Intergenerational Education Mobility in Africa: Has Progress Been Inclusive?

This paper employs nationally representative household survey data on parents of adult individuals to analyze the intergenerational transmission of education in nine Sub-Saharan African countries. The paper provides the levels, trends, and patterns of intergenerational persistence of educational attainment over 50 years, with a special focus on gender differences. The study finds a declining cohort trend in the intergenerational educational persistence in all the countries, particularly after the 1960s. The increase in educational mobility coincides with drastic changes in educational systems and a huge investment in human capital accumulation in the region following independence. Nevertheless, the education of parents' remains a strong determinant of educational outcomes among the children in all the countries. Ghana, Guinea, Nigeria, and Uganda experienced the highest intergenerational mobility, and the Comoros and Madagascar the lowest. In all the sample countries, more mobility is observed in the lower tail of the distribution of education. Intergenerational educational persistence is strong from mothers to children, and the effect is more pronounced among daughters than sons. The results highlight the need for targeted redistributive policies that improve intergenerational mobility in the region.

This paper employs nationally representative household survey data on parents of adult individuals to analyze the intergenerational transmission of education in nine Sub-Saharan African countries. The paper provides the levels, trends, and patterns of intergenerational persistence of educational attainment over 50 years, with a special focus on gender differences. The study finds a declining cohort trend in the intergenerational educational persistence in all the countries, particularly after the 1960s. The increase in educational mobility coincides with drastic changes in educational systems and a huge investment in human capital accumulation in the region following independence.
Nevertheless, the education of parents' remains a strong determinant of educational outcomes among the children in all the countries. Ghana, Guinea, Nigeria, and Uganda experienced the highest intergenerational mobility, and the Comoros and Madagascar the lowest. In all the sample countries, more mobility is observed in the lower tail of the distribution of education. Intergenerational educational persistence is strong from mothers to children, and the effect is more pronounced among daughters than sons. The results highlight the need for targeted redistributive policies that improve intergenerational mobility in the region.

