 1992 6252 0.200 2.798 1.308 14 Tanzania 1991-92 8327 0.187 2.798 1.001 15 Tanzania 1996 7969 0.202 3.036 1.114 15 Uganda 1995 7550 0.192 2.886 1.023 15 Zambia 1992 6209 0.259 3.879 2.108 15 Zambia 1996-97 7286 0.275 4.121 2.695 15 Zimbabwe 1994 5984 0.273 4.101 2.216 15 East Asia and Pacific Indonesia 1991 26858 0.296 2.665 1.051 9 Indonesia 1994 33738 0.258 3.352 1.585 13 Philippines 1993 12995 0.257 3.596 2.200 14 Middle East, North Africa, and Europe Egypt 1992 10760 0.266 3.452 1.943 13 Egypt 1995-96 15567 0.250 3.255 1.861 13 Kazakhstan 1995 4178 0.203 3.045 1.479 15 Morocco 1992 6577 0.286 4.571 3.163 16 Turkey 1993 8612 0.234 2.806 1.511 12 Unweighted average 10687 0.250 3.659 2.065 14 Unweighted std dev 13093 0.032 0.605 0.659 3.5 Median 7481 0.256 3.771 2.154 15 7 'We assume that the most "common variation" in the set of asset variables is a good proxy for a household's wealth. Filmer and Pritchett (1998a) defends this assumption, showing the asset index performs as well as a more traditional measures, such as household size adjusted consumption expenditures. Empirical estimates in that paper suggest that the asset index works as well, or better, as a proxy for long-run household wealth to predict childreii's enrollment than consumption expenditures. There are two key findings that suggest assets might work "better". First, the enrollment profile is consistently "flatter," that is it shows smaller gaps between rich and poor, when using expenditures as opposed to assets, which is consistent with a large transitory component in expenditures. Second, in three countries with surveys where the results of asset index and consumption expenditures could be compared for the same households, the comparison of OLS and instrumental variables estimates and of bounds from reverse regression suggest that consumption expenditures has considerably more measurement error as a proxy for predicting enrollments than does the asset index. We wish to stress that we do not imply that the asset index is a proxy for current standards of living, nor that it is appropriate for poverty analysis. The fourth column of Table 1 shows how well the first principal component of the asset variables (which is our asset index) "fits" the underlying variables, reporting the proportion of the variation captured. The proportion is remarkably stable, and reasonably high, at between 20 and 30 percent of the variance (from Uganda at .19 to Bolivia at .31)6. There is a generic problem with principal components analysis. While it is relatively easy to interpret the first principal component, an intuitive explanation of the second and 8 higher order components is more problematic. Analysts generally hope for only one factor. In our case, although the first eigen value is relatively high, it is not as high as we would have liked and the value of the second eigen value is also generally above 1, the commonly used cut-off value for "significant" components. This suggests that the "co-movement" of the assets is explained by more than one factor. We have no idea how to interpret this second principal component (especially in a consistent way across countries) and will ignore it for now in an uneasy truce with the data. We do believe, however, that it is not an unreasonable assumption that the "factor" which explains the largest amount of the " co-movement" of the different assets can be interpreted as a household's economic status7. The asset index is calculated separately for each country. Within each country individuals are sorted by the asset index and cutoffs for the bottom 40 percent, the middle 40 percent, and the top 20 percent are derived. Households are then assigned to each of these groups on the basis of their value of the asset index. From here on we will refer to these groups, without further apology, as "poor", "middle" and "rich" .' Since the principal components procedure normnalizes the mean of the index to zero for each country, the value of the index is zero for all countries. Therefore, in comparing the "poor" in Kenya to the "poor" in Turkey or India it is important to keep in mind that the measure is relative and 40 percent of the households are defined to be "poor" in every 6 Since random measurement error will tend to "flatten" the household wealth / enrollment relationship the fact that the fit is similar across countries is comforting as the cross-country comparisons are therefore not likely to be greatly affected by differing degrees of measurement error. 7 Since, by construction, principal components are orthogonal to one another, the "omitted variables" problem of ignoring the second principal component should not be severe. But this rationalization would not be true of omitted variable bias for additional control variables, such as urban residence, which may be correlated with either component. 8 While the cut-off is based on all individuals, the analysis is carried out only for those 15 to 19 so there can be more or less than 40 percent of that cohort in the bottom 40 percent of households. 9 country. Moreover, the gap between rich and poor could easily vary between countries so the Brazilian poor could well be relatively poorer than the Brazilian rich compared to the Egyptian poor relative to the Egyptian rich. C) Attainment Profiles 'We use the data for children aged 15 to 19 to create an "attainment profile" which shows graphically the proportion of individuals that completed each grade or higher (Figure 1). For example this means that the level at grade 1 shows the proportion that ever attended school and completed first grade. One minus this proportion is the proportion that never completed even one year of schooling.9 The slope of the enrollment profile is a simulation of drop-outs.'0 The difference between the proportion that completed grade 5 or higher and those that completed grade 6 or higher is an estimate of the proportion of all children that dropped out between 5th and 6th grade. This is not the usual drop-out rate, as the denominator is all children as opposed to the proportion of those reaching 5th. In the attainment profile figures the drop-out rate is the vertical drop between grades as a proportion of the absolute height. Figure 1 shows the attainment profiles for each of the 35 countries (some with profiles for more than one survey) with the profile of the poor, middle and rich identified. Since much of the paper is an exploration of the interesting results and patterns that emerge from these graphs we'll walk through the interpretation of the graphs by describing the first country, Benin in detail. 9 We are therefore not distinguishing between attending school but never completing even one grade and never having attended school at all. 0 This is a simulation because we are not observing an individual's progression through the school system but a cross section of attaimnents of this cohort. 