Policy Research Working Paper 10285 Growing Up Together Sibling Correlation, Parental Influence, and Intergenerational Educational Mobility in Developing Countries Md. Nazmul Ahsan M. Shahe Emran Hanchen Jiang Qingyang Han Forhad Shilpi Development Economics Development Research Group January 2023 Policy Research Working Paper 10285 Abstract This paper presents credible and comparable evidence on and Pacific), while others remained trapped in stagnancy intergenerational educational mobility in 53 developing (South Asia and Sub-Saharan Africa). The only region that countries using sibling correlation as a measure, and data experienced monotonically increasing sibling correlation is from 230 waves of Demographic and Health Surveys. It is the Middle East and North Africa. The recent approach of the first paper to provide estimates of sibling correlation Bingley and Cappellari (2019) is used to estimate the share in schooling for a large number of developing countries of sibling correlation due to intergenerational transmission. using high quality standardized data. Sibling correlation The estimates show that when the homogeneity and inde- is an omnibus measure of mobility as it captures observed pendence assumptions implicit in the standard model of and unobserved family and neighborhood factors shared by intergenerational transmission are relaxed, the estimated siblings when growing up together. The estimates suggest share is much larger. In the sample of countries, on average that sibling correlation in schooling in developing coun- 74 percent of sibling correlation can be attributed to inter- tries is much higher (average 0.59) than that in developed generational transmission, while there are some countries countries (average 0.41). There is substantial spatial hetero- where the share is more than 80 percent (most in Sub-Sa- geneity across regions, with Latin America and Caribbean haran Africa). This suggests a dominant role for parents in having the highest (0.65) and Europe and Central Asia determining the educational opportunities of their children. the lowest (0.48) estimates. Country level heterogeneity Evidence on the evolution of the intergenerational share, within a region is more pronounced. The evolution of sib- however, suggests a declining importance of the intergener- ling correlation suggests a variety of mobility experiences, ational transmission component in many countries, but the with some regions registering a monotonically declining pattern is diverse. In some cases, the trend in the intergen- trend from the 1970s birth cohort to the 1990s birth erational share is opposite to the trend in sibling correlation. cohort (Latin America and the Caribbean and East Asia This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at fshilpi@worldbank.org. 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 views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Growing Up Together: Sibling Correlation, Parental Inuence, and Intergenerational Educational Mobility in Developing Countries Md. Nazmul Ahsan, St. Louis University1 M. Shahe Emran, IPD at Columbia University Hanchen Jiang, University of North Texas Qingyang Han, Johns Hopkins University Forhad Shilpi, DECRG, World Bank First Drat: June 22, 2022; This Version: January 16, 2023 ABSTRACT This paper presents credible and comparable evidence on intergenerational educational mobil- ity in 53 developing countries using sibling correlation as a measure, and data from 230 waves of Demographic and Health Surveys. It is the rst paper to provide estimates of sibling correlation in schooling for a large number of developing countries using high quality standardized data. Sib- ling correlation is an omnibus measure of mobility as it captures observed and unobserved family and neighborhood factors shared by siblings when growing up together. The estimates suggest that sibling correlation in schooling in developing countries is much higher (average 0.59) than that in developed countries (average 0.41). There is substantial spatial heterogeneity across regions, with Latin America and Caribbean having the highest (0.65) and Europe and Central Asia the lowest (0.48) estimates. Country level heterogeneity within a region is more pronounced. The evolution of sibling correlation suggests a variety of mobility experiences, with some regions registering a monotonically declining trend from the 1970s birth cohort to the 1990s birth cohort (Latin America and the Caribbean and East Asia and Pacic), while others remained trapped in stagnancy (South Asia and Sub-Saharan Africa). The only region that experienced monotonically increasing sibling correlation is the Middle East and North Africa. The recent approach of Bingley and Cappellari (2019) is used to estimate the share of sibling correlation due to intergenerational transmission. The estimates show that when the homogeneity and independence assumptions implicit in the standard model of intergenerational transmission are relaxed, the estimated share is much larger. In the sample of countries, on average 74 percent of sibling correlation can be attributed to intergenerational transmission, while there are some countries where the share is more than 80 percent (most in Sub-Saharan Africa). This suggests a dominant role for parents in determining the educational opportunities of their children. Evidence on the evolution of the intergenerational share, however, suggests a declining importance of the in- tergenerational transmission component in many countries, but the pattern is diverse. In some cases, the trend in the intergenerational share is opposite to the trend in sibling correlation. JEL Codes: J0, D3, J62 Key Words: Sibling Correlation, Intergenerational Mobility, Education, Years of School- ing, Developing Countries, Intergenerational Share, Decomposition, DHS 1 Emails for correspondence: shahe.emran.econ@gmail.com. (M. Shahe Emran); fshilpi@worldbank.org (Forhad Shilpi). We would like to thank Lorenzo Cappellari for helpful discussion on econometric issues, and Ira Gang, Kunal Sen, Lewis Anderson and participants at a seminar at University of Windsor for useful comments on an earlier version. (1) Introduction A vast literature on intergenerational mobility and inequality of opportunity in economics and sociology focuses on the role of family and neighborhood background in shaping the life 2 chances of a child. This literature, however, primarily concentrates on developed countries, and the evidence base on developing countries remains relatively sparse. There has been a spurt in the interests among development economists in understanding the geography and evolution of intergenerational mobility in the recent years, partly spurred by the evidence of increasing inequality in the 1980s and 1990s in many developing countries (World Bank (2006), 3 Chancel et al. (2022)). Among the policymakers, there has been an increasing emphasis on 4 inequality of opportunity as opposed to inequality of outcome. The observed increase in cross-sectional inequality in socioeconomic outcomes is of serious concern when it reects increasing inequality of opportunity and declining intergenerational mobility. It is, however, dicult to build a credible and comparable cross-country evidence base on intergenerational mobility in developing countries because of data limitations. Two essen- tial building blocks are required for such an analysis: a measure of socioeconomic status of children and parents, and a measure of intergenerational mobility or inequality of economic 5 opportunity. Permanent income has been the preferred measure of socioeconomic status in the economic literature on developed countries, but reliable panel data for a long enough pe- 6 riod to estimate permanent income remain scarce in developing countries. In the absence of income data, it is feasible to use education data across a large number of developing countries, 2 For recent surveys of the economic literature see, Black and Devereux (2011), Bjorklund and Salvanes (2011), Solon (1999), Heckman and Mosso (2014), Mogstad and Torsvik (2021). For surveys of the sociology literature see Breen (2010) and Torche (2015a). 3 The focus on geography of intergenerational mobility also reects the inuential work of Chetty et al. (2014) which constructed an opportunity atlas with estimates of absolute mobility at the zip code level in the United States. 4 For example, German President Joachim Gauck declared equality of opportunity as the normative policy goal during his inauguration in 2012. US president Barack Obama in his 2014 state of the union address mentioned opportunity 10 times. 5 It has been increasingly emphasized in the literature that intergenerational mobility and inequality of opportunity are closely related, and they focus on the same fundamental question, even though these two strands of the literature developed independently. For discussions, see Bjorklund and Jantti (2020) and Emran and Shilpi (2021). 6 There are other diculties in measuring household and individual income in a developing country. First, because of large informal home-based economic activities, it is dicult to measure income. Second, It is dicult to measure individual income in an extended family living and eating together in the same household. For a discussion on this point, see Iversen et al. (2019). 1 but a major concern here is that parental (usually father's) education may give us only a partial measure of socioeconomic status of a child. To address this, one can include a vector of observable family characteristics in a model of intergenerational mobility by adopting the latent socioeconomic approach developed by Lubotsky and Wittenberg (2006) (henceforth 7 called LW approach following Emran and Shilpi (2021)). A second solution is oered by the literature on inequality of opportunity (henceforth IOP) that grew out of the inuential work of Roemer (1998) and Coleman et al. (1966). The IOP approach uses multiple indicators such as father's education and occupation, sex of a child, ethnicity etc. as the circumstances for which a child should not be held responsible as they did not choose them, rather inherited by birth. A practical challenge for both of these approaches (IOP and LW) is that the set of family background indicators one can use is dictated by the lowest common denominator across the surveys in dierent countries. In practice, these approaches thus rely on a limited number of observed family characteristics when studying inequality of opportunity and in- tergenerational mobility across countries (see the discussion by Bjorklund and Jantti (2020) on IOP). A related important limitation is that these approaches often fail to take into ac- count the eects of unmeasured and dicult to measure family characteristics (e.g., cultural inheritance) and neighborhood characteristics (e.g., school and peer quality). It is especially dicult to get reliable and comparable data on these tacit aspects of family and neighborhood environment when studying a large number of countries. In a forthcoming paper, Deutscher and Mazumder (2022) discuss 19 measures of intergen- erational mobility and nd that some of them are only weakly correlated with each other, suggesting that they measure very dierent concepts of economic mobility. When dierent studies use dierent measures of mobility capturing disparate economic concepts, it is im- possible to rank countries in terms of intergenerational mobility based on the evidence from individual country studies. Many recent studies on developing countries use measures of inter- generational association based on the conditional expectation function of children's schooling given father's schooling. However, even among the subset of studies focusing on the condi- tional expectation function of schooling, comparability is compromised in two ways. First, some studies use years of schooling, some use years of schooling normalized by standard de- 7 Notwithstanding its advantages, the LW approach has not been adopted widely yet. We are aware of only two studies that used this approach: Neidhofer et al. (2018) and Vosters and Nybom (2017). 2 8 viation, and some others rely on ranks in the schooling distribution. Second, even when based on the same measures of education and mobility, comparability across countries is often compromised by a lack of standardization of the data. Some surveys include data on years of schooling while others only categorical variables such as primary, secondary, etc. These dier- ent measures of schooling are likely to contain dierent degrees of measurement error and a study based on such data may not yield comparable estimates of intergenerational educational mobility across countries. To deal with these challenges, we use sibling correlation in education as a measure of mobility, and take advantage of data from Demographic and Health Surveys (DHS). Sibling correlation stands out as a measure on both conceptual and empirical grounds. Conceptually, sibling correlation is a broader measure of intergenerational mobility compared to most of 9 the other measures used in the literature. First, the standard measures such as correlation between parents and children's schooling attainment usually rely solely on father's education because of missing data on mother's education. Sibling correlation reects the eects of both father and mother along with other family members residing in the household such as grand- parents, uncles, and aunts. Second, similarity between sibling's educational outcomes does not only reect observable family characteristics, but also unobservable (to the researcher) factors shared by siblings such as family culture. It also captures other factors such as parenting style, aspiration, and risk attitude, even when the data sets do not contain any information on these variables. In this sense, sibling correlation is broader than the IOP and LW ap- proaches which can only use the observable characteristics measured in a survey. 