Introduction
Since the mid-1990s, the resurgence of growth has been remarkable in Sub-Sharan Africa. For instance, between 2000 and 2012, the region experienced a 3 percent per capita annual growth rate in Gross Domestic Product (Thorbecke, 2013). Recent evidence indicates that, while the observed growth has led to considerable poverty reduction, it has been accompanied by a rise in inequality in a number of countries, including Kenya, Uganda, and Zambia (Fosu, 2015). Accordingly, there is growing concern that the benefits of economic growth are not shared broadly. For policy making, it is important to understand whether the increase in inequality is an outcome of an economic structure that rewards hard work and risk taking, or whether it is a reflection of the existence of inequality of opportunity within society. The rise in inequality becomes a policy concern if it is an outcome of inequality of opportunity among individuals with different initial circumstances; for example children from poor families with ability and talent are unable to move beyond the position of their parents on the economic ladder through their own effort and choices (Rawls, 1971). Intergenerational persistence in socioeconomic status is the main mechanism through which inequality of opportunity persist in a society. For example, social mobility may differ based on gender, race, ethnicity, or region, suggesting differential access to opportunity across groups within a society. Equality of opportunity has come to be a key condition if a society is to achieve an acceptable level of equity: in its strongest form, equality of opportunity is a more relevant aspect of policy in a society than of inequalities of outcomes (Kanbur and Wagstaff, 2015;Rama et al., 2015). Greater inequality of opportunity may result in greater inequality in a society and affect public attitudes toward other social objectives such as growth and poverty reduction (Atkinson, 1980;Piketty, 1995;Corak, 2013). Because of this, the extent to which socioeconomic outcomes are transmitted from one generation to the next has long been of interest among development economists and policy makers.
Although understanding intergenerational mobility is important for policy, economic analysis of intergenerational mobility in developing countries is only in its infancy because of lack of appropriate data. In particular, intergenerational mobility studies on Sub-Saharan Africa are scarce. 1 The current study aims to fill this gap by taking advantage of recent nationally representative data that provide information on the social origin of adult individuals in nine Sub-Saharan Africa countries. We use education as an indicator of economic status. Using education as an indicator of economic status has four main advantages. First, the literature in both developed and developing countries identifies education as an important driver of labor market participation and, hence, income, more years of schooling is usually associated with higher income (see Chevalier et al., 2003;Blanden et al., 2005;Black and Devereux, 2011). Understanding the trends, levels and patterns of the persistence of the education attainment across generations therefore sheds light on overall mobility in economic status in a society. Second, schooling is an outcome on which it seems reasonable to assume respondents can reliably report on their parents. Third, data restrictions, especially in developing countries, are much less stringent; retrospective information on parents' education has been more widely available recently than information on parental incomes or occupations. Finally, using income to proxy socioeconomic outcomes in developing countries is found to be problematic. There is a serious concern about persistence of measurement errors in consumption and income data from developing countries (see Deaton, 1997;Glewwe, 2005;Lee, 2009 for detailed discussion). Education is also considered a vital policy instrument in the creation of equal access to economic opportunity, leading to higher social mobility and economic progress (Bowles, 1972;Becker and Tomes, 1979;Piketty, 1995;Hertz et al., 2007;Corak, 2013;Reeves, 2014). Furthermore, the region serves as an excellent case study for intergenerational education mobility. It endorsed the United Nations goal of Education for All agreed in Dakar in 2000, a commitment to provide quality basic education for all children and adults by 2015. Accordingly, many countries in the region undertook major reforms in education systems, including the abolition of school fees (Thakur, 1991;Tomasevski, 2006). Thus, gauging how such policy changes induce intergenerational mobility has important policy implications.
Drawing on nationally representative survey data, the study analyzes the trends, levels, and patterns of intergenerational im(mobility) in educational attainment in the Comoros, Ghana, Guinea, Madagascar, Malawi, Nigeria, Rwanda, Tanzania, and Uganda over 50 years, with a special focus on gender differences. The paper contributes to the existing literature in several ways. First, the study extends the existing evidence by creating comparable recent estimates for nine Sub-Saharan African countries so that we may begin to draw conclusions about the inclusiveness of the recent investment in education in the sampled countries. Using two widely applied measures of intergenerational persistence, we provide persistence estimates for 10 successive five-year birth cohorts at the aggregate and gender levels in each country. Second, by closely following the methodology of the two closest antecedent studies in developing countries (Hertz et al., 2007;Azam and Bhatt, 2015), we are able to rank countries in terms of intergenerational educational mobility. To the best of our knowledge, there exist no comparable estimates on these countries. 2 Third, unlike other studies in developing countries, the estimates presented in this study do not suffer from selection bias caused by imposing co-residence to construct parent and child pairs. Most of the studies on intergenerational mobility still rely on cohabitation to identify parent and child pairs. Using this method has two major consequences. First, using coresidents to identify parent and child pairs leads to a sample selection that biases the intergenerational elasticity downward. For instance, Francesconi and Nicoletti (2006) and Azam and Bhatt (2015) document a substantial bias in constructing father and son pairs in the United Kingdom and India, respectively. Second, coresidence over represents younger adults who are still living with parents, which restricts the analysis to an unrepresentative young population (Jalan and Murgai, 2007;Hnatkovska et al., 2013). The current study addresses this issue by using nationally representative data on educational attainment among adult individuals and their parents regardless of whether parents are alive or, if alive, reside in the same household. 3 Fourth, the existing evidence on the link between child and parental education by gender is largely unexplored. A handful of studies examine the intergenerational persistence of economic status between parents and daughters (see Grusky and DiPrete, 1990;Chadwick and Solon, 2002). In this study, we attempt to fill this gap in the literature and provide gender estimates (daughters-mothers, sons-mothers, daughters-fathers, and sons-fathers) of intergenerational educational persistence by five-year birth cohorts in each country.
The study uses two measures of intergenerational educational mobility: intergenerational elasticity and the partial correlation coefficient. The analysis shows the trends in intergenerational educational mobility across five-year birth cohorts for each sex in each country. There are several findings. Comparing the highest educational attainment, both measures accord in pointing out the importance of parental education in determining the educational attainment of children in all the countries. We find a declining cohort trend in the estimated intergenerational elasticity in all the countries, particularly after the 1960s. This implies greater educational mobility among more recent birth cohorts in all the countries. The declining trend after the 1960s coincides with the drastic changes in educational systems and the huge investment in human capital accumulation in the region since independence. We note a country difference: Nigeria, Guinea, Ghana, and Uganda experienced the highest intergenerational mobility, and the Comoros and Madagascar the lowest. The decline in intergenerational education persistence is strongest in the lower tail of the education distribution, and, daughters' educational attainment is more correlated with parental education. The greater intergenerational persistence among women compared with men is consistent with previous findings in other developing countries (Thomas, 1996;Branson et al., 2012;Ranasinghe, 2015;Emran and Shilpi, 2015). Furthermore, in all the countries except the Comoros, intergenerational persistence from mothers to children is stronger. This result contrasts with evidence from South Africa where the link between children and father's education is stronger than or the same as that of mothers (Lam, 1999;Girdwood and Leibbrandt, 2009). In line with the findings of Hertz et al. (2007) for 42 countries and Azam and Bhatt (2015) for India, we also show that the correlation coefficient between parents and child's schooling has been increasing or remaining constant across cohorts, mainly driven by educational inequality in the parents' generation. This result is not surprising in our context given that the correlation coefficient provides an absolute measure of intergenerational persistence after account is taken of a possible improvement in the distribution of education attainment because of education system reforms, such as the abolition of school fees, which increase average schooling and reduce variation in schooling. The education systems of all the countries in our sample have changed drastically since the 1960s. From a policy perspective, our result highlights the demand for targeted redistributive policies that can improve intergenerational mobility in the region. Moreover, putting in place a favorable environment for women who are less well off in terms of education might play a decisive role in promoting social mobility not only in the short run, but also in the next generation.
The rest of the paper proceeds as follows. Section 2 reviews the literature and places the study in the context of existing literature. Section 3 presents the analytical framework. Section 4 describes the data. Section 5 presents the results. Section 6 offers concluding remarks and describes potential policy implications.