10 Figure 1 Attainment profiles for ages 15 to 19, by economic group: Western and Central Africa Benin 1993 Burkina Faso 1991-92 I Cameroon 1991 1l 1l- °11 , 0.8 i 0.8 i 0.81 '0.6 00.6 a \0.6 '0.4 0.4 0.4 0.2 0.2- 0.2 _ 0 00 1 2 3 4 5 6 7 8 123456789 1 2 3 4 5 6 7 8 9 Grade I Grade Grade C.A.R. 1994-95 Cote d'lvoire 1994 Ghana 1993 0.8 0.8 0. _ 0.6 0- e , EO.6 0.6 - 0.4 ~~~~~~~04 04- 0.2 >>*>0.2 0.2 0 ~~~ ~ ~~~~0 +9 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 89 1 2 3 4 5 6 7 89 Grade Grade Grade Mali 1995-96 Niger 1992 Senegal 1992-93 . ~~~~I ] _ _ _ _ , 0.8 . 0.8 1 0.8 - .~0.6 - - 0.6 o0.6 - o04 E0.4 04 0.2 - 0.2 0.2 0 0~ 01 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 89 1 2 3 4 5 6 7 89 Grade I Grade Grade * Poorest . Middle A Richest 11 Figure 1 continued Attainment profiles for ages 15 to 19, by economic group: South Asia Bangladesh 1993-94 Bangladesh 1996-97 India 1992-93 1.l i 1 ol 0.8 0.8-- 0.8 'o0.6 - I 0.6 --< e 0.6l 0.6 0.6 0.6 20.4 0.4 0.4 0.2 0.2 0.2 0 0 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade Grade Nepal 1996 Pakistan 1990-91 0.8 -- 0.8 20.6 ~0.6 . - 0.4 0.4 0.2 0.2- - 0 0 I 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade * Poorest u Middle A Richest 12 Figure 1 continued Attainment profiles for ages 15 to 19, by economic group: Central America and Caribbean pominican Republic 1991 prnminican Republic 1996 Guatemala 1995 08 0 8 08 06 ~~~~~~~~ 06 06 0 4 0 4 04 04 2 04 2 04 02 02 02 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade Grade Haiti 1994-95 0.8 0 2 i 2 3 4 5 6 7 8 9 Grade * Poorest * Middle A Richest 13 Figure 1 continued Attainment profiles for ages 15 to 19, by economic group: South America Bolivia 1993-94 Northeast Brazil 1991 Brazil 1996 0 8.81 8 ---0.8 08|D O 10 6 00.6-o0.6-006 -----0- 0. 04-04 U.2 . . ) 1 0.2 0.2 00 0 . o - ! - o - 0 2 , 0 . 2 1 2 3 4 5 6 7 8 9 1 2 3 4 567891 2 3 4 5 6 7 8 9 Grade Grade Grade Colombia 1990 Colombia 1995 i Peru 1991-92 0.8-- 0.8 - -- s 0.6 1< = 0.6 | 408o.6 [ N 0.4 -\<, \; ; I :; 0.4 E . 0.4 0. 0.2 -0.2 -0.2 - - - 0 - 0 _ __ _0 ____ 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 89 1 2 3 4 5 6 7 8 9 Grade Grade Grade Peru 1996 [Northeast Brazil 19961 0.8 0.8 - - 0.6 e 0.6 0.4 0.4 0.2- -- 0.2 . - O-- 0- : 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade * Poorest . Middle A Richest 14 Figure 1 continued Attainment profiles for ages 15 to 19, by economic group: Eastern and Southern Africa Comoros 1996 Kenya 1993 Malawi 1992 - - -. ---------- -. - -.--- ..... --.- ------------- .----------- 0~8 0.8 -0.8- s 0~6 ;0 eg 06- 0g.6 024 02 4 02 0 20- 2 0 2 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade Grade f~~~~~~~~~~~ t.~~~~~~~ I Namlbia 1992 Rwanda 1992 Tanzania 1991-92 ll I~ ~ ' ............................................................---- .r....= 0.8- 08- 0.8 - O S606 \0 6 IC 4 0 404 a. 04 -- - - 0.04- --- - :04 \ 02 - 0.2 0.2 O 0 0- 1 2 3 4 5 6 7 8 9 I 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade Grade Tanzania 1996 Uganda 1995 Zambia 1992 0 8 0.84 1 0.81- 06 0.606 a. 4 0.4a. 04 02 -- - 02 0.2- 0 0 o__ _ _ _ _ _ I 2 3 4 5 6 7 8 9 I 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade Grade * Poorest u Middle A Richest 15 Figure 1 continued Attainment profiles for ages 15 to 19, by economic group: Eastern and Southern Africa continued Zambia 1996-97 Zimbabwe 1994 084 < > 0.8 N i O 0,6 t 04 a 404 02 - 02 0 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade Attainment profiles for ages 15 to 19, by economic group: East Asia and Pacific Indonesia 1991 I Indonesia 1994 Philippines 1993 08- 0.8 08. 0.6 0.6 0.6 a., 04 c_ 0.4 a.0.4 02 0.2 _ 0.2 0 0 I~~~~~ 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 67 89 I 1 2 3 4 5 6 7 8 9 Grade Grade Grade * Poorest * Middle & Richest 16 Figure 1 continued Attainment profiles for ages 15 to 19, by economic group: Middle East, North Africa, Central Asia and Europe Egypti 1992 Egypt 1995-96 Kazakstan 1995 08 ~~~ - -> } 086____+ _0 08 0 6 0 0.6 0 6 0.2 02. 0.2 o O O 0 4 a. 4 0. 04 ~2 02 0.2 0 . 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade Grade Morocco 1992 Turkey 1993 08 4 08 02 02 0 3 0 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 Grade Grade * Poorest . Middle A Richest 17 Figure 2 In Benin (whose attainment profile is reproduced in Benin 1993 Figure 2) only 26 percent of the poor aged 15 to 19 have completed of grade 1 or higher, so 74 percent have either 0.8 .o 0.6 >- . | attended school (or more precisely, not completed even one >,, 0.4 - ~ > < ( year of schooling). Completion of grade 5 or higher is only 0.2 J 7.9 percent and only 0.7 percent complete grade 9. Among 0 the rich, 80 percent complete grade 1 or higher but drop-out 1 2 3 4 5 6 7 8 9 Grade is such that only 54 percent complete grade 5. Even among * Poorest * Middle A Richest the rich only 17 percent complete grade 9. We define two "wealth gaps". First, the wealth gap in the completion of any given grade (which, graphically, is the vertical distance between economic groups). For example, the wealth gap in grade 1 completion is .54 in Benin (.74 for the rich versus .20 for the poor with no schooling), while the wealth gap at grade 5 is .46. Second, the wealth gap in median grade completed. Visually the median is where a horizontal line at .5 would cross the attainment profile hence the gap is, graphically, the horizontal distance between the two groups. The median grade completed of the poor in Benin is zero, while that of the middle group is 2, and that of the rich is 5 years. The wealth gap in attainment is 5 years. There are three main patterns that emerge from these figures and they are discussed in turn. * First, the role of "ever enrollment" versus drop-out in explaining education outcomes, * second, differing wealth gaps and attainment patterns across countries, 18 * and third, the use of the differences in attainment profiles as a diagnostic tool. II) Enrollment and drop-out among the poor Average attainment can be decomposed into two parts: the fraction of children that ever enroll and, conditional on having enrolled, the grade at which children leave school. We do not distinguish between those who never enrolled and those who may have enrolled but did not complete even one year of schooling, and in the following discussion the two descriptions are used interchangeably. A) Patterns of enrollment and dropout There are four patterns of enrollment and drop-out of the poor, which tend to follow regional patterns: * low ever enrollment and high drop-out (Western/Central Africa) * low ever enrollment and low drop-out (South Asia) * high ever enrollment and early high drop-out (Latin America) * high ever enrollment and late (East Africa) or very late drop-out (East Asia and Central Asia/North Africa/Europe). Table 2 presents the proportion of 15 to 19 year olds from the poorest 40 percent who completed (at least) grade 1, grade 5, primary school, and grade 9. Because the lengths of the cycles (primary versus lower secondary --sometimes together called "basic"-- and upper secondary) differ across countries we show the results both for the comparable number of grades (grade 5) and for the comparable cycle (primary)." Grade 9 was chosen as the highest "Presently we are using the