10 Third, sibling correlation captures broader neighborhood eects (including school and peers) that are not correlated with parental education but shared by the siblings growing up together in the same neighborhood. An important advantage of sibling correlation as a measure is that it is less susceptible to coresidency bias because we want to capture the factors shared by siblings when growing 8 For a comparative analysis of these alternatives measures of intergenerational educational mobility with a focus on possible conicts among them, please see Ahsan et al. (2022). They use data from China, India, and Indonesia. 9 For discussions on this point, see Bjorklund and Salvanes (2011), Bjorklund et al. (2010), Deutscher and Mazumder (2022). 10 IOP can include variables that represent factors not shared by the siblings. See the discussion by Bjorklund and Jantti (2020). 3 up together, going to the same neighborhood school, socializing with the same cohorts of peers. Thus, we would like to exclude the siblings who grew up far apart in dierent times (or dierent places like a sibling who left home for a boarding school), and might have experienced 11 dierent family, neighborhood, and school environment when growing up. This includes, for example, much older children of the household head who are not coresident because they are in college or working in a dierent location after graduating from college. We restrict the age 12 of children to 16-28 years in the survey year of a wave for our analysis. This also ensures 13 that possible sample truncation because of coresidency is small. We provide estimates of sibling correlation in educational attainment for 53 develop- ing countries and trace out the evolution of intergenerational educational mobility for three 14 decade-wise birth cohorts (1970s to 1990s). We use 230 waves of Demographic and Health Surveys (DHS) to build a comparable data base across countries. This ensures that the estimates are not dierent because of dierences in survey instruments or measurement of schooling (years of schooling vs. categorical). Another important advantage is that the in- formation on parental education is not based on recall of the children, and thus are likely to contain much less measurement error. The advantages of the Demographic and Health Sur- veys for cross-country comparison studies have been well appreciated in the recent literature (see, for example, Bhalotra and Rawlings (2013), and Lleras-Muney et al. (2022)). Another important contribution of this paper is that we take advantage of recent method- ological advances to provide a credible answer to a long-standing policy-relevant question: 11 This is supported by our cohort-wise estimates that show signicant changes in sibling correlation over time. Nicoletti and Rabe (2013) provide evidence that sibling correlation in academic attainment declines sharply as more distantly spaced siblings are considered. 12 Thus, the maximum age dierence between siblings in our sample can be 12 years. Bingley and Cappellari (2019) also use a maximum of 12 years age gap to dene their sample. This implies that the cohort variations in our analysis primarily come from dierent survey years of dierent waves of DHS for a given country. We note here that the main conclusions are robust to alternative age ranges. These alternative results are available from the authors. 13 It is well understood in the literature that the degree of sample truncation due to coresidency declines as we focus on younger children. See, for example, the recent inuential analysis on intergenerational educational mobility in Africa by Alesina et al. (2021) where they use census data (coresident samples) from IPUMS and restrict age of children to 18-24. 14 The only other paper we are aware of that provides estimates of sibling correlation for multiple developing countries is Dahan and Gaviria (2001) who report estimates for 16 Latin American countries. But their estimates are not comparable to the other estimates available in the literature because they use a dierent measure. They focus on the educational failure (lack of grade progression) of a child rather than educational attainment. 4 how long is the father's shadow cast on the siblings? A major focus of the literature on developed countries has been on the share of sibling correlation accounted for by father's ed- ucation or income (more broadly, parental education or income). A substantial literature on sibling correlation in income in developed countries suggests that the share of the intergener- ational transmission may be small (see Bjorklund and Salvanes (2011) and Solon (1999)). A small intergenerational share would imply that the focus of research and policy should be on the neighborhood and school factors rather than family (Solon (1999)). However, the recent analysis by Bingley and Cappellari (2019) suggests that this low estimate is due to the restric- tive assumptions of homogeneity and independence in the estimates of the intergenerational transmission from the parents to the children. They develop an approach that relaxes these assumptions, and nd that the share accounted for by the intergenerational transmission of income in sibling correlation in income is much higher. We provide estimates of intergen- erational share in sibling correlation in schooling using the classic methods (Solon (1999), Bjorklund et al. (2010), Mazumder (2008)) along with the Bingley and Cappellari (2019) approach. The evidence suggests four key conclusions. First, the estimates of sibling correlation in our sample of developing countries are, in general, substantially larger than the existing estimates for the developed countries. The average in our 53 country sample is 0.59, and the average for the top half of the distribution is 0.65. Based on 56 estimates available for 15 developed countries, the average sibling correlation in schooling is 0.41. Our estimates thus suggest a considerable gap in educational opportunities between the developing and developed 16 worlds. Second, there is signicant spatial heterogeneity at the regional level and across countries within a region. The Latin America and Caribbean region experienced the worst educational 15 The 56 estimates are for the same birth cohorts as our sample: 1960s to 1990s birth cohorts. Bjorklund and Salvanes (2011) report a range of 0.40-0.60 for developed countries with the estimates for the United States among the highest. Prag et al. (2019) report an average of 0.49 from a meta analysis of the studies on sibling correlation in income and education published between 1972-2018 (includes both developing and developed countries). 16 The existing cross-country evidence for developing countries based on father-child correlation in schooling also nds lower mobility in developing countries (see, for example, Neidhofer et al. (2018)). However, the existing evidence cannot answer whether this pattern holds for broader measures of socioeconomic status. We provide the rst evidence that this conclusion holds even when we capture a broad set of observed ad unobserved family and neighborhood factors. 5 opportunities with an estimated average sibling correlation of 0.65, and East Asia and Pacic is the second worst (0.64), while Europe and Central Asia had the lowest sibling correlation estimate (0.48). The within region heterogeneity is also substantial; for example, in Sub- Saharan Africa, the maximum estimate is 0.77 (Madagascar) and the minimum is 0.49 (South Africa). Third, the evolution of sibling correlation from the 1970s birth cohort to the 1990s birth cohort suggests a rich variety of mobility experiences. At the regional level, some experi- enced monotonic improvements in intergenerational educational mobility (Latin America and Caribbean, and East Asia and Pacic), while some others faced stagnation (South Asia, and Sub-Saharan Africa). Middle East and North Africa stands out as the only region to have a declining trend in educational opportunities (i.e., monotonically increasing sibling correla- tion from the 1970s to the 1990s). However, the regional average conceals a lot of country level heterogeneity. For example, notwithstanding a stagnant regional average in South Asia, Bangladesh achieved substantial gains in educational opportunities with sibling correlation declining from 0.67 (1970s) to 0.61 (1990s). In contrast, Pakistan experienced a declining intergenerational educational mobility with sibling correlation increasing from 0.60 (1970s) to 0.68 (1990s). Fourth, estimates of the role played by intergenerational transmission between father and children vary dramatically depending on the decomposition method used. Consistent with the analysis of Bingley and Cappellari (2019), we nd that the estimated share of intergenerational transmission in sibling correlation is considerably higher when we relax the homogeneity and independence assumptions implicit in the standard methods of decomposition. The estimates from the Bingley and Cappelari (2019) approach suggest an average of 74 percent across 53 countries, and in some countries, the share is higher than 80 percent (many of them in Sub- Saharan Africa). In contrast, the average share is only 34 percent according to the estimates 17 from Bjorklund et al. (2010) approach. The estimates across birth cohorts show that the share of intergenerational transmission has declined in many countries from the 1970s to the 1990s birth cohort. But there are 13 countries where the share has increased over decades, many of them (11) are located in the Sub-Saharan Africa region. 17 The average is 30 percent according to the Solon (1999) method, and 18 percent according to the Mazumder (2008) method. 6 The rest of the paper is structured as follows. Section (2) discusses the related literature and puts the contributions of this paper in perspective. Section (3) is devoted to the conceptual framework that describes the measure of sibling correlation and the decomposition methods for estimating the share of intergenerational transmission in sibling correlation. A special focus here is on the Bingley and Cappalleri (2019) approach. Section (4) discusses the advantages of the Demographic and Health Surveys for our cross-country analysis and provides a brief discussion of the estimation methods. Section (5), arranged in a number of subsections, reports and discusses the estimates of sibling correlation. Section (6) discusses the estimates of the share of the intergenerational transmission across regions and countries, and traces out the evolution over time from the 1970s birth cohort to the 1990s birth cohort. The paper ends with a summary of the results and the contributions of the paper in the conclusions. (2) Related Literature The economics literature on intergenerational mobility is grounded on the seminal contri- butions of Becker and Tomes (1986) that developed a model of intergenerational persistence in permanent income focusing on the role of human capital. The inequality of opportunity (IOP) strand of the literature builds on the foundation of the theory of distributive justice developed by Roemer (see Roemer (1998), and Roemer and Trannoy (2016)). The inequality of oppor- tunity (IOP) refers to the circumstances a child is born into, and emphasizes that inequality due to the circumstances is unjust and should be the focus of policy interventions. Although these two approaches grew largely independently, there has been an increasing appreciation 18 that they deal with fundamentally the same question. These two approaches can be best viewed as complementary. The IOP provides a theory of justice foundation, but does not identify the economic mechanisms which could be the policy levers. The Becker-Tomes model identies a set of such economic mechanisms. The sociological literature uses occupational prestige and class mobility with a focus on the long-term factors including the role of formal and informal institutions, especially in the labor market (see Torche (2015a), Breen (2010)). In the recent decades, many sociologists adopted the regression-based approach of economists and appeal to the Becker-Tomes model for theoretical underpinning of their results. 