Related literature
Intergenerational social mobility refers to the ability of children to climb higher than their parents on the socio economic status ladder when they become adults. Although the literature has widely focused on income, several social outcomes such as education, social class, health, or occupation can be used to study intergenerational mobility in a society (Bhalotra and Rawlings, 2011;Causa and Johansson, 2011;Ferreira et al., 2012;Bauer and Riphahn, 2007). Indeed, education is viewed as the main shaper of all other adulthood opportunities (Stiglitz, 2012). The literature on intergenerational mobility in education, occupation, and income is broad and extensive in developed countries. Black and Devereux (2011) update the previous works of Solon (1992Solon ( , 1999 and present a survey of the literature, along with the methodological challenges of the existing evidence in developed economies. Recent contributions on education mobility in developed countries include Ranasinghe (2015), Johnston et al. (2014), Checchi et al. (2013), andCobb-Clark andNguyen (2010). Hertz et al. (2007) extend the analysis of intergenerational educational mobility to 42 countries, including 19 developing countries, among which three are Sub-Saharan African countries (Ethiopia, Ghana, and South Africa), and present trends over 50 years. They find that the intergenerational regression coefficient has fallen over time, implying a high degree of intergenerational education mobility, but the correlation in educational attainment between children and their parents' remained unchanged over the period. They also document considerable regional differences, with Nordic and Latin American countries displaying the highest and the lowest intergenerational education mobility, respectively. Daude and Robano (2015) study education mobility in 18 Latin American countries and confirm the finding of Hertz et al. (2007). Hnatkovska et al. (2013) study education and occupation mobility in India by caste. They conclude that structural changes in India have coincided with a breaking down of caste-based barriers to socioeconomic mobility. Azam and Bhatt (2015) examine the intergenerational transmission of education in India and report a decline in educational persistence between fathers and sons over the last 45 years. In contrast, Emran and Shilpi (2015) find that India shows greater intergenerational education persistence than Latin America and that educational mobility remained unchanged between 1991 and 2006.
With the exception of South Africa, studies on the intergenerational transmission of education in Africa are almost nonexistent. 4 Some important early contributions on South Africa include Thomas (1996), Lam (1999), Case and Deaton (1999), Nimubona and Vencatachellum (2007), and Girdwood and Leibbrandt (2009). Overall, the studies find that parental education determines education outcomes among children and that there is substantial education persistence in the country, especially among black South Africans. Recent studies (Branson et al., 2012;Kwenda et al., 2015) document a decrease in intergenerational transmission of education over the last five decades in the country. To the best of our knowledge, the only cross-country study on intergenerational education mobility that includes other African countries is Hertz et al. (2007). Using data from Ethiopia, Ghana, and South Africa, the authors present evidence of lower educational persistence in African countries compared with Latin American countries. However, the smaller educational persistence rate in educational attainments between children and parents in these countries is not necessarily indicative of greater mobility. Rather, parental education dispersion was limited because of the low educational level in the population during their study period.
The evidence on education mobility across generations by sex is mixed. Lam (1999) finds that the effect of a mother's education on a child's education is no different relative to that of a father in South Africa. In contrast, Kwenda et al. (2015), Thomas (1996), andBranson et al. (2012) on South Africa, Ranasinghe (2015) and Crook (1995) on Australia, and Björklund et al. (2006) on Sweden present empirical evidence that the schooling of mothers has a bigger effect on children's educational attainment than that of fathers, while Girdwood andLeibbrandt (2009), Plug (2004), and Behrman and Rosenzweig (2005) find the paternal effect to be strong in the United States. The current study aims to extend the evidence for on nine Sub-Saharan African countries with recent data and provide an in-depth analysis of intergenerational education mobility over the long term. The study therefore complements the growing literature on international comparisons of intergenerational education mobility and fills the hitherto overlooked aspect of intergenerational mobility in developing countries particularly in Sub-Saharan Africa.

Analytical framework
There are several theoretical arguments on how parental educational background affects children's education. Educational decisions on children are determined by parental preference and credit constraints Tomes, 1979, 1986;Solon, 1992Solon, , 1999. These theoretical arguments identify many possible channels through which parental education affects children's education. For instance, parents affect their children through innate ability, which has an impact on educational attainment, an aspect first formalized by the seminal work of Becker and Tomes (1979). The nutrition and health status of a mother during pregnancy have a huge impact on a child's initial health endowments and, hence, outcomes in adulthood, including education (Currie, 2009(Currie, , 2011Hackman et al., 1983). For instance, a positive relationship between a mother's education and a child's birthweight, which is a strong predictor of health outcomes in adulthood, is found throughout the world (Currie and Moretti, 2003). The abilities of parents affect their own income and education outcomes, which determine the quality and quantity of investment in children, thereby affecting the educational attainment of the children Tomes, 1979, 1986). First, welleducated parents generally earn higher incomes, which may increase the investment in a child's education by relaxing resource constraints. Second, higher educational attainment may improve the productivity of parents in child development, thus enhancing activities that may positively affect the educational attainment of children. Finally, parental education directly influences the schooling of children through the choice of school, with an expectation that more able families send their children to more well-endowed schools. In this study, we are not trying to investigate the channels through which intergenerational educational correlations emerge. Our objective is to correlate the educational attainment of parents and children and to present comparisons of trends in intergenerational educational im(mobility) over time.

Identification issues
Intergenerational mobility studies have been fraught with econometric challenges that have arisen because of unobservable heterogeneity, including the inheritance of genetic endowments such as ability and preference across generations. The partial correlation observed in the data might be mainly driven by the transmission of preference and ability between parents and children. Previous studies attribute the partial, but high correlations between parents' and children's educational outcomes to nature and nurture, among other factors (Becker and Tomes, 1986;Haveman and Wolfe, 1995;Black and Devereux, 2011;Checchi et al., 2013). Nature refers to a genetic transmission of the ability of a parent to a child. Able parents have a higher chance of producing to have more able children who can attain higher levels of education without special parental investment. For instance, a child might learn skills through observation without any additional effort from parents (Haveman and Wolfe, 1995;Basu and Getachew, 2015). Nurture pertains to the amount of time and economic investments of parents on a child's human capital accumulation.
The standard approach in tackling unobserved heterogeneity is to use instrumental variables. The challenge is to identify exogenous variables that affect parental educational attainment, but do not have any effect on children's educational attainment. However, the instrumental variables used widely in the literature such as family background variables tend to affect children's outcomes, including our main interest here, education. Some studies use data on adoption (Plug, 2004;Plug and Vijverberg, 2003) and twins (Behrman and Rosenzweig, 2005) to isolate the effect of nature from the effect of nurture. However, these studies are limited to developed countries, where reliable data are available. Other studies compare the effect of nature and nurture on social mobility and find that nurture is relatively more important in explaining parent-child education transmission (Checchi et al., 2013;Haveman and Wolfe, 1995). In the absence of quasi-experimental data and credible instruments, we limit our analysis to the correlation between the educational attainment of parents and children. If factors of nature are time invariant, analyzing changes in intergenerational educational mobility over time is policy relevant without differentiating the effect of nature and nature (Heineck and Riphahn, 2009). Moreover, we are not aware of any analysis of intergenerational educational mobility in our sample countries. Thus, the pattern of partial correlation over time may be of independent interest.