UNESCO reporting of the structure of primary and secondary, which may or may not reflect country realities and moreover, may not have been relevant to the situation a decade ago. 19 grade to report because the truncation problem (that children are still in school and we are not observing completed spells of schooling) becomes more severe the higher the grade. In Western/Central Africa only between 4.6 (Mali) and 27 (Cote d'Ivoire) percent of poor children complete grade 5. This is a combination of low ever enrollment and substantial drop-out. For instance, in Benin 74 percent never completed even grade 1 and of those that did, oinly 30 percent complete grade five leaving overall completion of grade 5 at only 8 percerLt. In South Asia the fraction of poor children who didn't complete grade 1 is also very high, around 50 percent, but of those children that do start there is much higher retention. Having begun school between 55 (Bangladesh) and 80 (India) percent stay through to grade 5, but after that drop-out accelerates. The Latin American pattern is one of high initial enrollment, but very steep drop-out among the poor. The situations are strikingly similar especially within South America where almost all poor children start school: the percent never enrolled ranges from 4.2 (Bolivia) to 7.6 (Brazil), but subsequent drop out is high. In all four South American countries the attaimnent profile of the poor drops sharply while the middle and rich children stay in school. In Brazil only 49 percent of those that complete grade 1 go on to complete grade 5. The situation is even bleaker when looking at the entire 6 to 8 years of primary school. Of those that complete grade 1 only 16 percent in Brazil go on to complete primary school, in Bolivia only 30 percent do so. 20 Table 2: Completion and simulated transition proportions for the poorest 40 percent Didn't Completed Completed Completed Country Year Years in Grade I Grade 5 Primary Grade 9 complete grade 5 primary grade 9 primary even grade of those of those of those cycle I who who who completed completed completed 1 1 5 Western and Central Africa Benin 1993 6 0.264 0.079 0.036 0.007 0.736 0.300 0.136 0.091 Burkina Faso 1992-93 6 0.130 0.078 0.064 0.002 0.870 0.599 0.488 0.030 C.A.R. 1991 6 0.514 0.148 0.047 0.003 0.486 0.288 0.091 0.022 Cote d'lvoire 1994-95 6 0.419 0.271 0.158 0.039 0.581 0.646 0.376 0.144 Cameroon 1994 6 0.640 0.446 0.318 0.055 0.360 0.697 0.497 0.124 Ghana 1993 6 0.791 0.694 0.652 0.306 0.209 0.877 0.825 0.441 Mali 1995-96 6 0.119 0.046 0.025 0.001 0.881 0.386 0.208 0.032 Niger 1992 6 0.154 0.113 0.016 0.007 0.846 0.739 0.102 0.062 Rwanda 1992 8 0.730 0.469 0.147 0.037 0.270 0.643 0.201 0.080 Senegal 1992-93 6 0.199 0.143 0.108 0.017 0.801 0.716 0.542 0.118 South Asia Bangladesh 1993-94 5 0.497 0.274 0.274 0.063 0.503 0.551 0.551 0.230 Bangladesh 1996-97 5 0.588 0.356 0.356 0.080 0.412 0.606 0.606 0.223 India 1992-93 5 0.472 0.376 0.376 0.139 0.528 0.797 0.797 0.369 Nepal 1996 5 0.594 0.406 0.406 0.116 0.406 0.683 0.683 0.287 Pakistan 1990-91 5 0.328 0.250 0.250 0.065 0.672 0.761 0.761 0.260 Central America Dominican Republic 1991 6 0.912 0.560 0.427 0.111 0.088 0.614 0.469 0.198 Dominican Republic 1996 6 0.873 0.569 0.466 0.143 0.127 0.652 0.534 0.251 Guatemala 1995 6 0.680 0.236 0.182 0.022 0.320 0.347 0.268 0.091 Haiti 1994-95 6 0.724 0.161 0.099 0.018 0.276 0.223 0.137 0.110 South America Bolivia 1993-94 8 0.958 0.705 0.288 0.195 0.042 0.737 0.301 0.276 Northeast Brazil 1991 8 0.754 0.121 0.009 0.009 0.246 0.160 0.012 0.074 Brazil 1996 8 0.924 0.457 0.150 0.078 0.076 0.494 0.162 0.172 [Northeast Brazil] 1990 8 0.879 0.344 0.093 0.046 0.121 0.392 0.106 0.133 Colombia 1990 5 0.941 0.571 0.571 0.096 0.059 0.607 0.607 0.168 Colombia 1995 5 0.939 0.630 0.630 0.145 0.061 0.671 0.671 0.230 Peru 1991-92 6 0.974 0.813 0.624 0.217 0.026 0.834 0.641 0.267 Peru 1996 6 0.954 0.746 0.496 0.175 0.046 0.781 0.520 0.235 Eastern and Southern Africa Comoros 1996 6 0.576 0.280 0.173 0.014 0.424 0.485 0.300 0.048 Kenya 1993 7 0.963 0.835 0.520 0.102 0.037 0.866 0.540 0.122 Malawi 1992 8 0.666 0.291 0.066 0.011 0.334 0.437 0.099 0.038 Namibia 1992 7 0.918 0.528 0.223 0.047 0.082 0.575 0.243 0.090 Tanzania 1991-92 7 0.821 0.676 0.486 0.004 0.179 0.824 0.592 0.006 Tanzania 1996 7 0.803 0.618 0.376 0.004 0.197 0.770 0.468 0.007 Uganda 1995 7 0.784 0.390 0.130 0.027 0.216 0.498 0.165 0.069 Zambia 1992 7 0.819 0.524 0.255 0.008 0.181 0.640 0.311 0.015 Zambia 1992 7 0.858 0.537 0.254 0.033 0.142 0.626 0.296 0.061 Zimbabwe 1994 7 0.973 0.892 0.696 0.252 0.027 0.917 0.716 0.283 East Asia and Pacific Indonesia 1991 6 0.946 0.778 0.713 0.186 0.054 0.822 0.753 0.240 Indonesia 1994 6 0.959 0.787 0.730 0.190 0.041 0.821 0.761 0.242 Philippines 1993 6 0.973 0.801 0.735 0.320 0.027 0.824 0.755 0.400 Middle East, North Africa, and Europe Egypt 1992 6 0.718 0.639 0.571 0.374 0.282 0.889 0.796 0.586 Egypt 1995-96 6 0.745 0.631 0.572 0.396 0.255 0.846 0.767 0.628 Kazakhstan 1995 4 0.995 0.994 0.995 0.833 0.005 0.999 1.000 0.838 Morocco 1992 6 0.366 0.211 0.106 0.029 0.634 0.576 0.289 0.136 Turkey 1993 5 0.932 0.910 0.910 0.186 0.068 0.976 0.976 0.204 21 One of the most striking findings to emerge from these results is that the level of attainment of the poor in Latin America is lower, not only than East Asia, but even than Eastern/Southern Africa. Grade 5 completion among the poor is 46 percent in Brazil, 57 percent in the Dominican Republic, 63 percent in Colombia, and peaks at 75 percent in Peru. In contrast, it is 89 percent in Zimbabwe, 84 percent in Kenya, 69 percent in Ghana, and 62 percent in Tanzania. The only Eastern/Southern African country with lower attainment for its poor than Brazil is Uganda. The Eastern/Southern African countries have, by and large, relatively low drop out rates in the primary years. So, while the fraction who never enroll is similar to that in the South American countries, the better Eastern/Southern African countries retain higher proportions of the poor. This is especially clear in the flat portions in Figure 1 of the profile for the poor in Kenya, Tanzania, and Zimbabwe (and Ghana, which although it is in West Africa has the attainment patterns of Eastern/Southern Africa). The final pattern is relatively high attainment countries with both high enrollment and high retention through primary and beyond into lower secondary. The patterns differ between Indonesia and Turkey with sharp drop-offs in attainment between primary and secondary and the Philippines and Egypt with less sharp changes across primary to secondary, a difference we return to below. B) Reaching universal attainment and the poor Nearly every country in the world has set a goal to reach universal educational attainmnent through some level: primary, "basic," or even secondary. One important question is what remains to be accomplished to achieve this goal. Examining the attainment profiles in Figure 1 it is clear that in some countries it is practically only the poor who are not completing 22 primary school while in other countries it is both the middle and poorest groups who do not do so. In only a very few countries do the rich not already have universal basic attainment. In Table 3 we report the deficit from universal completion of grade 5 and of primary school. In Figure 1 the shortfall is the vertical distance shown from the horizontal line at universal completion (value of 1) to the level who have completed the grade in question. We then decompose this deficit into that fraction due to shortfalls of the poor, the middle, and the rich children. 