18 See, for example, the discussion by Deutscher and Mazumder (2022), and Bjorklund and Jantti (2020). We discuss the dierences between these two approaches later in the paper. 7 As noted in the introduction, the literature on developing countries mainly focuses on in- tergenerational educational mobility because of data limitations. Although there is a growing literature studying the persistence of educational attainment across generations at the country level, the studies that attempt to provide comparable estimates across a sample of developing 19 countries remain limited. The most widely known cross-country analysis of intergenerational educational mobility is Hertz et al. (2008) that provides estimates of relative mobility using in- tergenerational regression coecient (IGRC) and intergenerational correlation (IGC) between father and children for 42 countries. Neidhofer et al. (2018) report estimates of a variety of measures of absolute and relative educational mobility for 18 Latin American countries. A number of recent papers focus on Sub-Saharan African countries, see, for example, Alesina 20 et al. (2021), Azomahou and Yitbarek (2021), and Razzu and Wambile (2022). The most extensive analysis of intergenerational educational mobility around the world is oered in a recent book by Narayan et al. (2018) covering 153 countries. They provide estimates of a num- ber of absolute and relative educational mobility measures, but their main analysis is based on the IGRC estimates. Perhaps, more important, all the cross-country studies noted above focus on the intergenerational link between parents' and children's educational attainment, and none report estimates of sibling correlation. In more than two decades following the publication of the handbook of labor economics chapter by Solon (1999), there have been only a few studies on developing countries that use sibling correlation as a measure of educational mobility. This is puzzling because Solon (1999) and the subsequent surveys of the eld (e.g., Bjorklund and Salvanes (2011)) provide substantial discussions on the advantages of sibling correlation as a measure, especially in the data scarce environment common in developing countries. The most widely cited is a study by Dahan and Gaviria (2001) who report estimates of sibling correlation in schooling for 16 Latin American countries. But as noted earlier, they do not follow the methodology developed in Solon et al. (1991), and Solon (1999). They use a measure of educational failure (lack of grade 19 At the individual country level, recent contributions on intergenerational educational mobility in develop- ing countries include Kundu and Sen (2022), Azam and Bhatt (2015), Azam (2016), Emran and Shilpi (2015) on India, Fan et al. (2021), Emran and Sun (2015) on China, Torche (2015b) on Mexico, Assaad and Saleh (2018) on Jordan; Ahsan et al. (2023), Ahsan et al. (2021) on Indonesia. For surveys of this literature, please see Iversen et al. (2019), Torche (2019), and Emran and Shilpi (2021). 20 These studies rely on census data from IPUMS. 8 progression) rather than educational attainment of children, and construct an index dierent from the measure of mobility used by Solon (1999) and other studies. Their index is based on the index of segregation proposed by Kremer and Maskin (1996). Their estimates are thus not comparable to the other estimates of sibling correlation in the literature, including the estimates reported in this paper. In a meta analysis of sibling correlation estimates published between 1972-2018, Prag et al. (2019) identify only two studies on developing countries including that of Dahan and Gaviria (2001), the second study is on intergenerational educational mobility in post-reform India by Emran and Shilpi (2015). In contrast, the literature on sibling correlation in education and income in developed countries is substantial with contributions from both economists and sociologists. For surveys of this literature, please see Solon (1999), Bjorklund and Salvanes (2011), Bjorklund and Jantti (2020). Given the focus of the economic literature in developed countries on income, many of the existing studies provide estimates of sibling correlation in income. But the literature on sibling correlation in education is also large. Most of the estimates of sibling correlation in schooling in developed countries fall in the range of 0.40-0.60 (see Bjorklund and Salvanes (2011)). Among recent contributions, Grtz et al. (2021) report estimates of sibling correlation in education for 6 developed countries with an average estimate of 0.44, the lowest estimate of 0.36 (Finland) and the highest 0.51 (United States and Germany). (3) Conceptual Framework (3.1) Sibling Correlation For the estimation and interpretation of sibling correlations, we adopt a conceptual frame- work that has been the workhorse in the empirical literature on sibling correlations (see, Solon et al. (1991), Solon (1999), Bjorklund et al. (2002), Bjorklund and Lindquist (2010), Mazumder (2008) and (2011)). Following Solon (1999) and Bjorklund et al. (2010), we begin with a simple model of children's educational attainment: ˜if = µ + ΓXi + af + bif S (1) Where ˜if S is measure of educational attainment, usually years of schooling, of sibling i in family f , µ is the country specic component that captures the factors common to all children 9 of a country, and Xi is a set of individual characteristics elements of which depend on the propose of the analysis. Following Bjorklund et al. (2010), we include a gender dummy, and, following Bingley and Cappellari (2019), we include cohort dummies, but no other controls are included in Xi .21 af is the family and neighborhood component shared by all siblings in family f , and bif is the individual specic component for sibling i capturing i's deviation from the common family and country components. We dene demeaned years of schooling Sif as follows: ˜if − (µ + ΓXi ) = af + bif Sif = S (2) The focus of the analysis is on the importance of the family and neighborhood component (the family component, for short) af in explaining the variance in demeaned years of schooling Sif .22 The country mean µ represents the growth and structural change in a country that inuence all children the same way irrespective of their family background. The cohort dum- mies take out the cohort specic eects shared by the children of a cohort, but may vary across dierent cohorts. The inclusion of the country and cohort specic intercepts in the vector Xi implies that the measure of mobility based on sibling correlation in demeaned schooling refers to relative rather than absolute mobility. Assuming that af is independent of bif , the variance of Sif can be expressed as the sum of variances of the family and individual components as: 2 2 2 σs = σa + σb (3) The sibling correlation in education (denoted by ρs ) then can be expressed as: 2 σa ρs = 2 + σ2 (4) σa b Sibling correlation thus estimates the share of variance of children's education that can be attributed to common family and neighborhood background. Sibling correlation is a measure of mobility (more precisely immobility) because the fam- 21 Some studies on intergenerational mobility in education include age and age squared following the literature on intergenerational mobility in income. However, age and age squared are used in studies on income to mop up life-cycle eects. For education, such life-cycle eects are not relevant. 22 In the variance components analysis, af is usually called the family component as it represents the family xed eect. 10 ily and neighborhood factors shared by the siblings growing up together are not chosen by themselves, but they are born into it. Thus, this measure is consistent with the inequality of opportunity foundation of distributive justice a la Roemer (1998). As discussed by Emran and Shilpi (2015), the basic insight of the Becker and Tomes (1986) model that imperfections in the credit market lead to lower mobility also holds for sibling correlation. When the credit market is perfect and parents can borrow at a given interest rate r to nance children's education, optimal investment is independent of family background and depends only on the ability of a child. Under the assumption that the distribution of innate ability does not depend on family background, the variance in the average education of children 2 across families captured by σa would be approximately zero. Now, consider the credit market imperfections model of Becker et al. (2018) where the poor (less educated) parents has access to credit market for children's education, but have to pay a higher rate, and the rich (and highly educated) pay low interest rate: rl > r > r h with subscripts l and h referring to low educated and high educated parents. In this case, r represents the interest rate faced by the families in the middle of the distribution. Parents in the low educated families optimally invest less in children's education at a given ability level, and the average education of siblings increases with the level of parental education. This increases the variance in children's schooling across 2 families, thus making σa and sibling correlation positive. Note that the strength of sibling correlation increases with the degree of credit market imperfections as captured by dierences in the interest rates faced by dierent households. The important point here is that sibling correlation as measure of mobility is grounded in the political philosophy foundation of theory of justice developed by Roemer (1998), and also consistent with the main insights of the Becker-Tomes model. An important advantage of sibling correlation is that it captures all the observed and unobserved (including the unobservables) family and neighborhood factors shared by siblings while growing up together. This, however, does not mean that sibling correlation provides an upper bound for the eects of family and neighborhood factors on educational opportunities of children. As noted by Bjorklund et al. (2010), while sibling correlation is a broader measure, it is in fact a lower bound estimate of the eects of family and neighborhood background, because 11 it does not include the factors not shared by siblings.23 (3.2) Intergenerational Correlation vs. Sibling Correlation Given that there is a large literature on intergenerational persistence in education, a natural question to ask is how much of the sibling correlation can be accounted for by the intergener- ational transmission from parent's (usually father) to children. If the widely used measures of intergenerational educational mobility such as intergenerational regression coecient (IGRC) or intergenerational correlation (IGC) can explain most of the sibling correlation, then it would suggest primacy of the family and parents in shaping the educational opportunities of 24 children, and policy should focus on the family. The link between intergenerational trans- mission and sibling correlation has been a focus of the literature since the early contributions on sibling correlation in income in the United States by Solon et al. (1988) and Solon et al. (1991). A simple approach to understanding the role of the intergenerational transmission is to estimate sibling correlation with and without conditioning on parental education. Mazumder (2008) uses this approach to estimate the share of parental inuences in sibling correlation in income in the United States, but does not estimate the share of intergenerational correlation in sibling correlation in education of children. Emran and Shilpi (2015) adopts this approach to estimate the share of father's inuences in sibling correlation in post-reform India. A second and more widely used approach was developed earlier by Solon (1999). Following Solon (1999) and Bjorklund et al. (2010), we can derive the relation between sibling correlation and intergenerational correlation. We can decompose the family and neighborhood component 25 af into two orthogonal parts: p af = βSf + λR f (5) 23 Bjorklund and Jantti (2020) note that, for most of the data sets, sibling correlation is a broader measure than inequality of opportunity even though one can include some of the non-shared factors (e.g., birth order) as part of the vector of circumstances in an IOP approach. As noted earlier, in a cross-country analysis only a few indicators of circumstances are included because the feasible set is determined by the lowest common denominator. 24 IGRC is the slope estimate from a regression of children's schooling on parent's schooling. In contrast, IGC is the slope from a regression model where both children's and parent's schooling are normalized by their respective generation-specic standard deviation in schooling. IGC thus estimates Pearson Correlation. 25 As noted by Bjorklund and Jantti (2020), this decomposition of sibling correlation was rst derived by Solon (1999). But the sociology literature on sibling correlation contains informal discussion on this before the formal derivation by Solon (1999). 