Estimation strategies
In the literature, the measurement of the degree to which family educational background affects the educational attainment of children has been accomplished in different ways (see Fields and Ok, 2000;Ferreira et al., 2012). Perhaps the most basic measures are intergenerational correlation and intergenerational elasticity. The standard OLS regression model that relates educational attainment transmission from parents to children allows an estimation of these measures: where i = 1, · · · , I indexes families and j = 1, · · · , J children; E ij denotes years of schooling of a child j in a family i; EP i is the parental years of schooling in a family i; and β is the intergenerational regression coefficient, which is the parameter of interest; ε is a mean zero error term that is independently and identically distributed across generations and individuals. Equation 1 allows the quantification of the importance of parental educational attainment on children's years of schooling using two measures. The first measure is intergenerational elasticity (β). Intergenerational elasticity (IGE) shows the relationship between each additional year of schooling of the parents and their children.β measures intergenerational persistence, and 1 −β is a measure of intergenerational mobility. Higher-value intergenerational elasticity indicates higher intergenerational persistence and, hence, lower mobility. In this study, the estimations are carried out on five-year birth cohorts by sex in each country. Thus,β is the estimated intergenerational elasticity of each of five-year birth cohort among sons and daughters. Comparingβ across birth cohorts in each country measures how intergenerational education persistence has evolved in both sexes over time.
The intergenerational education correlation between E ij and EP i is an alternative measure of intergenerational elasticity that has also been widely used in the literature. The correlation coefficient (ρ) quantifies how much of the observed dispersion in children's education is explained by parental education. A higher value in the correlation coefficient also implies lower intergenerational mobility and higher intergenerational education persistence. Intergenerational elasticity equals the correlation coefficient between parent and child education weighted by the ratio of the standard deviations of education across generations. Thus the two measures, the correlation and the elasticity, will be equal provided that the standard deviation of years of schooling is the same across generations. The relationship between the two measures is as follows:β where σ p and σ c are the standard deviations of years of schooling of parents and children in each five-year birth cohort; σ pc is the covariance between the years of schooling of parents and children; and ρ pc is the correlation between the schooling of parents and children. An estimate of ρ that equals to 1 implies perfect intergenerational immobility, that is, child educational attainment is entirely influenced by the educational background of parents, while a ρ close to zero indicates a perfectly mobile society in which parental education has only limited or no effect on children's educational attainment. A decrease (increase) in intergenerational elasticity (β) may arise because of either a decrease (increase) in intergenerational correlation (ρ) or a decrease (increase) in the inequality of education across generations ( σc σp ). Thus, the main difference between IGE (β) and the correlation coefficient (ρ) is that the former factors out the cross-sectional inequality of education across generations and, hence, provides a relative measure of intergenerational mobility. In contrast, the estimated elasticity (β) provides an absolute measure of intergenerational persistence that is not affected by education policy changes in a country, for instance, the expansion of compulsory free primary education, and this reduces the possible variation in the measure. Hence, a change in the inequality of education across the generation of the parents and children will cause the two measures to evolve differently over time. Checchi et al. (2013) argue that a change inρ captures not only a change in the parent-child education correlation, but also other events in the education system, such as the expansion of compulsory primary education. To disentangle the effects of these events from the educational correlation between parents and children, they propose decomposingρ into three components: changes in the dispersion of the educational attainment of parents and children around the respective means, changes in children's educational attainment conditional on the educational attainment of the parents, and changes in the unconditional distribution of parental educational attainment. They argue that changes in children's schooling conditional on parent's education is the most relevant for policy. In the same vein, we model the effect of the highest levels of education of parents on the highest level of schooling of the children, across the five-year birth cohorts using an ordered probit model.
Let's define the educational attainment of children in a household as follows: where i = 1, · · · , I indexes families and j = 1, · · · , J children; E ij is the years of schooling of a child j in a family i; µ is the population mean; a i is a family component common to all children in a household i; and b ij is the individual specific component for a child that captures i s deviation from the family component. Because the individual component (b ij ) is orthogonal to the family component (a i ), one can express the family component as follows: where z i denotes family factors that are orthogonal to parental schooling. From equation 4, it follows that Equation (5) is widely known in the literature; it shows that the square of intergenerational correlation provides an estimate of the share of total variance in the educational attainment of children that can be explained by parental educational attainment only (Solon, 1999). As discussed above,β is affected by the relative variance of education in the two generations. Therefore, any change in the relative variance may lead to different ρ and β trend in a same society. For instance, Hertz et al. (2007) document that β fell over time (implying more mobility); yet, the correlation between children's and parents' educational attainment remained constant for half a century (implying no change in mobility). For this reason, it is a common practice to report both measures of educational persistence (see, for example, Ranasinghe, 2015;Checchi et al., 2013;Azam and Bhatt, 2015;Hertz et al., 2007).
In this study, we follow the same tradition and report both measures, intergenerational elasticity (β) and the correlation coefficient (ρ), across five-year birth cohorts in all the countries. Both measures can be easily extended to analyze different aspects of intergenerational mobility, such as mobility by geography or difference in sex, caste, and other aspects socioeconomic status (Ranasinghe, 2015;Hnatkovska et al., 2013;Azam and Bhatt, 2015;Bourguignon et al., 2007;Binder and Woodruff, 2002). With the objective of assessing a sex difference in intergenerational education persistence, we estimate the parameters in equations (1) and (2) for daughters and sons separately. To strengthen our analysis by looking at changes in children's educational attainment conditional on parental educational attainment and shedding light on the role of maternal and paternal education on children's human capital accumulation, we study the intergenerational transmission of children's highest level of educational attainment across cohorts. We define three categories of education for both the generation of the children and the generation of the parents: no schooling, primary, and secondary schooling and above. 5 Analyzing transitions in educational level between the two generations serves the same propose as the decomposition of Checchi et al. (2013). Theories on education inequality also support this approach by highlighting the importance of socioeconomic background in educational level transitions (Mare, 1980;Raftery and Hout, 1993). According to these theories, it is important to identify levels of education among children that are highly influenced by paternal or maternal education. Studying educational levels that are highly influenced by parental educational attainment is also important for policy within Sub-Saharan Africa. Following a series of education system reforms in many counties in the region, there has been a significant increase in primary education, but enrollment in higher education (secondary education and above) has remained low (Tomasevski, 2006). Hence, this analysis will give an indication where policy should focus to promote equity in the long run. Accordingly, we model the effect of the highest levels of educational attainment among parents on children's highest level of schooling across five-year birth cohorts. This entails estimating an ordered probit model. Let s * be an ordered response that takes on values of 0, 1, 2 denoting children's highest level of education (0 = no schooling, 1 = primary, 2 = secondary and above). The latent model underlying the ordered probit model for s * is as follows: where s * is unobservable; sp is the control for the highest level of education of parents; β is the corresponding unknown coefficient; and ε is the error term, which is assumed to be normally distributed across observations with mean and variance normalized at 0 and 1. Although s * is unobservable, we observe the highest level of children's educational attainment which is the category of the response, that is, as follows: where α 1 < α 2 are the unknown cutoff points. The estimation of the parameters α and β is performed using maximum likelihood (Wooldridge, 2010).