2 Again, there are regional patterns in the absolute level of the shortfall and in the fraction of that shortfall due to the different groups. Western/Central Africa has high levels of deficit from grade 5 attainment (around 80 percent) which are nearly evenly distributed across wealth groups. This counter-intuitive result stems partially from the fact that the asset index is defined on a household, not per capita basis. In these cases there are substantially more than 20 percent of children in the top 20 percent of households (the percentage ranges from 23 percent in Comoros to 27 percent in Cote d'Ivoire). In South Asia the attainment deficit is large but its distribution varies. In India and Pakistan the large wealth gaps are revealed in a concentration of the attainment deficit in the poor and middle groups. For India there is a 38 percent shortfall from completion of grade 5, of which 61 percent is due to children from the bottom 40 percent of households while only 4 percent is due to children from the richest 20 percent. 12 That is, for example, the fraction due to the poor is Sp*pp/S where Sp is the shortfall for the poorest group, pp is the proportion of 15 to 19 year olds that are in the poorest group, and S is the total shortfall. 23 Table 3: Shortfall from grade 5 and primary completion, and the proportion of that shortfall due to the shortfall in each economic group Shortfall from grade 5 completion Shortfall from primary completion Total Proportion Proportion Proportion Total Proportion Proportion Proportion Country Year due to due to due to due to due to due to poorest 40 middle 40 richest 20 poorest 40 middle 40 richest 20 percent percent percent percent percent percent Western and Central Africa Benin 1993 0.706 0.443 0.385 0.172 0.797 0.411 0.395 0.193 Burkina Faso 1992-93 0.749 0.443 0.414 0.143 0.775 0.435 0.413 0.152 C.A.R. 1991 0.643 0.478 0.388 0.134 0.796 0.432 0.393 0.175 Cote d'Ivoire 1994-95 0.552 0.434 0.395 0.171 0.674 0.410 0.403 0.187 Cameroon 1994 0.350 0.569 0.365 0.065 0.481 0.509 0.389 0.101 Ghana 1993 0.251 0.440 0.456 0.103 0.292 0.429 0.457 0.112 Mali 1995-96 0.804 0.411 0.422 0.167 0.848 0.398 0.418 0.184 Niger 1992 0.800 0.394 0.456 0.150 0.914 0.383 0.444 0.173 Rwanda 1992 0.455 0.417 0.403 0.180 0.780 0.391 0.421 0.188 Senegal 1992-93 0.640 0.498 0.390 0.112 0.694 0.478 0.398 0.124 South Asia Bangladesh 1993-94 0.524 0.463 0.447 0.090 0.524 0.463 0.447 0.090 Bangladesh 1996-97 0.465 0.500 0.394 0.106 0.465 0.500 0.394 0.106 India 1992-93 0.378 0.606 0.357 0.037 0.378 0.606 0.357 0.037 Nepal 1996 0.512 0.442 0.441 0.117 0.512 0.442 0.441 0.117 Pakistan 1990-91 0.500 0.531 0.403 0.066 0.500 0.531 0.403 0.066 Central America and Caribbean Dominican Republic 1991 0.264 0.636 0.266 0.099 0.360 0.607 0.301 0.092 Dominican Republic 1996 0.254 0.668 0.262 0.070 0.329 0.638 0.286 0.076 Guatemala 1995 0.450 0.589 0.345 0.066 0.508 0.559 0.368 0.073 Haiti 1994-95 0.557 0.494 0.361 0.145 0.663 0.446 0.396 0.158 Bolivia 1993-94 0.145 0.684 0.211 0.106 0.395 0.608 0.301 0.091 South America Northeast Brazil 1991 0.649 0.495 0.337 0.169 0.917 0.395 0.393 0.214 Brazil 1996 0.322 0.698 0.236 0.066 0.649 0.542 0.325 0.133 [Northeast Brazil] 1996 0.517 0.810 0.162 0.028 0.808 0.716 0.238 0.045 Colombia 1990 0.232 0.674 0.234 0.092 0.232 0.674 0.234 0.092 Colombia- 1995 0.191 0.737 0.196 0.068 0.191 0.737 0.196 0.068 Peru 1991-92 0.091 0.741 0.169 0.090 0.189 0.710 0.202 0.088 Peru 1996 0.118 0.756 0.173 0.071 0.244 0.723 0.207 0.071 Eastern arnd Southern Africa Comoros 1996 0.543 0.463 0.406 0.130 0.686 0.421 0.429 0.149 Kenya 1993 0.171 0.388 0.470 0.143 0.467 0.412 0.464 0.124 Malawi 1992 0.584 0.461 0.409 0.130 0.865 0.411 0.408 0.182 Namibia 1992 0.350 0.549 0.408 0.043 0.633 0.500 0.425 0.075 Tanzania 1991-92 0.255 0.477 0.409 0.116 0.430 0.449 0.414 0.138 Tanzania 1996 0.306 0.445 0.442 0.113 0.535 0.415 0.448 0.136 Uganda 1995 0.478 0.531 0.368 0.101 0.756 0.479 0.391 0.130 Zambia 1992 0.256 0.667 0.290 0.042 0.508 0.526 0.366 0.107 Zambia 1992 0.295 0.565 0.400 0.036 0.546 0.492 0.415 0.094 Zimbabwe 1994 0.078 0.551 0.358 0.092 0.216 0.563 0.384 0.053 East Asiza and Pacific Indonesia 1991 0.125 0.622 0.317 0.061 0.166 0.605 0.328 0.067 Indonesia 1994 0.118 0.627 0.289 0.084 0.153 0.615 0.309 0.076 Philippines 1993 0.093 0.738 0.205 0.057 0.127 0.716 0.217 0.066 Middle East, North Africa, and Europe Egypt 1992 0.214 0.688 0.248 0.064 0.272 0.642 0.284 0.074 Egypt 1995-96 0.220 0.673 0.258 0.069 0.265 0.646 0.277 0.077 Kazakhstan 1995 0.007 0.396 0.568 0.037 0.007 0.356 0.605 0.039 Morocce 1992 0.495 0.599 0.327 0.073 0.627 0.536 0.372 0.091 Turkey 1993 0.068 0.551 0.368 0.080 0.068 0.551 0.368 0.080 24 Consistent with the previous observations, the large drop-out rates of the poor in Latin America are revealed in the large proportions of the deficit that is due to the poor. In South America the fraction that do not complete grade 5 is between 12 (Peru) and 32 (Brazil) percent but over 70 percent of that shortfall is due to the poor. The attainment deficit problem for these countries is essentially keeping the poor in school. The Eastern/Southern African countries are again a contrast with those in both Western/Central Africa and Latin America. The shortfalls from grade 5 completion are much lower than those in Western/Central Africa or South Asia and somewhat higher than those in South America, but the distribution of the shortfall is more even. This is true especially when looking at the entire primary school cycle, where the fraction due to the poorest and middle groups is roughly equal. Finally, in East Asia the gaps are smaller, but concentrated among the poor. In the Philippines there is only a 13 percent shortfall from universal primary completion, of which 72 percent is due to the shortfall of the poor. III) "Wealth gaps" across countries The second prominent feature of the country profiles in Figure 1 is the uniform ranking of the rich, middle and poorest groups in terms of educational attainment. As discussed above there are two ways to define a wealth gap. First, the difference in the proportion of each group who complete any given grade. Second, the difference in the median attainment of rich and poor groups. 