12 p where βSf is the part due to parental education and λR f is the residual sibling eect. Taking variance of equation (5), we have: 2 σa 2 = β 2 σp 2 + σλR (6) 2 Dividing through by σs we get: 2 β 2 σp σ2 ρs = 2 + λR 2 = IGC 2 + Residual Sibling Correlation (7) σs σs 2 2 If one assumes stationary distributions across generations, then σp = σs and we have ρs = β 2 + Residual Sibling Correlation (8) In fact, Solon (1999) derived the decomposition under the assumption of stationary distri- butions as in equation (8), while Bjorklund et al. (2010) used equation (7). Residual sibling correlation represents all other factors shared by siblings but uncorrelated with parental education. Many studies on intergenerational mobility in developed countries used equations (7) and (8), and the conclusion from this literature is that only a small part of sibling correlation could be explained by the parental education. According to the estimates for years of schooling reported by Bjorklund and Jantti (2020), the IGC estimate for Sweden is 0.30 and sibling correlation is 0.43. The squared IGC is thus 0.09, only about 20 percent of sibling correlation is explained by IGC. However, equation (5) is motivated by the workhorse linear mobility equation for esti- mating IGC which imposes a number of assumptions that are likely to be rejected on both theoretical and empirical grounds. Recent theoretical advances suggest that the assumption of linearity is likely to be violated in many cases. Becker et al. (2015) develop a model of intergenerational educational persistence between parents and children where the mobility 26 equation can be concave (due to diminishing returns) or convex (due to complementarities). A concave or convex intergenerational persistence equation has two important implications: 26 Recent evidence suggests that the mobility equation is not linear in most of the cases. Emran et al. (2020) nds that the mobility equation in India is concave irrespective of gender. Ahsan et al. (2022) provides evidence suggesting concave or convex mobility equations for years of schooling in China, India, and Indonesia. 13 (i) the eects of parents on children as captured by IGC (β ) are heterogeneous across fami- lies; and (ii) the parameter β can be positively (for convex mobility function) or negatively (concave mobility function) correlated with parental education. Bingley and Cappellari (2019) develop a decomposition method that allows for heterogeneous β and arbitrary correlation be- p tween β and Sf . They show that, for sibling correlation in income, relaxation of the implicit assumptions in equation (5) makes a big dierence. To the best of our knowledge, we are the rst to implement the Bingley and Cappellari (2019) approach for estimating the intergenerational share in sibling correlation in education, and we do it for a large number of developing countries (53 countries) using comparable data from the DHS surveys. We provide a brief discussion of the Bingley and Cappellari (2019) approach below, and refer the reader to the original paper for details. (3.3) Decomposition of Sibling Correlation: Bingley and Cappellari (2019) Approach In the context of our set-up, Bingley and Cappellari (2019) replace equation (5) by the following random coecient specication: ¯ + βf S p + λR af = β (9) f f where ¯ β is average eect of parental education and βf is deviation of family f from the mean. This specication thus incorporates heterogeneity in the eects of parental education captured by the parameter βf . If we relax only the heterogeneity assumption but retain the assumption that the magnitudes of the parental eect is independent of the level of parental education, we have the following decomposition: ¯2 + σ 2 σ 2 σ 2 β β p ρs = 2 + λR2 (10) σs σs But as we discussed above, there are plausible theoretical models that suggest that βf p is correlated with Sf . Using a result on the exact variance of the product of two random variables due to Bohrnstedt and Goldberger (1969), Bingley and Cappellari (2019) derive the 14 following decomposition (under normality): ¯2 + σ 2 σ 2 + cov βf S p 2 β 2 β p f σλR ρs = 2 + 2 (11) σs σs p 2 Since cov βf Sf ≥ 0, assuming independence in equation (10) will in general underesti- mate the role of the intergenerational transmission of schooling. The evidence on intergener- ational income mobility in Denmark reported by Bingley and Cappellari (2019) suggests that the relaxation of the independence assumption is especially important; the estimated share of the intergenerational component (father's income) increases substantially as a result. The decomposition in equation (11) relaxes two important restrictive assumptions in the p standard specication (5): homogeneity in βf and independence between βf and Sf , but it relies on the normality assumption which is rejected by the data in most of the cases. Bingley and Cappellari (2019) nd that imposing normality tends to underestimate the share of intergenerational inuences in sibling correlation. They relax the normality assumption by using an unrestricted form of the intergenerational correlation between the children and parents. In our empirical analysis, we will report estimates from both the classic methods (Bjorklund et al. (2010), Mazumder (2008), Solon (1999)), and the method due to Bingley and Cappellari (2019). (4) Data and Estimation Methods A major hurdle for credible cross-country ranking of inequality of opportunity and inter- generational mobility is that data from dierent surveys may not be comparable. As noted earlier, the survey instruments used for education information by DHS are standardized across 27 countries which makes the data much more comparable. Even when trying to elicit the same information (say education of parents and children), dierent household surveys may contain dierent kinds of data. In the context of intergenerational educational mobility, there are two issues relevant here. First, whether data on educational attainment refer to years of schooling or education categories (primary, secondary etc.). The DHS data we use have information on years of schooling for both the parents (father) and children. Second, in many household surveys used for intergenerational mobility analysis in developing countries, data on parental 27 We take the information from the household roster which is the same in all DHS surveys. 15 education are based on children's recalled information, and thus may contain non-negligible measurement error (Emran and Shilpi (2021), Torche (2019)). This would tend to bias down- ward the estimated parents-children persistence in education and, in turn, lead to downward biased estimate of the share of the intergenerational transmission in sibling correlation. The DHS data on parents are not based on recall, and thus are much more reliable. There are 53 countries in our sample. We use 230 waves of DHS surveys to build our data base. We exclude 42 countries where at least one DHS survey is available but the sample size is small. The cut-o for inclusion is a minimum of 1000 observations in the sample. The trade-o between country coverage and sample size is well-appreciated in the literature. For a recent analysis of intergenerational educational mobility covering a large number of countries (153), see Narayan et al. (2018), but, as noted earlier, they do not provide estimates of sibling 28 correlation. In each wave of DHS, our sample includes children of age 16-28 in the survey year. The exclusion of relatively older age cohorts in each wave is motivated by two considerations. First, it reduces the possibility of sample truncation due to grown-up children leaving the household for work or to start a family. Second, as noted earlier, we would like to exclude children who are born far apart as they are likely to face dierent family, neighborhood, and school environ- ments. Among our 53 countries, there are 6 countries with fewer than 2000 observations, and 22 countries with sample size more than 5000. The total number of observations in our data set is 544624. The country level estimation samples include children from the 1960s to 1990s birth cohorts. But in many countries, the number of observations for the 1960s birth cohort is small because only a limited number of DHS surveys were administered in these countries in the 1990s and earlier. For the analysis of the evolution of educational mobility across cohorts we thus do not include 1960s observations, and focus on the three decade wise birth cohorts from 1970s to 1990s. The estimation method adopted by Bingley and Cappellari (2019) is the method of mo- ments. The data requirements for the analysis are more demanding because the Bingley and 28 The price of the extensive country coverage in Narayan et al. (2018) is that in 57 countries, the sample size is less than 1000 observations, and in 19 countries less than 500 observations. There are 25 countries with more than 5000 observations. The authors are very much aware of this trade-o and report the number of observations for each estimate so that a reader can make an informed judgment. The study by Hertz et al. (2008) include 42 countries with a minimum sample size of 1047 observations (Philippines). 16 Cappellari (2019) approach is based on family triads with father and two children in a family. We take the oldest two children from those families where the number of children is more than 2. To ensure that the siblings are not too far apart, we follow Bingley and Cappellari (2019) and restrict their age gap to a maximum of 12 years. The intergenerational transmission is estimated as the average of the persistence between father and the rst child, and between father and the second child in the sample. The birth cohorts are dened based on the birth year of the older sibling in a household. For the estimation of the share of intergenerational component, we do not impose the stationary distributions assumption across generations used 29 by Bingley and Cappellari (2019) as this assumption is rejected by our data. We also nd that the estimated share can be more than 100 percent if we incorrectly impose stationary dis- tribution assumption within the Bingley and Cappellari (2019) approach. The estimates from the Mazumder (2008) method for the share of intergenerational component are implemented using Restricted Maximum Likelihood (REML) in a mixed eects model. (5) Evidence: Geography of Sibling Correlation and Evolution over Three Birth Cohorts (5.1) Geography of Sibling Correlation across Regions and Nations The estimates suggest that there are substantial regional variations in intergenerational educational mobility as measured by sibling correlation. Figure 1 presents the average sib- ling correlation estimates for six regions of the world. The country specic estimates are reported in Table 1. The estimates suggest that intergenerational educational mobility for the 1960s-1990s birth cohorts is the lowest in the Latin American and Caribbean countries with 30 an average sibling correlation of 0.65. Compare this with an average of 0.41 for developed countries noted before. This evidence on Latin America and Caribbean is interesting as the countries in this region also experienced some of the highest income inequality during this period (De Ferranti (2004)). Thus, high cross-sectional inequality was coupled with low inter- generational mobility, a doubly undesirable distributional outcome. Among the countries in 29 In the income data used by Bingley and Cappellari (2019), the null hypothesis of stationary distributions is not rejected. Stationary distributions are also assumed by Solon (1999). 30 The only other study that reports estimates of sibling correlation in schooling for Latin American countries is Dahan and Gaviria (2001). However, as discussed earlier, their estimates are not comparable to our estimates or other estimates in the literature. 17 this region, Guatemala has the unfortunate distinction of having the highest sibling correla- tion in schooling: 0.71, and the country with the lowest sibling correlation is the Dominican Republic with an estimate of 0.57. Intergenerational educational mobility is also low (comparable to Latin America) in East 31 Asia (average estimate 0.64) and South Asia (average estimate 0.62). Among the East Asian countries, Cambodia and Vietnam have the lowest intergenerational educational mobility, with a sibling correlation estimate of 0.66 in both countries. The sibling correlation estimate for Philippines is the lowest in this region (0.60). In South Asia, the estimates are very close in four out of ve countries, ranging from 0.62 (Nepal) to 0.64 (Bangladesh). Afghanistan enjoys the highest intergenerational educational mobility with an estimate of 0.56. We have two countries from the Middle-East and North Africa region for which the re- 32 quired DHS data were available: the Arab Republic of Egypt and Jordan. The estimate suggests that sibling correlation is much lower compared to the three regions discussed above (Latin America and Caribbean, South Asia, and East Asia and Pacic). Sibling correlation in schooling is 0.48 in Jordan and 0.54 in Egypt which are smaller than, for example, the estimate for the most mobile country in South Asia, Afghanistan (0.56). For Sub-Saharan Africa, we have 30 countries (please see Table 1 for the list of the coun- tries), with an average sibling correlation of 0.59. On average, Sub-Saharan Africa is more mobile than Latin America, South Asia and East Asia, but the mean estimate hides substan- tial heterogeneity across countries. The highest estimate is 0.77 for Madagascar which is also the highest among our 53 countries. There are three other countries with estimates of 0.70 or higher: Chad (0.74), Nigeria (0.70), and Ethiopia (0.70). The lowest estimate is 0.49, for South Africa. The region with the highest intergenerational educational mobility is Europe and central 33 Asia; the average sibling correlation is 0.48. Among the 5 countries from this region in 31 Our East Asia sample does not include Japan because DHS surveys do not cover Japan. The countries included are: Cambodia, Indonesia, Philippines, and Vietnam. The South Asia sample includes Afghanistan, Bangladesh, India, Nepal, and Pakistan. 