Data
We use data from Comoros -Enquête intégrale auprès des ménages ( , 1944for Ghana, 1933for Guinea, 1936for Madagascar, 1942for Malawi and Nigeria, 1931for Rwanda, 1941for Tanzania, and 1937 After data cleaning, the total sample size ranges from 32,730 in Ghana to 6,778 in Tanzania; a total of more than 145,000 adult children (ages between 20-69) are represented in the study. The data in all the countries are organized into five-year birth cohorts based on the children's years of birth. In table 1, we also present the minimum sample size in each country, which corresponds to the smallest sample size of five-year birth cohort in each country. In all the countries, years of schooling is coded as the number of years associated with the highest grade completed, and repeated grades are not counted. Parental educational attainment is the average of the years of schooling of the mothers and fathers. All the surveys contain information on the educational attainment of parents in two separate variables that differentiate the education of parents who are coresiding or not residing in the household. We used the personal identification numbers of fathers and mothers to create a pair of parents and children if the parents and children are still living together in a household. Using this information to create pairs of parents and children imposes a co-residence condition that reduces the sample size significantly. In addition to this information, all the surveys have another question that collects information on retrospective parental educational attainment among household members who are not co-residing in the household regardless of whether the parent is alive.
Combining these two variables, we have been able to identify parental schooling for more than 95 percent of adult respondents in each country. Because of the lack of long panel or administrative data, most studies in the literature have used cross-sectional data and co-residence to identify child-parent pairs, mostly father-son pairs (see, for example, Emran and Shilpi, 2015;Hnatkovska et al., 2013;Jalan and Murgai, 2007). Using co-residence identification has three important implications for the analysis. First, because the distribution of education across both generations is different in the subsample of adults who live together with their parents and versus the total population, sample selection problems arise that bias the intergenerational elasticity downward. For instance, Francesconi and Nicoletti (2006) and Azam and Bhatt (2015) document a bias because of the co-residence condition that ranges from 12% to 39% and 17% in constructing father-son pairs in the United Kingdom and India, respectively. Second, the co-residency criteria over identify younger adult children who are still living with their parents; this might not lead to a representative adult population sample (Jalan and Murgai, 2007;Hnatkovska et al., 2013). Third, children-parent pairs that are constructed using co-residence identification does not allow cohort-wise long-term trend analysis of intergenerational persistence.  The last two columns of table 1 report the average years of schooling of parents and children in the first and last five-year birth cohorts. In all countries except Madagascar, the average years of schooling have increased rapidly over the past 50 years among both children and parents. Excluding Madagascar, we document four to six and two to five years of schooling gain between the first and the last five-year birth cohorts of children and parents, respectively. We observe the least year of schooling gain, about one year, among both children and parents in Madagascar. Figure 1 provides a visual illustration of the educational attainment of children (daughters and sons) and their parents across five-year birth cohorts. In all the countries, children, on average, have higher educational attainment than their parents. With the exception of the youngest cohorts, there has been an upward trend in the average years of schooling among both sons and daughters across cohorts. In general, average years of schooling is higher among sons than daughters, and the gender difference is significant. The gender difference remains similar across the two generations, women (mothers and daughters) show significantly fewer years of schooling than their men counterparts (fathers and sons). Overall, the proportion of children with no schooling has declined over the 50 years. We note a boom in primary education in all the countries, particularly among children born after the 1960s. This coincides with the policy changes in educational systems and a huge investment in human capital accumulation in the region after independence (Thakur, 1991). In all our sample countries, the proportion of children who completed tertiary education was small. The proportion of daughters who complete tertiary education is less than 10% among all cohorts and countries. We note the same trend among sons except in Ghana and Guinea, where we observe a slight improvement in the two youngest cohorts. Similarly, a large proportion of parents show a lower level of education, no education, and primary education in all the countries across cohorts (see table A1-A9 in Appendix A). Parent-child differentials in the distribution of the highest educational attainment suggest improvement in education mobility or a weak link between the educational persistence of parents and children over time, particularly among the youngest birth cohorts.

Results
We present the estimates of the two intergenerational educational persistence measures, intergenerational elasticity (β) and the correlation coefficient (ρ), in six stages. First, we present our baseline estimates at the country level using a pooled sample of all children in each country. Second, because our sample is comparable with the datasets used by Hertz et al. (2007) to rank 42 countries in five regions, we pool our data at the regional level and rank Sub-Saharan Africa in terms of intergenerational educational persistence among other regions. Third, we discuss the trend in intergenerational education mobility across five-year birth cohorts using both measures. Ranking each country among other nations on which comparable estimates are available follows. Fourth, we explore the potential differences in intergenerational education mobility across gender. Fifth, we explore the potential difference of paternal and maternal educational attainment in influencing child's educational attainment across five-year birth cohorts in each country. The final section presents the estimates of the order probit model of children's highest level of education. Table 2 presents the estimates of the intergenerational elasticity and correlation coefficients for the pooled sample in each country. The results reveal two main findings of interest. First, for all specifications considered, parental education has a statistically significant effect on children's educational attainment in all the countries. The estimates imply that, despite the increase in years of schooling in almost all the countries over the last 50 years, parental education plays a crucial role in children's education attainment.