25 Table 4: Wealth gaps in attainment Wealth gaps (rich-poor) in the proportion Median grade completed who completed Wealth Country Year Grade I Grade 5 Primary Grade 9 Bottom 40 Middle 40 Top gap (top - 20 bottom) Western and Central Africa Benin 1993 0.532 0.462 0.383 0.162 0 2 5 5 Burkina Faso 1992-93 0.556 0.514 0.490 0.200 0 0 6 6 C.A.R. 1991 0.354 0.499 0.386 0.073 1 3 5 4 Cote d'lvoire 1994-95 0.353 0.389 0.387 0.233 0 4 6 6 Cameroon 1994 0.316 0.457 0.474 0.358 4 6 8 4 Ghana 1993 0.150 0.197 0.210 0.273 7 7 9 2 Mali 1995-96 0.476 0.436 0.374 0.041 0 0 4 4 Niger 1992 0.388 0.355 0.281 0.086 0 0 4 4 Rwanda 1992 0.117 0.180 0.224 0.155 4 5 6 2 Senegal 1992-93 0.576 0.550 0.522 0.216 0 0 6 6 South Asia Bangladesh 1993-94 0.386 0.520 0.520 0.384 0 4 8 8 Bangladesh 1996-97 0.296 0.431 0.431 0.407 2 5 8 6 India 1992-93 0.483 0.556 0.556 0.592 0 7 10 10 Nepal 1996 0.233 0.338 0.338 0.314 3 3 8 5 Pakistan 1990-91 0.570 0.602 0.602 0.487 0 5 9 9 Central America and Caribbean Dominican Republic 1991 0.066 0.316 0.415 0.495 5 8 9 4 Dominican Republic 1996 0.118 0.348 0.417 0.498 5 8 10 5 Guatemala 1995 0.279 0.635 0.656 0.489 2 6 9 7 Haiti 1994-95 0.208 0.529 0.498 0.291 2 5 6 4 South America Bolivia 1993-94 0.032 0.232 0.565 0.573 6 9 10 4 Northeast Brazil 1991 0.078 0.435 0.199 0.199 2 4 6 4 Brazil 1996 0.068 0.438 0.426 0.302 4 7 8 4 [Northeast Brazil] 1996 0.089 0.421 0.315 0.199 4 6 7 3 Colombia 1990 0.039 0.330 0.330 0.353 5 7 8 3 Colombia 1995 0.049 0.311 0.311 0.429 5 8 9 4 Peru 1991-92 0.017 0.152 0.304 0.404 6 9 9 3 Peru 1996 0.037 0.219 0.430 0.432 5 8 9 4 Eastern end Southern Africa Comoros 1996 0.331 0.414 0.382 0.169 2 4 6 4 Kenya 1993 -0.002 0.040 0.182 0.213 7 6 8 1 Malawi 1992 0.223 0.372 0.238 0.092 2 3 6 4 Namibia 1992 0.052 0.387 0.509 0.345 5 6 8 3 Tanzania 1991-92 0.128 0.197 0.259 0.149 6 7 7 1 Tanzania 1996 0.160 0.238 0.321 0.102 5 6 7 2 Uganda 1995 0.147 0.377 0.396 0.219 4 5 7 3 Zambia 1992 0.166 0.430 0.515 0.160 5 7 7 2 Zambia 1992 0.137 0.418 0.526 0.315 5 6 8 3 Zimbabwe 1994 0.015 0.071 0.245 0.461 7 8 9 2 East Asia and Pacific Indonesia 1991 0.046 0.190 0.239 0.475 6 8 9 3 Indonesia 1994 0.036 0.173 0.224 0.501 6 8 9 3 Philippines 1993 0.024 0.177 0.230 0.424 7 9 10 3 Middle E]ast, North Africa, and Europe Egypt 1992 0.245 0.299 0.337 0.410 7 9 10 3 Egypt 1995-96 0.233 0.304 0.342 0.401 7 10 11 4 Kazakhstan 1995 0.004 0.005 0.004 0.082 10 10 10 0 Morocco 1992 0.548 0.622 0.630 0.390 0 5 8 8 Turkey 1993 0.054 0.064 0.064 0.332 5 6 9 4 26 Table 4 reports the gap in the proportion of the 15 to 19 year old cohort who have completed grade 1, grade 5, primary school, and grade 9. In Western/Central-Africa there are large gaps at the primary level but by grade 9 attainment has fallen for the rich so the wealth gap closes. In South Asia the wealth gap starts large and stays large ranging from .34 (Nepal) to .60 (Pakistan) for primary school completion and from .31 (Nepal) to .59 (India) for grade 9 completion. In South America the wealth gap is less than .10 at grade 1, but gets progressively larger. For example, by the end of primary school the wealth gap in completion has reached .43 in Brazil, while in Bolivia by grade 9 it has reached .57. The wealth gaps in the Eastern/Southern African countries are relatively small, even through primary completion (except for Zambia and Uganda). The second way of defining the wealth gap highlights striking differences across countries. In the attainment profiles, the difference in median grade completed is the horizontal gap at .5 (i.e. half the population).13 Figure 3 shows the median attainment of rich and poor for each country. The last four columns of Table 4 report the median grade attainment of the poor, middle and rich, as well as the difference between the rich and the poor for each country. Perhaps not surprisingly, as countries move from low to high levels of attainment, the gap starts out high, grows, peaks, and then falls when the entire population enrolls and stays in school. '3 This is calculated from the data and is not truncated at grade 9 as in the figures. This does imply that the differences are, if anything larger due to upper censoring of those still enrolled. 27 Grade CDow J _ 3 w rs Oso 00 0 o Niger 1992 Mali 1995-96 Benin 1993 Ls i Burkina Faso 1992-93 Os Cote dilvoire 1994 .I qi Senegal 1992-93 (I, MoM ocF o 1992 _ ____oo _ _ ] _ O_ o Pakistan 1990-91 -* India 1992-93 0. C-AR. 1994-95 _o Bangladesh 1993-94 P. Northeast Brazil 1991 Haiti 1994-95 _ Comoros 1996 Malawi 1992 Guatemala 1995 o 0D Bangladesh 1996-97 _. Nepal 199611 111 Rwanda 1992 =w Uganda 19 Cameroon 1991 _ _ _ _ iI = X Brazil 1996 Tanzania 1996 __ Zambia 1992 _ Zambia 1996-97 1 Colombia 1990 _ _ CP Namibia 1992 CD Dominican Republic 1991 o Dominican Republic 1996 _| CD Colombia 1995 M 0. Peru 1996 _ - Turkel 1993 _ _ _ _- __ Tanzania 1991-92 = C"o Ken;a 1993 -- - - - Peru 1991-92 _. Indonesia 1991 w Indonesia 1994 _ Bolivia 1993-94 __ __ ___ 0~~~~~~~~~~~~~~~~~~~~~~ Q ~~Zimbabwe 19 Philippines 1993 __ Egypt 1992 ___ Egypt 1995-96 __ Kazakstan 1995 Co In Western/Central Africa the median grade completed of the bottom 40 percent is zero, as less than half of the poor in these countries ever finish even one year of schooling. However, since the rich do not achieve very high levels of schooling either the wealth gap ranges from 4 to 6 years. The wealth gap is the highest in the world in South Asia where the poor are not going to, nor staying in, school. The median grade completed is zero in all countries but Nepal (3 years) and Bangladesh in 1996-97 (2 years). However, the richer groups in these countries have high levels of attainment. India has the world's largest gap of 10 years with the poor having median grade completed of zero, while for the rich attainment is 10 years. This is followed closely by Pakistan at 9 years, and Bangladesh in 1996-97 at 5. The Latin American countries have smaller, but for their average attainment, enormous wealth gaps. Haiti has a pattern similar to those in Western/Central Africa with median grade completed of 2 for the poor and only 6 for the rich, while Guatemala has a pattern like that in South Asia with a gap of 7 (2 for the poor versus 9 for the rich). The inequality in attainment in South America results in a wealth gap of 4 years in all four countries with the median grade completed ranging from 4 to 6 for the poor, and from 8 to 10 for the rich. Again there is the striking comparison between Latin America and the much poorer countries in Eastern/Southern Africa. The bottom 40 percent in Eastern/Southern Africa have considerably higher educational attainment than the bottom 40 percent in Central or South America. The median grade completed in Kenya, Ghana, and Zimbabwe is 7 in contrast to 4 in Brazil, and 5 in Colombia and Peru. The Eastern/Southern African pattern of high initial enrollment and high retention of all groups through primary leads to low wealth gaps ranging from 1 (Kenya) to 3 (Uganda and Zambia). The wealth gap is equal to 3 for the two East Asian countries. This is due not to especially low attainment for the poorest and middle groups, but rather to higher levels of attainment of the richest group. The wealth gap in median grade completed is 4 in Egypt and Turkey, again due largely to the high attainment of the richest group. IV) ALttainment profiles as a diagnostic The attainment profiles are also useful as a