32 As discussed in the data section, we excluded the countries with DHS survey if the sample size is less than 1000 observations. 33 This average estimate is slightly higher than the average of 0.44 for 6 developed countries reported by Grtz et al. (2021). The developed countries are: Finland, Norway, Germany, United States, United Kingdom, and Sweden. The countries in our sample are: Albania, Armenia, Kyrgyz Republic, Tajikistan, and Turkiye. 18 our sample, Kyrgyz Republic comes out at the top with an estimated sibling correlation of 0.38 which is also the lowest among the 53 countries in Table 1. Turkiye (previously known as Turkey) and Armenia share the unfortunate distinction of the lowest intergenerational mobility in this region with an estimated sibling correlation of 0.55. (5.2) Evolution Over Time: Estimates from Decade Wise Birth Cohorts As noted earlier, in many countries, the sample size for the 1960s birth cohort is too small for credible estimation of sibling correlation. We thus focus on the three decade wise birth cohorts, from the 1970s to the 1990s. The children born in the 1970s are likely to face sig- nicantly dierent economic and educational policies when compared to the children born in the later decades. There were two major policy developments in the 1980s and the following decades that might have aected the educational opportunities of children. First, there were economic liberalization and reform across many developing countries including trade liberal- ization, privatization and deregulation. The reform yielded impressive economic growth and substantial reductions in poverty in many countries, but at the same time increased income inequality (World Bank (2006)). Second, there were dramatic expansion of schools in many 34 developing countries. Many countries also implemented compulsory primary and secondary schooling in the decades of 1980s-2000s. As a result, signicant gains in school enrollment and schooling attainment were achieved over these decades (World Bank (2018)). Did the poverty reduction and the expansion of schooling and other educational policies outweigh the countervailing eects of inequitable growth and might have actually expanded the educational opportunities for the children in the later decades? Are there important regional dierences in the evolution of inequality of educational opportunities over these decades? We make some progress on these questions in this section. Figure 2 presents the estimates of sibling correlation for the six regions dis-aggregated 35 by the decade of birth (1970s-1990s birth cohorts). The rst impression that jumps out of 34 For a comprehensive discussion on the school expansion in developing countries, see chapter 2 titled The great school expansion- and those it has left behind in World Bank (2018). For recent analysis of the eects of school construction on intergenerational mobility, see Mazumder et al. (2019) and Ahsan et al. (2023). 35 The countries in a region in Figure 2 may vary from Figure 1, as we included only those countries for which estimates for all three decades are available. For example, Figure 2 does not include Brazil where the last DHS survey was done in 1986, and as a result, we do not have enough observations for the 1980s and 1990s birth cohorts. A cohort-wise graph including all countries can be found in the online appendix. 19 Figure 2 is that there are substantial regional heterogeneity in the evolution of inequality of educational opportunity. Of the 6 regions, 2 show monotonic improvements over the three decades, they are Latin America and Caribbean, and East Asia and Pacic. The largest decline in sibling correlation is experienced in the Latin America and Caribbean region (14.71 percent reduction, from 0.68 in the 1970s to 0.58 in the 1990s), with East Asia and Pacic also achieving a substantial decline (9.23 percent reduction, from 0.65 (1970s) to 0.59 (1990s)). The substantial improvements in intergenerational educational mobility in Latin American countries is a welcome development because of its historically high income inequality levels (De Ferranti (2004)). In fact, all 5 countries in our Latin America and Caribbean sample registered better intergenerational mobility for the 1990s birth cohort compared to that for the 1970s birth cohort. However, even after substantial decline over three decades, the estimated sibling correlation for the 1990s birth cohort remains much larger in Latin America compared to the estimate for Europe and Central Asian countries in our sample (0.42 in the 1990s). Middle East and North Africa stands out as the only region where we observe a mono- tonically increasing average sibling correlation from the 1970s birth cohort to the 1990s birth 36 cohort. Although sibling correlation was low for the 1970s cohort in these countries (0.50), it increased by 14 percent to 0.57 in the 1990s cohort which is close to the estimate of 0.58 for the Latin America and Caribbean region for the same birth cohort. In contrast, the changes in sibling correlation in South Asia and Sub-Saharan Africa are not monotonic across dierent birth cohorts. More important, the magnitudes of changes are rather small: a less than 2 percent decline in the sibling correlation estimate from the 1970's cohort to the 1990's cohort in both regions. In South Asia, the estimated sibling correlation declined marginally from 0.64 in the 1970s to 0.63 in the 1990s, although the cohorts born in the 1980s experienced a slightly better outcome (sibling correlation 0.62). This picture of stagnation in South Asia, however, conceals important heterogeneity; for example, the trajectories of change over time are opposite in Bangladesh vs. Pakistan. Sibling correlation declined substantially in Bangladesh from 0.67 in the 1970s cohort to 0.61 in the 1990s cohort, while Pakistan experienced a substantial increase from 0.60 in the 1970s to 0.68 36 A caveat here is that we have two countries from this region in our sample so the average estimate may not be representative of other countries of this region. But Egypt is by far the largest country in the region. These countries have about 20 percent of the region's population. 20 in the 1990s cohort. Evidence on India, by far the largest country in the region, suggests that intergenerational mobility remained largely unchanged over the three birth cohorts. This is striking because following extensive economic reforms including dramatic trade liberalization and domestic deregulation in 1991, India reaped impressive economic growth and poverty reduction in the decades of 1990s and 2000s during which the children of the 1980s and 1990s 37 birth cohorts went to school. The evidence thus suggests that the gains in growth and poverty reduction failed to translate into better educational opportunities for the children of 38 liberalization in India. As noted earlier, Sub-Saharan Africa as a region also did not experience any substantial improvements over the three decades. Again, the average estimates conceal substantial coun- try level diversity in mobility experiences. We observe some of the most dramatic declines in intergenerational educational mobility in this region. For example, sibling correlation in Mozambique increased from 0.52 in the 1970s cohort to 0.68 in the 1990s cohort, and in Nige- ria from 0.64 (1970s) to 0.74 (1990s). There are also a number of countries in this region that experienced substantial improvements. For example, sibling correlation in Uganda declined from 0.65 (1970s) to 0.55 (1990s), and in Tanzania, from 0.56 (1970s) to 0.48 (1990s). Out of 27 countries in this region for which we have estimates for both the 1970s and the 1990s cohorts, 16 countries registered improvements, while 10 experienced a setback in intergenera- tional educational mobility. 37 Based on Indian government ocial poverty line, the proportion of poor people in rural areas declined from 47 percent in 1983 to 28 percent in 2004-2005. The corresponding decline in urban India is from 42 percent in 1983 to 26 percent in 2004=2005. See Bank (2011). 38 This evidence of no signicant improvements in educational opportunities in India is in contrast to the evidence of substantial improvements based on the estimates of intergenerational regression coecient (IGRC) in educational attainment reported by Azam and Bhatt (2015), Jalan and Murgai (2008), Kishan (2018), and Maitra and Sharma (2010). However, this conclusion is consistent with the analysis of Emran and Shilpi (2015) which uses two rounds of DHS (called NFHS in India) surveys (1992/93 and 2006) and focuses on the 16-27 year old children in the survey year. Using sibling correlation and intergenerational correlation (IGC), they show that there has been almost no change in educational opportunities from 1992/93 to 2006. Since IGRC and IGC are partial measures and cannot take into account many factors shared by the siblings, one can make a plausible argument in favor of the conclusions based on sibling correlation estimates. We will discuss later the changes in the share of intergenerational correlation (IGC) over time in our sibling correlation estimates. Please see section 6 below. 21 (6) How Long is the Father's Shadow? Estimating the Intergenerational Share in Sibling Correlation To understand the importance of intergenerational transmission between the father and children, we primarily rely on the Bilgley and Cappellari (2019) approach. For comparison, we also report estimates from the three traditional methods used in the literature: Solon (1999), Bjorklund et al. (2010), and Mazumder (2008). A comparison of the three traditional methods shows that the estimates from the Mazumder (2008) approach are the lowest in magnitude, while the estimates from the Bjorklund et al. (2010) approach are the largest. Recall that Bjorklund et al. (2010) do not impose the stationary distribution assumption, unlike Solon (1999). As discussed earlier, our estimates using the Bingley and Cappellari (2019) approach also do not impose the stationarity assumption. We focus on a comparison of the estimates from Bjorklund et al. (2010) and Bingley and Cappellari (2019) methods in our discussion below. (6.1) Geography of Intergenerational Share Figure 3A presents the estimated share of the intergenerational transmission for our six regions based on the Bingley and Cappellari (2019) method. The corresponding shares from the Bjorklund et al. (2010) method are in Figure 3B. A comparison of these two methods suggests three major conclusions. First, the estimates from the Bingley and Cappellari (2019) approach are much larger: the lowest estimate is 0.70 (MENA region), while the highest estimate from the Bjorklund et al. (2010) approach is only 0.40 (East Asia and Pacic). The average intergenerational share for the 53 countries is 74 percent according to the Bingley and Cappellari (2019) approach, while it is only 34 percent according to the Bjorklund et 39 al. (2010) approach. This is consistent with the evidence on income mobility in Denmark reported by Bingley and Cappellari (2019), and vindicates, in a much wider context, their argument that the low estimates in the existing literature are driven by restrictive homogeneity and independence assumptions. Second, the ranking of regions may change depending on the method of decomposition used. For example, according to the Bjorklund et al. (2010) method, the share of the intergenerational transmission is larger in East Asia and Pacic (40 percent) 39 The estimates from the other two traditional methods are even lower, and in particular, the method due to Mazumder (2008) seems to yield very low estimates. 