Intergenerational education mobility at the country level
There exists an intergenerational link in educational outcomes: for instance, a one year difference in parental schooling is associated with a 0.74-year difference in children's education in Madagascar. In terms of the estimated intergenerational elasticity, Tanzania and the Comoros show the highest and the lowest intergenerational education mobility, respectively. On average, an additional year of parental schooling is associated with a 0.47 and a 0.91-year difference in children's years of schooling in Tanzania and the Comoros, respectively. Second, as discussed above, gender is an important determinant of educational attainment in many developing and developed countries.
In low-income countries, girls tend to receive less education than boys (Behrman and Knowles, 1999;Alderman, Harold and King, Elizabeth M, 1998). In Sub-Saharan Africa, boys are still 1.6 times more likely to complete secondary education than their girl counterparts (Klugman et al., 2014). This is also in line with our observation in the sample countries in figures 1 and 2, women (mothers and daughters) show significantly fewer years of schooling than their men counterparts (fathers and sons) in all the countries. Accordingly, in column (2), we control for gender. We find that estimated intergenerational elasticity declined slightly in Ghana, Guinea, Madagascar, Malawi, Nigeria, Rwanda, and Uganda, while it increased slightly in the Comoros and Tanzania, suggesting lower educational mobility or higher educational persistence among daughters than sons in most countries. The next set of results in column 3, table 2 includes other control variables that are used in the literature: age and number of children in a household. Moreover, with the objective of capturing the cohort effect, we also include the square of age. The results in column 3, table 2 show that the addition of the controls does not affect the estimated intergenerational elasticity in any significant way in the countries, though it leads to slight increase in the explanatory power of the regression. Despite the inclusion of such powerful controls, the qualitative results remain unchanged; parental education plays a vital role in children's educational attainment in all the countries. Parents education is average of mother's and father's years of schooling. † Regression include gender of children.
‡ In addition to gender this regression includes age, age square and the number of children in a family.

Intergenerational education mobility at the regional level
To rank Sub-Saharan Africa among world regions, we have pooled the sample at the regional level. We find a regional correlation coefficient of 0.51, indicating that parental years of schooling account for about 51% of the inequality in children's years of schooling. Our estimate is above the global average of 0.42 (for 42 countries) documented by Hertz et al. (2007), and is comparable with their estimates of 0.39, 0.44, 0.46 in Asia, Western Europe and the United States, and Eastern Europe, respectively. Sub-Saharan Africa has lower estimated intergenerational educational persistence (ρ) than Latin America. Overall, mobility in Sub-Saharan Africa is lower than Europe and the United States and Eastern Europe and higher than Latin America. Our estimate of intergenerational correlation (ρ) is also higher than the African average of 0.36 estimated by Hertz et al. (2007). Therefore, using correlation coefficient of parental background explains a significantly higher share of the variation in the educational attainment of children in Sub-Saharan Africa than before. 7 Similarly, we estimate an intergenerational elasticity of 0.66 for Sub-Saharan Africa, indicating that additional years of parental schooling, on average, increases children's years schooling by 0.66. This estimate is higher than the estimates of Hertz et al. (2007) 0.52 in Western Europe and United States and 0.38 in Eastern bloc Europe, and is lower than the intergenerational elasticity of 0.83 in Latin America and 0.69 in Asia (see table 3). The intergenerational elasticity estimate of the current study is lower than the estimates of Hertz et al. (2007) (0.8) for four African countries (Egypt, Ethiopia, Ghana, and South Africa). Our estimates of the two measures paint a picture that is consistent with previous estimates on developing countries. While intergenerational elasticity demonstrates that an extra year of parental schooling adds fewer years of schooling to children's education now than before, the regional correlation coefficient estimate is higher than previous estimates of Hertz et al. (2007), thereby, telling a bleaker story of mobility in the region. This apparent contradiction can be explained based on Eq.2: two countries can have the same intergenerational elasticity estimates, but the correlation coefficient can be different if the educational inequality in the generation of the parents and children varies over time, for instance because of, education policy changes affecting children's generation (see section 5.3).
As discussed in section 3, Eq.5, the square of intergenerational correlation provides an estimate of the share of the total variance in schooling that can be explained by parental years of schooling alone. We estimate that parents education alone can explain 31% of variations in the years of schooling of daughters and 21% of sons in the region. This estimate is higher than the available estimates for developed countries that indicate that parental education explains only 9%-21% of the total variation in children's years of schooling and lower than the estimates for India, where parental education explains 27%-29% and 31%-39% of total the variations in the year of schooling of sons and daughters, respectively (Bjorklund and Salvanes, 2010;Emran and Shilpi, 2011).  Hertz et al. (2007). Number of countries refers to the sample counties in each region.