diagnostic as to where key concerns in the systern are. When the issue of increasing enrollment rates or educational attainment is discussed there is often a tendency to talk about "access" to schooling, where access is narrowly defined as the physical availability of schools. This was almost certainly the key issue some years ago when there just were not enough schools or teachers to go around. However, it is increasingly unlikely that in many countries the physical presence or absence of schools is a major constraint on expanding enrollment and attainment, particularly of the poor. Many analysts have concluded that even in very poor countries improving the quality of schooling is now the critical dimension for expanding enrollments. The figures presented here provide three stylized facts that are consistent with this conjecture, two from the primary level and o:ne from the transition from primary to secondary. A) Primary First, if a child went to school and then stopped going it is very likely that he or she could have continued to go to school. The first column of Table 5 presents the proportion of the shortfall in primary completion that is due to drop-out (i.e. the ratio of the difference between grade 1 and primary completion and the shortfall in primary completion). 30 Table 5: Attainment and dropout rates Shortfall Proportion who completed Bottom 40 percent only, due to at least grade I simulated dropout rates dropout, in the last between in the Country Year bottom 40 All Top 20 rural year of primary second year percent male primary secondary of secondary ~~~~~~............... ..... .. .- ---.......... ......... ............. ........ -ec n m a le ..pr . .. ... ....i.....a... Western and Central Africa Benin 1993 0.237 0.510 0.777 0.546 0.250 0.229 Burkina Faso 1992-93 0.071 0.323 0.603 0.186 0.735 0.651 C.A.R. 1991 0.490 0.686 0.946 0.684 0.629 0.506 Cote dlvoire 1994-95 0.310 0.602 0.830 0.417 0.488 0.248 Cameroon 1994 0.472 0.787 0.964 0.287 0.417 0.458 Ghana 1993 0.399 0.830 1.000 0.060 0.128 0.165 Mali 1995-96 0.097 0.294 0.484 0.461 0.509 0.259 Niger 1992 0.140 0.250 0.412 0.862 0.169 0.037 Rwanda 1992 0.683 0.786 0.826 0.444 0.746 0.538 Senegal 1992-93 0.102 0.434 0.417 0.243 0.750 0.156 South Asia Bangladesh 1993-94 0.307 0.666 0.881 0.215 0.437 0.231 Bangladesh 1996-97 0.360 0.725 0.890 0.185 0.341 0.240 India 1992-93 0.153 0.695 0.956 0.100 0.180 0.145 Nepal 1996 0.317 0.632 0.894 0.173 0.206 0.212 Pakistan 1990-91 0.104 0.577 0.936 0.111 0.331 0.192 Central America and Caribbean Dominican Republic 1991 0.846 0.948 1.000 0.237 0.308 0.317 Dominican Republic 1996 0.761 0.933 1.000 0.181 0.256 0.308 Guatemala 1995 0.609 0.835 0.986 0.228 0.716 0.342 Haiti 1994-95 0.694 0.848 0.824 0.387 0.482 0.397 South America Bolivia 1993-94 0.940 0.981 1.000 0.239 0.323 0.449 Northeast Brazil 1991 0.752 0.813 0.661 0.716 0.000 0.493 Brazil 1996 0.911 0.962 1.000 0.321 0.476 0.496 [Northeast Brazil] 1996 0.866 0.911 1.000 0.367 0.507 0.474 Colombia 1990 0.863 0.967 0.741 0.168 0.407 0.270 Colombia 1995 0.836 0.970 0.972 0.147 0.379 0.200 Peru 1991-92 0.931 0.986 1.000 0.232 0.273 0.253 Peru 1996 0.909 0.977 1.000 0.334 0.195 0.288 Eastern and Southern Africa Comoros 1996 0.488 0.731 0.917 0.382 0.529 0.542 Kenya 1993 0.923 0.956 0.943 0.269 0.461 0.637 Malawi 1992 0.643 0.736 0.895 0.487 0.834 0.529 Namibia 1992 0.894 0.933 0.773 0.390 0.430 0.628 Tanzania 1991-92 0.652 0.872 0.921 0.189 0.985 0.415 Tanzania 1996 0.685 0.865 0.952 0.219 0.980 0.424 Uganda 1995 0.751 0.847 0.940 0.481 0.615 0.462 Zambia 1992 0.757 0.909 0.925 0.367 0.856 0.791 Zambia 1992 0.809 0.911 1.000 0.352 0.703 0.567 Zimbabwe 1994 0.911 0.978 0.890 0.153 0.457 0.333 East Asia and Pacific Indonesia 1991 0.813 0.974 0.988 0.083 0.602 0.137 Indonesia 1994 0.847 0.979 0.996 0.073 0.605 0.144 Philippines 1993 0.898 0.987 0.996 0.083 0.240 0.177 Middle East, North Africa, and Europe Egypt 1992 0.342 0.842 0.984 0.105 0.082 0.079 Egypt 1995-96 0.406 0.866 0.988 0.093 0.100 0.076 Kazakhstan 1995 0.000 0.994 1.000 0.000 0.001 0.000 Morocco 1992 0.291 0.652 1.000 0.499 0.295 0.262 Turkey 1993 0.249 0.951 1.000 0.011 0.671 0.098 31 The striking results here are the high numbers for Latin America (60 to 94 percent), EasternmSouthern Africa (excluding Ghana, 75 to 92 percent) and East Asia (85 and 90 percent),. In these countries, the main explanation for why children do not complete primary school is not that they don't start school, it is the fact that they drop-out. In the Western/Central African and South Asian countries, dropout explains less than half the shortfall from universal primary education for the poor, with values ranging from 10 percent in India to 49 percent in Comoros (which although it lies off the Eastern coast has the attainment patterns of Western/Central Africa). ,Of course this approach can only address the question of the physical availability of schools, not true access to education. In particular it is possible that children did attend first grade, but in classes of 100 or more, with no materials, indifferent (or worse) teaching, and deteriorating buildings. Not surprisingly there will be high drop-out, due not to physical availability but to access to an education, which when properly defined, includes quality. A second approach to the impact of school availability would be to look directly at the relationship between the presence of schools and enrollment. For example, using the DHS data from India, Filmer and Pritchett (1998b) found in state by state regressions that there was only a weak relationship between the availability of schools and enrollment rates. In this paper we don't replicate the analysis country by country but, as a heuristic indication, we report the proportion of "Rich Rural Males" (RRM) who have completed at least grade 1. Since poverty is not completely regionally concentrated this group is likely to suffer from a similar lack of physical access to schools as less socially favored groups (the poor and females). If RRM have high enrollment this provides some evidence on the degree to which other groups are falling short due to their status, not school availability. 