22 than that in South Asia (0.34). But the share is identical in these two regions (76 percent) according to the Bingley and Cappellari (2019) estimates. The disaggregated country level estimates of the intergenerational share are reported in Table 3 using the Bingley and Cappellari (2019) method. The estimated share is high in most of the countries (more than 60 percent in every case), and there are some countries where 80 percent or more of the sibling correlation can be attributed to the intergenerational transmission between father and the children. They are Philippines (0.81) and Vietnam (0.82) in South East Asia, Bangladesh (0.82) and Pakistan (0.81) in South Asia, and Benin (0.82), cameroon (0.85), the Republic of Congo (0.85), Madagascar (0.85), Mozambique (0.82), Senegal (0.80), and Togo (0.82) in Sub-Saharan Africa. Interestingly, none of the countries in the Latin America and Caribbean region have such a high share of the intergenerational transmission even though some of these countries have very high sibling correlation. (6.2) Evolution of Intergenerational Share We next look at the evolution of the intergenerational share across the three birth cohorts in the six regions. Figure 4 presents the results based on the Bingley and Cappellari (2019) method. It is striking that in every region, the share of intergenerational transmission declined from the 1970s cohort to the 1990s cohort, even though in some cases the magnitude is negligible (for example, Latin America and Caribbean where the share was 74 percent in the 1970s and 73 percent in the 1990s) . This can be interpreted as a declining role of parents in shaping the educational opportunities of children over time. The evolution of the share over time oers some contrasting patterns when compared to the evidence on sibling correlation across cohorts discussed earlier. The share of the intergenerational transmission remained virtually unchanged in the Latin America and Caribbean despite the substantial decline in the sibling correlation we discussed above. In the Middle East and North Africa region, the intergenerational share declined substantially across the cohorts which stands in sharp contrast to the monotonically increasing magnitudes of sibling correlation. The individual country level estimates of the intergenerational share across cohorts show a variety of mobility experiences. Although the share of intergenerational transmission declined in most of the cases, there are some countries, especially in the Sub-Saharan Africa, which experienced a higher share in the 1990s (out of the 13 countries with a higher share, 11 are 23 in Sub-Saharan Africa). In South Asia, all countries experienced a decline in the intergen- erational share, with India registering the largest reduction. The evidence suggests that the evolution of intergenerational share does not depend systematically on the level or evolution of sibling correlation in a country. The share of intergenerational transmission declined in both Pakistan and Bangladesh even though their trajectories for sibling correlation were opposite (an increasing sibling correlation in Pakistan, and a declining one in Bangladesh). (7) Conclusions We provide comparable estimates of intergenerational educational mobility for 53 devel- oping countries using sibling correlation as an omnibus measure, and data from 230 waves of Demographic and Health Surveys. Sibling correlation is an omnibus measure because it captures all the observed and unobserved family and neighborhood factors shared by the sib- lings when growing up together. Sibling correlation is thus a much broader measure compared to the other widely used measures in the literature such as intergenerational regression co- ecient, intergenerational correlation (Pearson correlation), and intergenerational rank-rank slope. Another important advantage is that sibling correlation is less susceptible to the biases caused by coresidency restrictions in the surveys as missing older children who grew up many years ago in dierent family and neighborhood environment does not bias the estimates. The Demographic and Health Surveys provide high quality data on schooling of children and fa- ther (years of schooling, not categorical), and the data on father's schooling are not based on children's recall. To the best of our knowledge, this is the rst paper to provide estimates of sibling correlation in schooling for a large number of developing countries using high quality data standardized across countries. The estimates suggest that sibling correlation in schooling in developing countries is much higher (average 0.59) than that in developed countries (average 0.41). We nd substantial spatial heterogeneity across regions, Latin America and Caribbean with the highest (0.65) and Europe and Central Asia with the lowest (0.48) estimates. Country level heterogeneity within a region is more pronounced. The evolution of sibling correlation suggests a variety of mobility experiences, with some regions registering a monotonically declining trend from the 1970s birth cohort to the 1990s birth cohort (Latin America and Caribbean and East Asia and Pacic), while others remained trapped in stagnancy (South Asia and Sub-Saharan Africa). 24 The only region that experienced monotonically increasing sibling correlation is Middle East and North Africa implying consistently declining educational mobility across cohorts. We take advantage of the recent approach of Bingley and Cappellari (2019) to estimate the share of sibling correlation due to intergenerational transmission of schooling from parents to children. We nd that relaxing the homogeneity and independence assumptions implicit in the standard methods of decomposition makes the estimated share much larger. In our sample, at least 60 percent of sibling correlation can be attributed to the intergenerational component, while there are some countries where the share is more than 80 percent (most in Sub-Saharan Africa). The average intergenerational share for the 53 countries is 74 percent. This suggests a dominant role for the parents in shaping educational opportunities of children, providing an argument in favor of causal analysis and economic policies focusing on the family. Evidence on the evolution of the intergenerational share, however, suggests a declining importance of the intergenerational transmission component in many countries, but the pattern is very diverse. In some cases, the trend in the intergenerational share is opposite to the trend in sibling correlation. References Ahsan, M. N., Shilpi, F., and Emran, S. (2023). Public Primary School Expansion, Gender- Based Crowding Out, and intergenerational educational mobility. Working paper, DECRG, The World Bank. Ahsan, N., Emran, M. S., and Shilpi, F. (2021). Complementarities and Intergenerational Ed- ucational Mobility: Theory and Evidence from Indonesia. MPRA Paper 111125, University Library of Munich, Germany. Ahsan, N. M., Emran, M. S., Jiang, H., Murphy, O., and Shilpi, F. J. (2022). When Mea- sures Conict: Towards a Better Understanding of Intergenerational Educational Mobility. Working paper, The World Bank. Alesina, A., Hohmann, S., Michalopoulos, S., and Papaioannou, E. (2021). Intergenerational Mobility in Africa. Econometrica, 89(1):135. Assaad, R. and Saleh, M. (2018). Does Improved Local Supply of Schooling Enhance Inter- 25 generational Mobility in Education? Evidence from Jordan. World Bank Economic Review, 32(3):633655. Azam, M. (2016). Intergenerational Educational Persistence among Daughters: Evidence from India. The B.E. Journal of Economic Analysis & Policy, 16(4):116. Azam, M. and Bhatt, V. (2015). Like Father, Like Son? Intergenerational Educational Mobility in India. Demography, 52(6):19291959. Azomahou, T. T. and Yitbarek, E. (2021). Intergenerational Mobility in Education: Is Africa Dierent? Contemporary Economic Policy, Forthcoming. Bank, W. (2011). Perspectives on poverty in India: Stylized facts from survey data. The World Bank. Becker, G., Kominers, S. D., Murphy, K., and Spenkuch, J. (2015). A Theory of Intergenera- tional Mobility. MPRA Paper 66334, University Library of Munich, Germany. Becker, G. and Tomes, N. (1986). Human Capital and the Rise and Fall of Families. Journal of Labor Economics, 4(3):139. Becker, G. S., Kominers, S. D., Murphy, K. M., and Spenkuch, J. L. (2018). A Theory of Intergenerational Mobility. Journal of Political Economy, 126(S1):725. Bhalotra, S. and Rawlings, S. (2013). Gradients of the Intergenerational Transmission of Health in Developing Countries. The Review of Economics and Statistics, 95(2):660672. Bingley, P. and Cappellari, L. (2019). Correlation of Brothers' Earnings and Intergenerational Transmission. The Review of Economics and Statistics, 101(2):370383. Bjorklund, A. and Jantti, M. (2020). Intergenerational mobility, intergenerational eects, sibling correlations, and equality of opportunity: A comparison of four approaches. Research in Social Stratication and Mobility, 70:100455. Bjorklund, A., Lindahl, L., and Lindquist, M. J. (2010). What More Than Parental Income, Education and Occupation? An Exploration of What Swedish Siblings Get from Their Parents. The B.E. Journal of Economic Analysis & Policy, 10(1):140. Bjorklund, A. and Salvanes, K. G. (2011). Education and Family Background: Mechanisms and Policies. In Hanushek, E., Machin, S., and Woessmann, L., editors, Handbook of the Economics of Education, volume 3, pages 201247. Elsevier. Black, S. E. and Devereux, P. J. (2011). Recent Developments in Intergenerational Mobility. 26 In Handbook of Labor Economics, volume 4, chapter 16, pages 14871541. Elsevier. Bohrnstedt, G. and Goldberger, A. (1969). On the Exact Covariance of Products of Random Variables. Journal of the American Statistical Association, 64(328):14391442. Breen, R. (2010). Educational expansion and social mobility in the 20th century. Social Forces, 89(2):365388. Chancel, L., Piketty, T., Saez, E., and Zucman, G. (2022). World inequality report 2022. Harvard University Press. Chetty, R., Hendren, N., Kline, P., and Saez, E. (2014). Where is the land of Opportunity? The Geography of Intergenerational Mobility in the United States. The Quarterly Journal of Economics, 129(4):15531623. Coleman, J. S., Campbell, E. Q., Hobson, C. J., McPartland, J., Mood, A. M., Weinfeld, F. D., and York, R. L. (1966). Equality of Educational Opportunity. Government Printing Oce, Washington, DC. Dahan, M. and Gaviria, A. (2001). Sibling correlations and intergenerational mobility in latin america. Economic Development and Cultural Change, 49(3):53754. De Ferranti, D. M. (2004). Inequality in Latin America: breaking with history? world bank publications. Deutscher, N. and Mazumder, B. (2022). Measuring Intergenerational Income Mobility: A Synthesis of Approaches. Journal of Economic Literature. Emran, M. S., Jiang, H., and Shilpi, F. J. (2020). Gender Bias and Intergenerational Edu- cational Mobility : Theory and Evidence from China and India. Policy Research Working Paper Series 9250, The World Bank. Emran, M. S. and Shilpi, F. (2015). Gender, Geography, and Generations: Intergenerational Educational Mobility in Post-Reform India. World Development, 72:362380. Emran, M. S. and Shilpi, F. (2021). Economic approach to intergenerational mobility: Mea- sures, methods, and challenges in developing countries. In Iversen, V., Krishna, A., and Sen, K., editors, Social mobility in developing countries: Concepts, methods, and determinants. Oxford University Press. Emran, M. S. and Sun, Y. (2015). Magical transition? intergenerational educational and occupational mobility in rural China : 1988-2002. Policy Research Working Paper Series 27 7316, The World Bank. Fan, Y., Yi, J., and Zhang, J. (2021). Rising intergenerational income persistence in china. American Economic Journal: Economic Policy, 13(1):20230. ¨tz, M., Barclay, K. J., Wiborg, O. N., Lyngstad, T. H., Karhula, A., Erola, J., Pra Gra ¨g, P., Laidley, T., and Conley, D. (2021). Sibling Similarity in Education Across and Within Societies. Demography, 58(3):10111037. Heckman, J. and Mosso, S. (2014). The Economics of Human Development and Social Mo- bility. Annual Review of Economics, 6(1):689733. Hertz, T., Jayasundera, T., Piraino, P., Selcuk, S., Smith, N., and Verashchagina, A. (2008). The Inheritance of Educational Inequality: International Comparisons and Fifty-Year Trends. The B.E. Journal of Economic Analysis & Policy, 7(2):148. Iversen, V., Krishna, A., and Sen, K. (2019). Beyond poverty escapes − social mobility in developing countries: A review article. World Bank Research Observer, 34(2):239273. Jalan, J. and Murgai, R. (2008). Intergenerational Mobility in Education in India. Technical report. Kishan, P. K. V. (2018). Is the Past Still Holding Us Back? A Study on Intergenerational Education Mobility in India. IIMA Working Papers WP 2018-01-03, Indian Institute of Management Ahmedabad, Research and Publication Department. Kremer, M. and Maskin, E. (1996). Wage inequality and segregation by skill. Technical report, National bureau of economic research Cambridge, Mass., USA. Kundu, A. and Sen, K. (2022). Multigenerational Mobility Among Males in India. Review of Income and Wealth. Lleras-Muney, A., Tarozzi, A., Aurino, E., and Tinoco, B. (2022). The Rise and Fall of SES Gradients in Heights Around the World. Available at SSRN 4127021. Lubotsky, D. and Wittenberg, M. (2006). Interpretation of Regressions with Multiple Proxies. The Review of Economics and Statistics, 88(3):549562. Maitra, P. and Sharma, A. (2010). Parents and children: Education across generations in India. Technical report. Mazumder, B. (2008). Sibling similarities and economic inequality in the US. Journal of Population Economics, 21(3):685701. 