Cohort analysis
In this section, we investigate the trends in intergenerational mobility in educational attainment of each of the five-year birth cohorts based on the number of years of schooling for both generations. Table 4 reports the results. In line with our previous observations in table 2, parental education has a statistically significant effect on the child's education across most birth cohorts in all the countries. The intergenerational persistence of education has generally decreased over the last five decades in all the countries, but the trend has not been consistent. In all the countries, there has been a significant improvement in education mobility from the 1960s onward. Nigeria, Guinea, Ghana and Uganda have recorded the highest gains in intergenerational mobility between those born from 1940s to the 1990s. The decline in the relationship between the education of parents and children is quite impressive in Nigeria, where the intergenerational elasticity between the youngest and the oldest cohort declined by 65% between 1942 and 1991 (table 4). For the Comoros, Guinea, and Rwanda, we document a small intergenerational educational persistence rate for the oldest cohorts. However, a lower persistence rate in educational attainment in these countries does not necessarily reflect high social mobility among older cohorts; rather parental education does not vary across households because of the low years of schooling, and parental education can only explain small proportion of the variation in child schooling. For example, children born between 1931 and 1935 in Rwanda have an average 1.4 years of schooling, while their parents have only 0.2 years of schooling. This is consistent with our observations that the explanatory power of the relationship between parental and children's years of schooling has been limited among older cohorts and increased among the youngest cohorts in these countries (table 4, column 4).
Our sample includes individuals who are continuing their education. This represents either delay in schooling or the pursuit to higher education. If it is caused by delayed completion among children in households with well-educated parents, the intergenerational elasticity will be biased downward. On the other hand, if children in household with less well educated parents are the ones taking more years to complete their education, the covariance between parental education and children's education increases, and the intergenerational elasticity will be upwardly biased. In light of this, we repeat our analysis by excluding the youngest cohort that is made up of children ages between 20 and 24 from each country where the current enrollment is higher. 8 Our results are relatively unaffected, and the bias is fairly small. In the youngest cohort, the true value of years of schooling is less than 1 year of schooling on average than what we observe in the youngest cohort if we use 20 years as the lower age cutoff. 9 Parents education is average of mother's and father's years of schooling.
Significance levels: * : 10% * * : 5% * * * : 1% For the standardized measure of intergenerational mobility, the correlation coefficient, a declining trend is not visible. This result is similar to the findings of previous studies in developing countries (see, for example, Hertz et al., 2007;Azam and Bhatt, 2015;Daude and Robano, 2015). A plausible explanation for the discrepancy between the two measures is a change in the dispersion of the years of schooling across the two generations (parents and their children). To examine this possibility, we present, in Figure 3, the trend in the standard deviations of years of schooling of both generations and the two measures of intergenerational educational persistence. The result clearly shows that, while the dispersion of education in the children's generation has decreased from the 1960s onward, the inequality in parental education has increased. This finding is expected: if nearly all parents were initially uneducated and then a small proportion, especially young parents, gain access to education, the variance in years of schooling will increase. For all the countries but Tanzania and Ghana among recent cohorts, the variance in children's years of schooling is always greater than that of the parents. This leads to a ratio of the standard deviation of parental schooling to that of their children of less than 1, because of which the correlation coefficient (ρ) is less than the intergenerational elasticity (β) among almost all the cohorts in every country. The combined effects, that is, the lower intergenerational correlation and the rise in the dispersion of parental education, explain the slight increase in intergenerational correlation. These patterns are similar to those reported by Hertz et al. (2007) and Azam and Bhatt (2015). The general declining trend of intergenerational elasticity (β) across cohorts partly reflects the improvement in the education systems and policies in many countries in the region. The education systems in Sub-Saharan Africa expanded substantially after independence in the 1960s, which almost doubled primary-school enrollments in many of the countries (Thakur, 1991). The expansion of primary education was facilitated through the expansion of public education. Government expenditure on education grew substantially during the period (UNESCO, 1970). During the decade, we observe a decline in intergenerational educational persistence in almost all countries in our sample (see, for instance the Comoros, Ghana, Guinea, and Nigeria estimates in table 4). In the 1980s, the recurrent balance of payment failures and economic regression limited the public expenditure on education. We observe a rebound in the intergenerational persistence rate in our sample countries. A revival of public education funding occurred again in the 1990s and the education systems in many countries in the region underwent dramatic policy changes. One of the most dramatic educational policy changes, for instance, was the abolition of primary school fees in Ghana, Malawi, and Uganda in the 1990s and in Benin, Burundi, Lesotho, Liberia, Mozambique, Rwanda, Sierra Leone, Tanzania, and Zambia in the 2000s (UNESCO, 2011). In line with this, we observe a decline in intergenerational elasticity among the youngest cohorts in all countries except Malawi and Tanzania. The positive relationship between public education expenditure and individual educational attainment has been extensively documented in the literature on both developing and developed countries (Black and Devereux, 2011). Hertz et al. (2007) also provides a survey of the existing literature that reports changes in education policy and intergenerational education mobility. There is also empirical evidence that higher public expenditure on primary education is boosting education mobility across generations in many countries. Thus, it is plausible to hypothesize that our result of rising educational mobility across cohorts using intergenerational education elasticity to some extent reflects the inclusiveness of the policy changes in the region to create equal educational opportunities for children from different parental education backgrounds over time.
Economic theory suggests three possible drivers of intergenerational mobility across countries, namely, income inequality, the returns to education and public education expenditure. Without inferring any causality, this section shows the correlation between the two measures of intergenerational mobility and the potential drivers in each country. Figure 4 suggests a positive correlation educational persistence and income inequality measured by the Gini coefficient. Countries that show lower educational mobility over the 50 years tend to experience a higher level of income inequality (figure 4). Contrary to the prediction of theory, we observe a negative correlation between returns to education and intergenerational educational persistence. In countries where we observe greater mobility, the returns to education tend to be smaller. One plausible explanation might be the credit constraints affecting poor households in a country where the returns to education are higher. Children with higher parental years of schooling probably have higher incomes and, will have the capacity to invest more on children's education relative to poor households. The results also suggest a negative relationship between educational persistence and public expenditure on education as a share of total government expenditure, implying that progressive public investment on education helps to foster equal opportunity in education among all children, including children with different parental educational backgrounds (figure 4). .7 .8 .9 Coefficient ( To compare levels of intergenerational educational persistence and rank the countries in our sample in terms of im(mobility) in educational attainment, we follow the approach of Hertz et al. (2007) and derive the simple average ofβ andρ across five-year birth cohorts in each country. 10 Using the intergenerational correlation (ρ) our result shows that most of the countries, except Rwanda and the Comoros, show greater intergenerational educational mobility than Latin American countries, but lower mobility than Western Europe, the United States and Eastern European countries. 11 The estimates for Rwanda and the Comoros show that these countries are more mobile than most developed countries. However, parents in both countries show fewer average years of schooling even among the youngest birth cohorts, and parental schooling can explain only a small proportion of the variation in children's schooling (see figure 1, figure 2, and table 4). India data from Azam and Bhatt (2015). Australia data from Ranasinghe (2015). Continued on next page. . .