32 Table 5 shows large gaps in the low enrollment countries between rich rural males and the average enrollment. In Pakistan RRM enrollment is 94 percent while the average is 58 percent (a 36 percentage point gap), in India RRM enrollment is 96 percent while average is 69 percent (a 37 percentage point gap), in Cote d'Ivoire RRM enrollment is 83 percent versus 60 percent (a 23 percentage point gap), and in C.A.R RRM enrollment is 95 percent versus an average of 69 percent (a 36 percentage point gap). Expanding enrollment of the average to that of RRM would nearly eliminate the proportion of children who completed less than 1 year of schooling in most countries, except for the very lowest performers such as Mali or Benin. B) Transition to secondary The third point about availability is that the attainment profiles do not suggest that the lack of availability of secondary schools, or rationing of secondary places, plays a large role in most countries, although it is a central phenomenon in some. Table 5 reports the simulated drop-out rate of the poor between the second-to-last and the last year of primary, between the end of primary and the end of the first year of secondary, and between the first and second years of secondary.'4 If secondary school places were rationed (either officially by an exam or in practice by the lack of facilities) one would expect to see the drop-out across the transition between primary and secondary to be much larger than either before or after the transition point. This is indeed the pattern in some countries, identifiable graphically by a steep drop at the end of the primary cycle. In Turkey the drop-out rate is 1.1 percent before the transition, '4 Again, simulated as we are not observing the dropout behavior on an individual child, rather this is the value implied from the cross section of 15 to 19 year olds. For example, in a country in which the primary cycle is 6 years, the simulated drop-out rate in the last year of primary is the ratio of the completion of grade 5 to the completion of grade 6, the rate between primary and secondary cycles is the ratio of the completion of grade 6 to the 33 67 percent across the transition, and 9.8 percent after it. In Tanzania the drop-out rate is 22 percent the year before, but 98 percent across the transition as almost no poor child made it to secondary'5. High drop-out rates across the primary transition are also prominent in Indonesia and Guatemala. In nearly ever other country, however, drop-out is noticeably higher, but not dramatically so, across the transition than in the year before or after. Most of the graphs are characterized by very smooth slopes that make it difficult to distinguish visually which is the transition year. For example, in Nepal the drop-out rate is 17 percent, 21 percent and 21 percent in the three years. In the Philippines the drop-out rate is 8.3 before the end of primary, increases to 24 percent across the transition, and then falls slightly to 18 percent. This pattern of a higher rate in the transition followed by a high, but lower, subsequent rate is true of Haiti, C.A.R., Zambia., and Uganda. In the Dominican Republic the drop-out rate is 18 percent, 26 percent and 31 percent, so drop-out is higher after the first year of secondary than in the transition before. This pattern true of Bolivia and Brazil as well. The analysis reveals the importance of the entire attainment profile. Looking only at average enrollment between primary and secondary one might be tempted to conclude that the large gap indicated the problem was across the transition. However, examining the profile might reveal that drop outs within primary imply that the "excess" of primary school leavers over secondary entrants is quite low. completion of grade 7, and the rate in the second year of secondary is the ratio between the completion of grade 7 and the completion of grade 8. 5 Keep in mind these are 15-19 year olds in the year of the survey and hence reflects the situation some years prior. 34 Figure 4 Figure 4 illustrates the Brazil 1993 Indonesia 1994 point by showing the > 0.8 < l 1 0.8 j average profiles for the ao 0.6 ----- ----- . -. . 0.6 tpoorest 40 percent in '; 0.4- ... ---- 0.4 - -;_ Brazil and Indonesia. In 0.2 - 0.2 - Brazil, dropout across 0o -0 l the transition may be 1 2 3 4 5 6 7 89 1 2 3 4 5 6 7 8 9 Grade Grade high (48 percent), but it _ Poorest n Middle A Richest is high all throughout the primary school years. In Indonesia, dropout is ten percentage points higher across the transition (61 percent) but the underlying profile is vastly different as there is a very small amount of dropout within the primary school years and virtually all of the drop-off is across the transition. Of course this use of the attaimnent profile is diagnostic and can only point to issues for more detailed examination. For instance Lavy (1997) has shown that the lack of secondary school facilities can influence drop-out at the primary levels even before the end of primary by lowering the expected return to additional primary years. Therefore, we cannot conclude from high primary drop-out that the physical lack of secondary facilities is not an important issue. 35 Conclusion In this paper we have documented striking cross-country patterns in education enrollment and attainment in 35 countries. While many others have examined the differences in enrollment rate behavior between the rich and poor, a major advantage of this analysis is that the data are comparable. The attainment data are derived consistently and, while the levels of the asset index are not directly comparable across countries, they are derived using an identical methodology. There are two overall conclusions that emerges from these results. First, many (if not most) countries the bulk of the deficit from universal enrollment up to primary (or basic) comes from the poor. The achievement of higher levels of enrollment for this group is an exercise in social inclusion, reaching out and bringing the poorest into what is already the norm for the rich and, in many cases, the middle class. Second, the evidence suggests that, except in the very poorest settings, the key to closing wealth gaps in enrollment and attainment will require actions which raise the demand for schooling of the poor. Raising the quality of schooling received at the primary level is likely to be the key ingredient to attract and retain poor children in school. 36 References Ahuja, Vinod and Deon Filmer, 1996 "Educational Attainment in Developing Countries: New Estimates and Projections Disaggregated by Gender". Journal of Educational Planning and Administration Vol X(3):229-254. Barro, Robert and Jong-Wha Lee, 1993. "International Comparisons of Educational Attainment," Journal of Monetary Economics, 32:363-394. Behrman, Jere R. and James C. Knowles, 1997. "How Strongly is Child Schooling Associated with Household Income?" University of Pennsylvania and Abt Associates. Mimeo. Castro-Leal, Florencio, Julia Dayton, Lionel Demery, and Kalpana Mehra, 1997, "Public social spending in Africa: do the poor benefit?" mimeo, PRMPO, The World Bank. Dubey, Ashutosh and Elizabeth King. 1994. "A New Cross-Country Education Stock Series Differentiated by Age and Sex", mimeo, The World Bank Filmer, Deon and Lant Pritchett, 1998a. "Estimating wealth effects without income of expenditure data -- or tears: Educational enrollment in India," mimeo, DECRG, The World Bank. Washington, DC. Filmer, Deon and Lant Pritchett, 1998b, "Determinants of Education Enrollment in India: Child, Household, Village and State Effects," mimeo, DECRG, The World Bank. Washington, DC. Hammer, Jeffrey, 1998. "Health Outcomes across wealth groups in Brazil and India," mimeo, DECRG, The World Bank. Washington, DC. Lavy, Victor. 