28 Mazumder, B., Rosales-Rueda, M., and Triyana, M. (2019). Intergenerational Human Capital Spillovers: Indonesia's School Construction and Its Eects on the Next Generation. AEA Papers and Proceedings, 109:243249. Mogstad, M. and Torsvik, G. (2021). Family Background, Neighborhoods and Intergener- ational Mobility. NBER Working Papers 28874, National Bureau of Economic Research, Inc. Narayan, A., der Weide, R. V., Cojocaru, A., Lakner, C., Redaelli, S., Mahler, D. G., Rama- subbaiah, R. G. N., and Thewissen, S. (2018). Fair Progress? Economic Mobility across Generations Around the World. Number 28428 in World Bank Publications. The World Bank. Neidhofer, G., Serrano, J., and Gasparini, L. (2018). Educational inequality and intergener- ational mobility in Latin America: A new database. Journal of Development Economics, 134(C):329349. Nicoletti, C. and Rabe, B. (2013). Inequality in pupils' test scores: How much do family, sibling type and neighbourhood matter? Economica, 80(318):197218. Prag, P., Akimova, E., and Monden, C. (2019). Intergenerational Transmission of Education: A Meta-Analysis of Sibling Correlations Published Between 1972-2018. Technical report, University of Oxford. Razzu, G. and Wambile, A. (2022). Four Decades of Intergenerational Educational Mobility in Sub-Saharan Africa. The Journal of Development Studies, 58(5):931950. Roemer, J. and Trannoy, A. (2016). Equality of Opportunity: Theory and Measurement. Journal of Economic Literature, 54(4):12881332. Roemer, J. E. (1998). Equality of Opportunity. Harvard University Press. Solon, G. (1999). Intergenerational mobility in the labor market. In Ashenfelter, O. and Card, D., editors, Handbook of Labor Economics, volume 3 of Handbook of Labor Economics, pages 17611800. Elsevier. Solon, G., Corcoran, M., Gordon, R., and Laren, D. (1988). Sibling and Intergenerational Correlations in Welfare Program Participation. Journal of Human Resources, 23(3):388 396. Solon, G., Corcoran, M., Gordon, R., and Laren, D. (1991). A Longitudinal Analysis of Sibling 29 Correlations in Economic Status. Journal of Human Resources, 26(3):509534. Torche, F. (2015a). Intergenerational Mobility and Equality of Opportunity. European Journal of Sociology, 56(3):343371. Torche, F. (2015b). Intergenerational Mobility and Gender in Mexico. Social Forces, 94:5633 587. Torche, F. (2019). Educational mobility in developing countries. WIDER Working Paper Series 2019-88. Vosters, K. and Nybom, M. (2017). Intergenerational Persistence in Latent Socioeconomic Sta- tus: Evidence from Sweden and the United States. Journal of Labor Economics, 35(3):869 901. World Bank (2006). World development report 2006: Equity and development. The World Bank. World Bank (2018). World Development Report 2018: Learning to Realize Education's Promise. Number 28340 in World Bank Publications - Books. The World Bank Group. 30 Figure 1: Sibling Correlations by Regions Notes: This figure presents the average sibling correlation estimates for six regions of the world using the full sample. Data come from 53 developing countries in the Demographic and Health Surveys (DHS). Specific countries included in each region are reported in Table 1. The dashed line represents the average sibling correlation estimates for all countries in our sample (0.60). For comparison purposes, the average sibling correlation for developed countries in the current literature is 0.41. 31 Figure 2: Sibling Correlations by Regions and Cohorts Notes: This figure presents the average sibling correlation estimates for six regions of the world dis-aggregated by different decades of birth cohorts (the 1970s, 1980s, and 1990s). Data come from 53 developing countries in the Demographic and Health Surveys (DHS). Specific countries included in each region are reported in Table 1. The dashed line represents the average sibling correlation estimates for all countries in our sample (0.60). For comparison purposes, the average sibling correlation for developed countries in the current literature is 0.41. 32 Figure 3A: Proportion of Sibling Correlations Explained by Intergenerational Transmission by Regions (Bingley and Cappellari 2019 Method) Figure 3B: Proportion of Sibling Correlations Explained by Intergenerational Transmission by Regions (Bjorklund et al. 2010 Method) Notes: This figure presents the average estimated share of the intergenerational component for six regions of the world using the full sample. Data come from 53 developing countries in the Demographic and Health Surveys (DHS). Specific countries included in each region are reported in Table 1. Panel A uses the Bingley and Cappellari (2019) method, while Panel B uses the Bjorklund et al. (2010) method. The dashed line in Panel A plots the average estimated share of the intergenerational component for all countries using the full sample (0.74). 33 Figure 4: Proportion of Sibling Correlations Explained by Intergenerational Transmission by Regions and Cohorts (Bingley and Cappellari 2019 Method) Notes: This figure presents the average estimated share of the intergenerational component using the Bingley and Cappellari (2019) method for six regions of the world dis-aggregated by different decades of birth cohorts (the 1970s, 1980s, and 1990s). Data come from 53 developing countries in the Demographic and Health Surveys (DHS). Specific countries included in each region are reported in Table 1. The dashed line plots the average estimated share of the intergenerational component for all countries using the full sample (0.74). 34 Table 1: Country-Specific Sibling Correlation Estimates (Full Sample) Sibling Sibling Country S.E. N Country S.E. N Corr. Corr. East Asia & Pacific Sub-Saharan Africa Cambodia 0.662 0.006 10521 Benin 0.600 0.010 5480 Indonesia 0.625 0.004 33209 Burkina Faso 0.620 0.013 3498 Philippines 0.595 0.007 15064 Burundi 0.539 0.013 3897 Vietnam 0.664 0.011 3692 Cameroon 0.627 0.012 3663 Chad 0.735 0.013 2710 Europe & Central Asia Congo, Rep. 0.532 0.018 1781 Albania 0.471 0.020 2311 Cote d'Ivoire 0.584 0.018 1565 Armenia 0.546 0.018 3215 Ethiopia 0.696 0.007 7047 Kyrgyz Republic 0.379 0.031 1455 Gabon 0.549 0.023 1212 Tajikistan 0.439 0.020 3238 Ghana 0.588 0.014 3313 Turkey 0.547 0.009 8292 Guinea 0.552 0.016 2589 Kenya 0.541 0.010 7559 Latin America & Caribbean Lesotho 0.581 0.011 4382 Bolivia 0.678 0.009 6971 Liberia 0.532 0.019 1771 Brazil 0.698 0.012 2967 Madagascar 0.769 0.010 4376 Colombia 0.606 0.006 17607 Malawi 0.598 0.010 6164 Dominican 0.566 0.009 8190 Mali 0.604 0.013 3910 Republic Guatemala 0.713 0.007 6553 Mozambique 0.540 0.016 2910 Haiti 0.680 0.008 6022 Namibia 0.524 0.019 2581 Peru 0.617 0.004 34974 Niger 0.675 0.015 1505 Nigeria 0.699 0.007 11380 Middle East & North Africa Rwanda 0.536 0.010 6848 Egypt, Arab Rep. 0.542 0.005 27042 Senegal 0.598 0.012 3850 Jordan 0.481 0.008 17023 Sierra Leone 0.517 0.019 2209 South Africa 0.490 0.020 2407 South Asia Tanzania 0.528 0.012 5619 Afghanistan 0.563 0.009 7585 Togo 0.521 0.020 1615 Bangladesh 0.641 0.007 12031 Uganda 0.620 0.012 4176 India 0.629 0.002 151142 Zambia 0.643 0.010 5595 Nepal 0.623 0.010 5574 Zimbabwe 0.589 0.015 3369 Pakistan 0.633 0.003 40964 Notes: This table presents the sibling correlation estimates for each of the 53 developing countries in the Demographic and Health Surveys (DHS) using the full sample. 35 Table 2: Country-Specific Sibling Correlation Estimates (By Cohorts) Country 1970 1980 1990 Country 1970 1980 1990 East Asia & Pacific Sub-Saharan Africa Cambodia 0.674 0.657 0.685 Benin 0.653 0.595 0.579 Indonesia 0.659 0.616 0.523 Burkina Faso 0.584 0.638 0.506 Philippines 0.614 0.616 0.550 Burundi n.a 0.584 0.500 Vietnam 0.682 0.616 n.a Cameroon 0.608 0.629 0.629 Chad 0.743 0.727 0.760 Europe & Central Asia Congo, Rep. 0.466 0.532 0.625 Albania n.a 0.502 0.389 Cote d'Ivoire 0.633 0.592 0.640 Armenia 0.468 0.592 0.492 Ethiopia 0.793 0.656 0.630 Kyrgyz Republic 0.379 0.438 0.342 Gabon 0.475 0.614 0.576 Tajikistan n.a 0.461 0.424 Ghana 0.592 0.617 0.500 Turkey 0.550 0.561 0.433 Guinea 0.597 0.515 0.559 Kenya 0.560 0.584 0.503 Latin America & Caribbean Lesotho 0.592 0.592 0.483 Bolivia 0.692 0.653 n.a Liberia n.a 0.535 0.522 Brazil 0.709 n.a n.a Madagascar 0.824 0.790 n.a Colombia 0.666 0.582 0.546 Malawi 0.609 0.602 0.585 Dominican 0.595 0.538 0.435 Mali 0.626 0.619 0.573 Republic Guatemala 0.750 0.697 0.680 Mozambique 0.526 0.522 0.681 Haiti 0.697 0.674 0.678 Namibia 0.554 0.535 0.516 Peru 0.668 0.591 0.546 Niger 0.670 0.684 0.645 Nigeria 0.643 0.666 0.741 Middle East & North Africa Rwanda 0.546 0.550 0.522 Egypt, Arab Rep. 0.523 0.581 0.588 Senegal 0.588 0.584 0.588 Jordan 0.474 0.468 0.551 Sierra Leone n.a 0.517 0.515 South Africa 0.521 0.383 0.501 South Asia Tanzania 0.559 0.563 0.484 Afghanistan n.a 0.523 0.591 Togo 0.446 0.570 0.519 Bangladesh 0.673 0.614 0.609 Uganda 0.648 0.666 0.552 India 0.658 0.631 0.631 Zambia 0.664 0.681 0.620 Nepal 0.638 0.618 0.588 Zimbabwe 0.569 0.609 0.625 Pakistan 0.595 0.633 0.679 Notes: This table presents the sibling correlation estimates for each of the 53 developing countries in the Demographic and Health Surveys (DHS) dis-aggregated by different decades of birth cohorts (the 1970s, 1980s, and 1990s). 36 Table 3: Country-Specific Proportion of Sibling Correlations Explained by Intergenerational Transmission Estimates (Full Sample) Country Prop. S.E. N Country Prop. S.E. N East Asia & Pacific Sub-Saharan Africa Cambodia 0.688 0.010 10521 Benin 0.824 0.018 5480 Indonesia 0.780 0.006 33209 Burkina Faso 0.720 0.021 3498 Philippines 0.812 0.011 15064 Burundi 0.660 0.024 3897 Vietnam 0.824 0.018 3692 Cameroon 0.848 0.021 3663 Chad 0.721 0.018 2710 Europe & Central Asia Congo, Rep. 0.853 0.039 1781 Albania 0.757 0.043 2311 Cote d'Ivoire 0.791 0.033 1565 Armenia 0.767 0.030 3215 Ethiopia 0.713 0.011 7047 Kyrgyz Republic 0.661 0.066 1455 Gabon 0.771 0.045 1212 Tajikistan 0.641 0.037 3238 Ghana 0.786 0.023 3313 Turkey 0.774 0.016 8292 Guinea 0.704 0.034 2589 Kenya 0.768 0.020 7559 Latin America & Caribbean Lesotho 0.717 0.021 4382 Bolivia 0.720 0.012 6971 Liberia 0.690 0.033 1771 Brazil 0.752 0.018 2967 Madagascar 0.845 0.014 4376 Colombia 0.729 0.009 17607 Malawi 0.752 0.016 6164 Dominican 0.691 0.015 8190 Mali 0.745 0.021 3910 Republic Guatemala 0.740 0.011 6553 Mozambique 0.823 0.032 2910 Haiti 0.640 0.014 6022 Namibia 0.749 0.034 2581 Peru 0.755 0.006 34974 Niger 0.704 0.027 1505 Nigeria 0.704 0.009 11380 Middle East & North Africa Rwanda 0.735 0.021 6848 Egypt, Arab Rep. 0.740 0.010 27042 Senegal 0.802 0.024 3850 Jordan 0.667 0.015 17023 Sierra Leone 0.668 0.030 2209 South Africa 0.786 0.040 2407 South Asia Tanzania 0.659 0.021 5619 Afghanistan 0.618 0.016 7585 Togo 0.825 0.043 1615 Bangladesh 0.820 0.010 12031 Uganda 0.735 0.020 4176 India 0.744 0.003 151142 Zambia 0.746 0.016 5595 Nepal 0.658 0.015 5574 Zimbabwe 0.707 0.024 3369 Pakistan 0.808 0.006 40964 Notes: This table presents the estimated share of the intergenerational component using the Bingley and Cappellari (2019) method for each of the 53 developing countries in the Demographic and Health Surveys (DHS) using the full sample. 37 Table 4: Country-Specific Proportion of Sibling Correlations Explained by Intergenerational Transmission Estimates (By Cohorts) Country 1970 1980 1990 Country 1970 1980 1990 East Asia & Pacific Sub-Saharan Africa Cambodia 0.731 0.696 0.659 Benin 0.867 0.840 0.809 Indonesia 0.787 0.791 0.774 Burkina Faso 0.844 0.688 0.727 Philippines 0.816 0.830 0.810 Burundi n.a 0.740 0.639 Vietnam 0.829 0.867 n.a Cameroon 0.742 0.841 0.871 Chad 0.735 0.753 0.677 Europe & Central Asia Congo, Rep. 0.926 0.863 0.728 Albania n.a 0.858 0.762 Cote d'Ivoire 0.737 0.845 0.732 Armenia 0.940 0.746 0.555 Ethiopia 0.749 0.781 0.679 Kyrgyz Republic 0.496 0.786 0.501 Gabon 0.890 0.672 0.775 Tajikistan n.a 0.700 0.587 Ghana 0.853 0.781 0.796 Turkey 0.819 0.767 0.701 Guinea 0.452 0.787 0.753 Kenya 0.656 0.746 0.877 Latin America & Caribbean Lesotho 0.716 0.727 0.817 Bolivia 0.761 0.709 n.a Liberia n.a 0.680 0.699 Brazil 0.758 n.a n.a Madagascar 0.859 0.816 n.a Colombia 0.778 0.755 0.706 Malawi 0.764 0.767 0.745 Dominican Republic 0.701 0.727 0.838 Mali 0.855 0.745 0.699 Guatemala 0.759 0.796 0.722 Mozambique 0.914 0.765 0.759 Haiti 0.709 0.644 0.629 Namibia 0.697 0.760 0.823 Peru 0.771 0.771 0.738 Niger 0.767 0.704 0.705 Nigeria 0.707 0.700 0.693 Middle East & North Africa Rwanda 0.727 0.753 0.715 Egypt, Arab Rep. 0.781 0.746 0.656 Senegal 0.820 0.810 0.792 Jordan 0.769 0.662 0.543 Sierra Leone n.a 0.764 0.597 South Africa 0.877 0.836 0.592 South Asia Tanzania 0.581 0.672 0.711 Afghanistan n.a 0.725 0.561 Togo 0.832 0.839 0.837 Bangladesh 0.854 0.861 0.718 Uganda 0.784 0.755 0.704 India 0.860 0.759 0.686 Zambia 0.726 0.777 0.741 Nepal 0.680 0.684 0.640 Zimbabwe 0.688 0.687 0.721 Pakistan 0.823 0.814 0.762 Notes: This table presents the estimated share of the intergenerational component using the Bingley and Cappellari (2019) method for each of the 53 developing countries in the Demographic and Health Surveys (DHS) dis- aggregated by different decades of birth cohorts (the 1970s, 1980s, and 1990s). 