What is important, education among mothers or education among fathers?
As discussed in section 2, much is still not known about the relative importance of the education of mothers and the education of fathers on children's educational outcomes, but the existing evidence reveals some suggestive patterns. With the objective of looking at the differential effect of the education of mothers and the education of fathers on the intergenerational mobility of daughters and sons, we carry out the same analysis on each sample. Figure 5 and Appendix B present the results. There are two notable findings. First, in all countries except the Comoros, both measures coincide in pointing out the stronger effects of maternal years of schooling relative to paternal education. This finding echoes previous findings in other countries such as South Africa, Australia, and Sweden (Kwenda et al., 2015;Ranasinghe, 2015;Thomas, 1996;Branson et al., 2012). Figure 5 shows that, in all countries except Madagascar, maternal schooling has a stronger effect on daughters than on sons. The results suggest that extra maternal years of schooling have an important role in determining the educational outcomes among daughters than sons (see tables B10-B18 in Appendix B). This finding is similar to the findings of previous studies results in other parts of the world. Several studies in developed countries underlined that mothers education strongly affects the educational attainment of female children relative to male children (see, for example, Crook, 1995 on Australia; Björklund et al., 2006 on Sweden). The differences in the effects of maternal and paternal years of schooling on children's educational outcomes might emerge from the different roles played by mothers and fathers in family life, in the labor market, and role model effect in each country. It is likely that mothers become the natural role model for a daughter, and fathers for sons. Social norms regarding gender roles might also play a big role on who, sons or daughters, obtain more years of schooling. Second, in line with our other findings, we note a decline in father-child and mother-child intergenerational elasticity (β) across birth cohorts after the 1960s. The mother-child intergenerational elasticity declined more than father-child's elasticity in Ghana, Guinea, Madagascar, Nigeria, Rwanda, and Uganda over the 50 years. In the Comoros, Malawi, and Tanzania, we document more gains in mobility from fathers to children.

Intergenerational mobility in educational attainment
As discussed in section 3, estimates of intergenerational elasticity and the correlation coefficient do not allow us to identify the level of children's educational attainment that is more affected by parental education. To investigate this, we estimate an order probit model for children's highest level of education in each country for each five-year birth cohort. The tables in Appendix C present the probability of a child achieving primary educational attainment or above, conditional on her mother or father's education across each five-year birth cohort in each country. In all countries, the omitted category of mother's and father's education is parents with no schooling. Despite family background, we document a convergence toward zero in the probability of children attaining no education in all countries. The results show that downward mobility, that is, attaining no schooling if parents have at least primary education, is negative across cohorts, suggesting an upward mobility from no schooling to an upper level of educational attainment. Concerning primary education, we find a narrowing, but not closing gap in the probability of childrens attaining primary education across all family educational backgrounds. For instance, in the Comoros, the probability of attaining a primary education when a mother has primary education as well declines from 13 percent in the oldest cohort to 2.5 percent in the youngest cohort. We observe a similar trend in primary education in all the countries except Ghana, Nigeria, and Uganda. This result suggests a narrowing, but not closing gap in primary education across children with a different parental educational background. However, in Ghana, Nigeria, and Uganda, the children of parents who have primary education experienced upward mobility and had a greater chance of obtaining a secondary education or above. Furthermore, children in more well educated households (parents who have secondary education or above) have a greater chance of obtaining a higher education than the children of parents who have no education or who have completed only primary education. Thus, children with poor parental education still have a lower prospect of attaining higher education. The overall results suggest that all the counties experienced upward intergenerational educational mobility over the last 50 years. However, the observed mobility in almost all the countries is concentrated in the lower tail of the education distribution, primary education. It is plausible that this is the result of the expansion in primary education in the region after independence in the 1960s. In line with our results in previous sections, we document evidence that maternal education is more important in influencing children's education relative to schooling of fathers in all the countries but the Comoros.

Conclusion
Drawing on nationally representative survey data, we study the intergenerational im(mobility) of educational attainment in the Comoros, Ghana, Guinea, Madagascar, Malawi, Nigeria, Rwanda, Tanzania, and Uganda over 50 years, with a particular focus on gender differences. The overall results indicate that there has been a significant improvement in intergenerational educational mobility during the last five decades, particularly after the 1960s. We document a country difference: Nigeria, Guinea, Ghana, and Uganda experienced the highest intergenerational mobility, and the Comoros and Madagascar the lowest. Nevertheless, the educational attainment of parents remains a strong determinant of children's schooling outcomes. We also find considerable gender differences in the persistence of education across generations, which are masked in the country estimations; the educational attainment of daughters is more closely correlated with parental years of schooling relative to educational attainment of sons. On paternal or maternal effects, the education of mothers is significantly more important than the education of fathers in shaping the educational attainment of both daughters and sons, though the effect is much stronger among daughters. Furthermore, we document more mobility in the lower tail of education distribution. In the countries, children from all family backgrounds exhibit a greater chance of attaining primary schooling, while children from more well educated family backgrounds have a greater chance of obtaining more schooling beyond primary education.
From a policy perspective, our results suggest a need for targeted redistribution policies that improve intergenerational mobility in the region. Moreover, putting in place an inclusive environment for women (mothers) who are less well off in human capital accumulation might play a decisive role in promoting social mobility in the long run. While primary education enrollment in our sample has generally increased over the five decades, access to secondary schooling is far from universal. Putting in place policies that promote access to secondary education is therefore a priority among the educational systems in the countries under investigation. Evidence from developed countries suggests that making secondary education mandatory better promotes education outcomes among the next generation. However, the policies in each country should be context specific.
There are two caveats. Because it was difficult to find valid instrumental variables to address the genetic correlations (ability and preference) between parents and their offspring, the study does not distinguish the effects of nature and nurture. We limit our analysis to investigating the correlation between the educational attainment of parents and children without implying any causality. Second, for all countries in our sample, we rely on single cross-sectional surveys and study intergenerational mobility among five-year age cohorts. Hence, we cannot assess the extent of measurement error, if any, in the education variable of both parents and their children. Future research might therefore involve examining the importance of these elements to document the importance of the recent education policy changes in the region in promoting social mobility.            [1955][1956][1957][1958][1959] 193 0.021 1950-1954 [1945][1946][1947][1948][1949]     The reference group is parents with is no education.