1997. "School supply constraints and children's educational outcomes in rural Ghana," Journal of Development Economics, 51:[291]-314. Montgomery, Mark, Kathleen Burke, Edmundo Paredes, and Salman Zaidi, 1997. "Measuring Living Standards with DHS Data: Any Reason to Worry?" mimeo, Research Division, The Population Council. New York, NY. Nehru, Vikram, Eric Swanson and Ashutosh Dubey, 1993. "New Database on Human Capital Stock in Developing and Industrial Countries: Sources, Methodology, and Results," Journal of Development Economics, 46:379-401 Patrinos, Harry Anthony. 1997. "Differences in Education and Earnings Across Ethnic Groups in Guatemala," Quarterly Review of Economics and Finance 37:809-821. 37 Table A-1: Proportion who have completed grade or higher Bottom 40 percent Middle 40 percent lop 20 percent Year Grade I Grade 5 Primary Grade 9 Grade I Grade 5 Primary Grade 9 Grade I Grade 5 Primnary Grade 9 W estern and Central A frica ~ ~ ~ ......... ...... . . --- - -- ----........................ .................... ..... ...- ....... Bcnin 1993 0.264 0.079 0.036 0.007 0.530 0.312 0.203 0.043 0.797 0.541 0,419 0.169 Burkina Fasu 1992-93 0.130 0.00788 0.064 0.002 0.2,52 0.I7 80. 1 50 0.023 0.686 0.5920.553 0,203 C.AR. 1991 0.514 0.148 0.047 0.003 0.729 0.367 0.206 0.037 0.868 0.647 0.433 0.076 Cote dIlvoire 1994-95 0.419 0.271 0.158 0.039 0.635 0.447 0.311 0.115 0.772 0.660 0.545 0.272 Camneroon 1994 0.640 0.446 0,318 0.055 0.819 0.685 0.540 0.156 0.956 0.903 0.792 0,413 Ghana 1993 0.791 0.694 0.652 0.306 0.799 0.714 0.666 0.319 0.941 0.891 0.862 0.579 Mali 1995-96 0.119 0.046 0.025 0.001 0.248 0.139 0.100 0.006 0.595 0.482 0.399 0.043 Niger 1992 0.154 0.113 0.016 0.007 0.175 0.129 0.032 0.011 0.541 0.468 0.297 0.093 Rwanda 1992 0.730 0.469 0.147 0.037 0.802 0.552 0.199 0.067 0.846 0.649 0.371 0.192 Senegal 1992-93 0.199 0.143 0.108 0.017 0.454 0.367 0.301 0.057 0.776 0,693 0.630 0.232 South Asia Bangladesh 1993-94 0.497 0.274 0.274 0.063 0.682 0.464 0.464 0.148 0.883 0.794 0.794 0.447 Bangladesh 1996-97 0.588 0.356 0.356 0.080 0.755 0.550 0.550 0.i74 0.885 0.788 0.788 0.487 India 1992-93 0.472 0.376 0.376 0.139 0.761 0.684 0.684 0.363 0.954 0.932 0.932 0.730 Nepal 1996 0.594 0.406 0.406 0.116 0.551 0.414 0.414 0.139 0.827 0.743 0.743 0.430 Pakistan 1990-91 0.328 0.250 0.250 0.065 0.614 0.522 0.522 0.209 0.898 0.852 0.852 0.552 Central America and Caribbean Dominican Republic 1991 0.912 0.560 0.427 0.111 0.967 0.828 0.734 0.399 0.978 0.876 0.843 0.606 Dominican Republic 1996 0.873 0.569 0.466 0.143 0.962 0.831 0.760 0.402 0.991 0.917 0.883 0.641 Guatemala 1995 0.680 0.236 0.182 0.022 0.894 0.632 0.557 0.181 0.959 0.871 0.839 0.511 Haiti 1994-95 0.724 0.161 0.099 0.018 0.894 0.512 0.363 0.105 0.932 0.690 0.597 0.308 South America Bolivia 1993-94 0.958 0.705 0.288 0.195 0.995 0.927 0.716 0.588 0.989 0.937 0.853 0.768 Northeast Brazil 1991 0.754 0.121 0.009 0.009 0.855 0.438 0.074 0.074 0.832 0.556 0.208 0.208 Brazil 1996 0.924 0.457 0.150 0.078 0.986 0.801 0.449 0.277 0.992 0.895 0.576 0.381 [Northeast Brazil] 1996 0.879 0.344 0.093 0.046 0.968 0.721 0.358 0.223 0.967 0.766 0.408 0.245 Colombia 1990 0.941 0.571 0,571 0.096 0.983 0.870 0.870 0.321 0.980 0.902 0.902 0.449 Colombia 1995 0.939 0.630 0.630 0.145 0.989 0.906 0.906 0.443 0.989 0.942 0.942 0.574 Peru 1991-92 0.974 0.813 0.624 0.217 0.993 0.963 0.907 0.502 0.991 0.965 0.928 0.621 Peru 1996 0.954 0.746 0.496 0.175 0.988 0.951 0.879 0.462 0.991 0.964 0.926 0.608 Eastern and Southern Africa Comoros 1996 0.576 0.280 0.173 0.014 0.762 0.475 0.299 0.059 0.907 0.694 0.555 0.183 Kenya 1993 0.963 0.835 0.520 0.102 0.946 0.801 0.463 0.103 0.961 0.85 0.702 0.315 Malawi 1992 0.666 0.291 0.066 0.011 0.715 0.393 0.106 0.014 0.889 0.664 0.304 0. 103 Natnibia 1992 0.918 0.528 0.223 0.047 0.933 0.657 0.355 0.122 0.970 0.915 0.732 0.392 Tanzania 1991-92 0.821 0.676 0.486 0.004 0.876 0.736 0.548 0.023 0.949 0.873 0.744 0.153 Tanzania 1996 0.803 0.618 0.376 0.004 0.861 0.665 0.406 0.014 0.964 0.857 0.697 0.107 Uganda 1995 0.784 0.390 0.130 0.027 0.871 0.533 0.215 0.060 0.930 0.767 0.525 0.246 Zamnbia 1992 0.819 0.524 0.255 0.008 0.946 0.816 0.540 0.038 0.985 0.954 0.770 0.168 Zambia 1992 0.858 0.537 0.254 0.033 0.909 0.711 0.445 0.100 0.995 0.955 0.779 0.347 Zimbabwe 1994 0.973 0.892 0.696 0.252 0.979 0.930 0.794 0.397 0.988 0.963 0.941 0.713 East Asia and Pacific Indonesia 1991 0.946 0.778 0.713 0.186 0.987 0.905 0.869 0.423 0.993 0.967 0.952 0.661 Indonesia 1994 0.959 0.787 0.730 0.190 0.987 0.916 0.884 0.445 0.995 0.960 0.953 0.691 Philippines 1993 0.973 0.801 0.735 0.320 0.992 0.954 0.933 0.618 0.997 0.978 0.965 0.744 Middle East, North Africa, and Europe Egypt 1992 0.718 0.639 0.571 0.374 0.907 0.857 0.792 0.565 0.963 0.938 0.909 0.784 Egypt 1995-96 0.745 0.631 0.572 0.396 0.927 0.845 0.799 0.628 0.978 0.935 0.913 0.798 Kazakhstan 1995 0.995 0.994 0.995 0.833 0.991 0.989 0.989 0.883 0.999 0.999 0.999 0.915 Morocco 1992 0.366 0.211 0,106 0.029 0.777 0.601 0.426 0.164 0.914 0.833 0.735 0.419 Turkey 1993 0.932 0.910 0.910 0.186 0.953 0.933 0.933 0.372 0.986 0.974 0.974 0.5 17 38 Policy Research Working Paper Series Contact Title Author Date for paper WPS1965 Manufacturing Firms in Developing James Tybout August 1998 L. Tabada Countries: How Well Do They Do, 36869 and Why WPS196E Sulfur Dioxide Control by Electric Curtis Carlson August 1998 T. Tourougui Utilities: What Are the Gains from Dallas Burtraw 87431 Trade? Maureen Cropper Karen L. Palmer WPS196J Agriculture and the Macroeconomy Maurice Schiff August 1998 A. Vald6s Alberto Valdes 35491 WPS1968 The Economics and Law of Rent Kaushik Basu August 1998 M. Mason Control Patrick Emerson 30809 WPS1 969 Protecting the Poor in Vietnam's Dominique van de Walle September 1998 C. Bernardo Emerging Market Economy 31148 WPS197IJ Trade Liberalization and Endogenous Thomas F. Rutherford September 1998 L. Tabada Growth in a Small Open Economy: David G. Tarr 36896 A Quantitative Assessment WPS1971 Promoting Better Logging Practices Marco Boscolo September 1998 T. Tourougui in Tropical Forests Jeffrey R. Vincent 87431 WPS1972 Why Privatize? The Case of Geroge R. G. Clarke September 1998 P. Sintim-Aboagye Argentina's Public Provincial Banks Robert Cull 38526 WPS1973 The Economic Analysis of Sector Sethaput Suthiwart- September 1998 C. Bernardo Investment Programs Narueput 31148 WPS1974 Volatility and the Welfare Costs Pierre-Richard Agenor September 1998 S. King-Watson of Financial Market Integration Joshua Aizenman 33730 WPS1 975 Acting Globaily While Thinking Gunnar S. Eskeland September 1998 C. Bernardo Locally: Is the Global Environment Jian Xie 31148 Protected by Transport Emission Control Programs WPS1976 Capital Flows to Central and Stijn Claessens September 1998 R. Vo Eastern Europe and the Former Daniel Oks 33722 Soviet Union Rossana Polastri WPS1977 Economic Reforms in Egypt: Rania A. Al-Mashat September 1998 S. Dy Emerging Patterns and Their David A. Grigorian 32544 Possible Implications WPS19'78 Behavioral Responses to Risk Jyotsna Jalan September 1998 P. Sader in Rural China Martin Ravallion 33902 Policy Research Working Paper Series Contact Title Author Date for paper WPS1979 Banking on Crises: Expensive Gerard Caprio, Jr. September 1998 P. Sintim-Aboagye Lessons from Recent Financial 38526 Crises