38 Table A1: Countries from the Demographic and Health Surveys (DHS) Region Country Used Region Country Used Cambodia Yes Angola No Indonesia Yes Benin Yes Lao PDR No Botswana No Myanmar No Burkina Faso Yes East Asia & Papua New Guinea No Burundi Yes Pacific Philippines Yes Cameroon Yes Samoa No Cape Verde No Thailand No Central African Republic No Timor-Leste No Chad Yes Vietnam Yes Comoros No Albania Yes Congo Yes Congo Democratic Armenia Yes No Republic Azerbaijan No Cote d'Ivoire Yes Georgia No Equatorial Guinea No Kazakhstan No Eritrea No Europe & Kyrgyz Republic Yes Eswatini No Central Asia Moldova No Ethiopia Yes Romania No Gabon Yes Tajikistan Yes Gambia No Turkey Yes Ghana Yes Turkmenistan No Guinea Yes Sub-Saharan Ukraine No Kenya Yes Africa Uzbekistan No Lesotho Yes Bolivia Yes Liberia Yes Brazil Yes Madagascar Yes Colombia Yes Malawi Yes Dominican Republic Yes Mali Yes Ecuador No Mauritania No El Salvador No Mozambique Yes Guatemala Yes Namibia Yes Latin America Guyana No Niger Yes & Caribbean Haiti Yes Nigeria Yes Honduras No Nigeria (Ondo State) No Jamaica No Rwanda Yes Mexico No Sao Tome and Principe No Nicaragua No Senegal Yes Paraguay No Sierra Leone Yes Peru Yes South Africa Yes Trinidad and Tobago No Sudan No Egypt Yes Tanzania Yes Jordan Yes Togo Yes Middle East & Morocco No Uganda Yes North Africa Tunisia No Zambia Yes Yemen No Zimbabwe Yes Afghanistan Yes Bangladesh Yes India Yes South Asia Maldives No Nepal Yes Pakistan Yes Sri Lanka No Notes: Data come from the Demographic and Health Surveys (DHS). 53 countries are used and accessed between April 2021 and July 2021. 42 countries are not used in the analytic sample where at least one DHS survey is available but the sample size is small. The total number of observations used in the analytic sample is 544624. 39 Table A2: Country-Specific Sibling Correlation Estimates (1970 Cohort) Sibling Sibling Country S.E. N Country S.E. N Corr. Corr. East Asia & Pacific Sub-Saharan Africa Cambodia 0.674 0.013 2271 Benin 0.653 0.020 1028 Indonesia 0.659 0.006 12924 Burkina Faso 0.584 0.022 1314 Philippines 0.614 0.012 4435 Burundi n.a n.a n.a Vietnam 0.682 0.013 2286 Cameroon 0.608 0.037 399 Chad 0.743 0.024 658 Europe & Central Asia Congo, Rep. 0.466 0.050 275 Albania n.a n.a n.a Cote d'Ivoire 0.633 0.030 507 Armenia 0.468 0.034 1011 Ethiopia 0.793 0.010 2106 Kyrgyz Republic 0.379 0.057 468 Gabon 0.475 0.037 478 Tajikistan n.a n.a n.a Ghana 0.592 0.025 980 Turkey 0.550 0.013 3375 Guinea 0.597 0.033 635 Kenya 0.560 0.022 1904 Latin America & Caribbean Lesotho 0.592 0.024 787 Bolivia 0.692 0.011 3329 Liberia n.a n.a n.a Brazil 0.709 0.014 1843 Madagascar 0.824 0.011 1658 Colombia 0.666 0.011 4009 Malawi 0.609 0.020 1285 Dominican 0.595 0.013 3097 Mali 0.626 0.027 842 Republic Guatemala 0.750 0.011 2117 Mozambique 0.526 0.025 1347 Haiti 0.697 0.015 1272 Namibia 0.554 0.028 846 Peru 0.668 0.007 10144 Niger 0.670 0.053 158 Nigeria 0.643 0.040 404 Middle East & North Africa Rwanda 0.546 0.021 1521 Egypt, Arab Rep. 0.523 0.009 9541 Senegal 0.588 0.026 838 Jordan 0.474 0.014 4118 Sierra Leone n.a n.a n.a South Africa 0.521 0.025 1325 South Asia Tanzania 0.559 0.023 1617 Afghanistan n.a n.a n.a Togo 0.446 0.029 1130 Bangladesh 0.673 0.009 4275 Uganda 0.648 0.023 876 India 0.658 0.006 13288 Zambia 0.664 0.016 1710 Nepal 0.638 0.016 1816 Zimbabwe 0.569 0.026 1150 Pakistan 0.595 0.009 6202 Notes: This table presents the sibling correlation estimates for each of the 53 developing countries in the Demographic and Health Surveys (DHS) using the 1970s birth cohort sample. 40 Table A3: Country-Specific Sibling Correlation Estimates (1980 Cohort) Sibling Sibling Country S.E. N Country S.E. N Corr. Corr. East Asia & Pacific Sub-Saharan Africa Cambodia 0.657 0.009 6402 Benin 0.595 0.016 2439 Indonesia 0.616 0.007 10148 Burkina Faso 0.638 0.017 1636 Philippines 0.616 0.011 4746 Burundi 0.584 0.021 1396 Vietnam 0.616 0.021 1232 Cameroon 0.629 0.017 1755 Chad 0.727 0.020 887 Europe & Central Asia Congo, Rep. 0.532 0.022 1212 Albania 0.502 0.032 1106 Cote d'Ivoire 0.592 0.029 620 Armenia 0.592 0.022 1735 Ethiopia 0.656 0.011 3006 Kyrgyz Republic 0.438 0.036 575 Gabon 0.614 0.034 485 Tajikistan 0.461 0.023 1469 Ghana 0.617 0.020 1588 Turkey 0.561 0.015 3265 Guinea 0.515 0.028 983 Kenya 0.584 0.017 2587 Latin America & Caribbean Lesotho 0.592 0.013 3001 Bolivia 0.653 0.015 3118 Liberia 0.535 0.027 787 Brazil n.a n.a n.a Madagascar 0.790 0.010 2084 Colombia 0.582 0.009 9258 Malawi 0.602 0.015 2470 Dominican 0.538 0.014 4279 Mali 0.619 0.019 1595 Republic Guatemala 0.697 0.016 1281 Mozambique 0.522 0.024 1176 Haiti 0.674 0.014 2594 Namibia 0.535 0.029 1072 Peru 0.591 0.006 19330 Niger 0.684 0.019 1031 Nigeria 0.666 0.011 4833 Middle East & North Africa Rwanda 0.550 0.015 3381 Egypt, Arab Rep. 0.581 0.008 11306 Senegal 0.584 0.016 2157 Jordan 0.468 0.012 7316 Sierra Leone 0.517 0.026 1029 South Africa 0.383 0.050 343 South Asia Tanzania 0.563 0.018 1952 Afghanistan 0.523 0.015 2727 Togo 0.570 0.039 430 Bangladesh 0.614 0.010 5228 Uganda 0.666 0.017 1489 India 0.631 0.003 53132 Zambia 0.681 0.018 1325 Nepal 0.618 0.016 2171 Zimbabwe 0.609 0.021 1481 Pakistan 0.633 0.004 27858 Notes: This table presents the sibling correlation estimates for each of the 53 developing countries in the Demographic and Health Surveys (DHS) using the 1980s birth cohort sample. 41 Table A4: Country-Specific Sibling Correlation Estimates (1990 Cohort) Sibling Sibling Country S.E. N Country S.E. N Corr. Corr. East Asia & Pacific Sub-Saharan Africa Cambodia 0.685 0.016 1751 Benin 0.579 0.019 1915 Indonesia 0.523 0.014 4540 Burkina Faso 0.506 0.055 241 Philippines 0.550 0.014 4301 Burundi 0.500 0.018 2484 Vietnam n.a n.a n.a Cameroon 0.629 0.021 1502 Chad 0.760 0.018 1058 Europe & Central Asia Congo, Rep. 0.625 0.043 291 Albania 0.389 0.036 1115 Cote d'Ivoire 0.640 0.057 154 Armenia 0.492 0.065 428 Ethiopia 0.630 0.016 1935 Kyrgyz Republic 0.342 0.094 304 Gabon 0.576 0.056 249 Tajikistan 0.424 0.030 1744 Ghana 0.500 0.036 577 Turkey 0.433 0.038 742 Guinea 0.559 0.026 949 Kenya 0.503 0.019 2568 Latin America & Caribbean Lesotho 0.483 0.036 593 Bolivia n.a n.a n.a Liberia 0.522 0.026 909 Brazil n.a n.a n.a Madagascar n.a n.a n.a Colombia 0.546 0.014 3729 Malawi 0.585 0.017 2178 Dominican 0.435 0.044 540 Mali 0.573 0.022 1332 Republic Guatemala 0.680 0.012 2690 Mozambique 0.681 0.041 279 Haiti 0.678 0.013 2144 Namibia 0.516 0.063 279 Peru 0.546 0.019 2645 Niger 0.645 0.041 308 Nigeria 0.741 0.010 5396 Middle East & North Africa Rwanda 0.522 0.023 1362 Egypt, Arab Rep. 0.588 0.018 2934 Senegal 0.588 0.040 354 Jordan 0.551 0.015 5047 Sierra Leone 0.515 0.026 1139 South Africa 0.501 0.043 578 South Asia Tanzania 0.484 0.030 1213 Afghanistan 0.591 0.011 4857 Togo 0.519 0.038 410 Bangladesh 0.609 0.018 1712 Uganda 0.552 0.021 1710 India 0.631 0.003 72192 Zambia 0.620 0.018 2011 Nepal 0.588 0.024 1317 Zimbabwe 0.625 0.037 518 Pakistan 0.679 0.009 5583 Notes: This table presents the sibling correlation estimates for each of the 53 developing countries in the Demographic and Health Surveys (DHS) using the 1990s birth cohort sample. 42 Table A5: Country-Specific Proportion of Sibling Correlations Explained by Intergenerational Transmission Estimates (1970 Cohort) Country Prop. S.E. N Country Prop. S.E. N East Asia & Pacific Sub-Saharan Africa Cambodia 0.731 0.024 2271 Benin 0.867 0.037 1028 Indonesia 0.787 0.009 12924 Burkina Faso 0.844 0.043 1314 Philippines 0.816 0.021 4435 Burundi n.a n.a n.a Vietnam 0.829 0.022 2286 Cameroon 0.742 0.073 399 Chad 0.735 0.039 658 Europe & Central Asia Congo, Rep. 0.926 0.134 275 Albania n.a n.a n.a Cote d'Ivoire 0.737 0.050 507 Armenia 0.940 0.076 1011 Ethiopia 0.749 0.017 2106 Kyrgyz Republic 0.496 0.112 468 Gabon 0.890 0.096 478 Tajikistan n.a n.a n.a Ghana 0.853 0.046 980 Turkey 0.819 0.026 3375 Guinea 0.452 0.060 635 Kenya 0.656 0.036 1904 Latin America & Caribbean Lesotho 0.716 0.048 787 Bolivia 0.761 0.017 3329 Liberia n.a n.a n.a Brazil 0.758 0.022 1843 Madagascar 0.859 0.017 1658 Colombia 0.778 0.017 4009 Malawi 0.764 0.037 1285 Dominican 0.701 0.023 3097 Mali 0.855 0.053 842 Republic Guatemala 0.759 0.017 2117 Mozambique 0.914 0.056 1347 Haiti 0.709 0.028 1272 Namibia 0.697 0.052 846 Peru 0.771 0.011 10144 Niger 0.767 0.093 158 Nigeria 0.707 0.054 404 Middle East & North Africa Rwanda 0.727 0.043 1521 Egypt, Arab Rep. 0.781 0.017 9541 Senegal 0.820 0.054 838 Jordan 0.769 0.032 4118 Sierra Leone n.a n.a n.a South Africa 0.877 0.054 1325 South Asia Tanzania 0.581 0.034 1617 Afghanistan n.a n.a n.a Togo 0.832 0.069 1130 Bangladesh 0.854 0.015 4275 Uganda 0.784 0.041 876 India 0.860 0.009 13288 Zambia 0.726 0.028 1710 Nepal 0.680 0.024 1816 Zimbabwe 0.688 0.046 1150 Pakistan 0.823 0.017 6202 Notes: This table presents the estimated share of the intergenerational component using the Bingley and Cappellari (2019) method for each of the 53 developing countries in the Demographic and Health Surveys (DHS) using the 1970s birth cohort sample. 43 Table A6: Country-Specific Proportion of Sibling Correlations Explained by Intergenerational Transmission Estimates (1980 Cohort) Country Prop. S.E. N Country Prop. S.E. N East Asia & Pacific Sub-Saharan Africa Cambodia 0.696 0.014 6402 Benin 0.840 0.028 2439 Indonesia 0.791 0.012 10148 Burkina Faso 0.688 0.029 1636 Philippines 0.830 0.019 4746 Burundi 0.740 0.041 1396 Vietnam 0.867 0.036 1232 Cameroon 0.841 0.028 1755 Chad 0.753 0.033 887 Europe & Central Asia Congo, Rep. 0.863 0.048 1212 Albania 0.858 0.065 1106 Cote d'Ivoire 0.845 0.055 620 Armenia 0.746 0.035 1735 Ethiopia 0.781 0.021 3006 Kyrgyz Republic 0.786 0.086 575 Gabon 0.672 0.057 485 Tajikistan 0.700 0.055 1469 Ghana 0.781 0.032 1588 Turkey 0.767 0.024 3265 Guinea 0.787 0.067 983 Kenya 0.746 0.030 2587 Latin America & Caribbean Lesotho 0.727 0.025 3001 Bolivia 0.709 0.020 3118 Liberia 0.680 0.050 787 Brazil n.a n.a n.a Madagascar 0.816 0.015 2084 Colombia 0.755 0.015 9258 Malawi 0.767 0.026 2470 Dominican 0.727 0.023 4279 Mali 0.745 0.032 1595 Republic Guatemala 0.796 0.026 1281 Mozambique 0.765 0.047 1176 Haiti 0.644 0.021 2594 Namibia 0.760 0.053 1072 Peru 0.771 0.010 19330 Niger 0.704 0.032 1031 Nigeria 0.700 0.016 4833 Middle East & North Africa Rwanda 0.753 0.032 3381 Egypt, Arab Rep. 0.746 0.013 11306 Senegal 0.810 0.033 2157 Jordan 0.662 0.025 7316 Sierra Leone 0.764 0.049 1029 South Africa 0.836 0.136 343 South Asia Tanzania 0.672 0.032 1952 Afghanistan 0.725 0.030 2727 Togo 0.839 0.070 430 Bangladesh 0.861 0.017 5228 Uganda 0.755 0.029 1489 India 0.759 0.005 53132 Zambia 0.777 0.029 1325 Nepal 0.684 0.023 2171 Zimbabwe 0.687 0.034 1481 Pakistan 0.814 0.007 27858 Notes: This table presents the estimated share of the intergenerational component using the Bingley and Cappellari (2019) method for each of the 53 developing countries in the Demographic and Health Surveys (DHS) using the 1980s birth cohort sample. 44 Table A7: Country-Specific Proportion of Sibling Correlations Explained Explained by Intergenerational Transmission Estimates (1990 Cohort) Country Prop. S.E. N Country Prop. S.E. N East Asia & Pacific Sub-Saharan Africa Cambodia 0.659 0.020 1751 Benin 0.809 0.031 1915 Indonesia 0.774 0.023 4540 Burkina Faso 0.727 0.095 241 Philippines 0.810 0.026 4301 Burundi 0.639 0.033 2484 Vietnam n.a n.a n.a Cameroon 0.871 0.032 1502 Chad 0.677 0.024 1058 Europe & Central Asia Congo, Rep. 0.728 0.061 291 Albania 0.762 0.080 1115 Cote d'Ivoire 0.732 0.078 154 Armenia 0.555 0.084 428 Ethiopia 0.679 0.023 1935 Kyrgyz Republic 0.501 0.159 304 Gabon 0.775 0.095 249 Tajikistan 0.587 0.053 1744 Ghana 0.796 0.066 577 Turkey 0.701 0.075 742 Guinea 0.753 0.062 949 Kenya 0.877 0.041 2568 Latin America & Caribbean Lesotho 0.817 0.070 593 Bolivia n.a n.a n.a Liberia 0.699 0.048 909 Brazil n.a n.a n.a Madagascar n.a n.a n.a Colombia 0.706 0.022 3729 Malawi 0.745 0.027 2178 Dominican 0.838 0.093 540 Mali 0.699 0.034 1332 Republic Guatemala 0.722 0.019 2690 Mozambique 0.759 0.055 279 Haiti 0.629 0.021 2144 Namibia 0.823 0.115 279 Peru 0.738 0.028 2645 Niger 0.705 0.056 308 Nigeria 0.693 0.011 5396 Middle East & North Africa Rwanda 0.715 0.043 1362 Egypt, Arab Rep. 0.656 0.025 2934 Senegal 0.792 0.069 354 Jordan 0.543 0.023 5047 Sierra Leone 0.597 0.039 1139 South Africa 0.592 0.072 578 South Asia Tanzania 0.711 0.052 1213 Afghanistan 0.561 0.017 4857 Togo 0.837 0.073 410 Bangladesh 0.718 0.025 1712 Uganda 0.704 0.037 1710 India 0.686 0.004 72192 Zambia 0.741 0.027 2011 Nepal 0.640 0.032 1317 Zimbabwe 0.721 0.054 518 Pakistan 0.762 0.013 5583 Notes: This table presents the estimated share of the intergenerational component using the Bingley and Cappellari (2019) method for each of the 53 developing countries in the Demographic and Health Surveys (DHS) using the 1990s birth cohort sample. 45