diff --git "a/data/part_2/0155082622.json" "b/data/part_2/0155082622.json"
new file mode 100644--- /dev/null
+++ "b/data/part_2/0155082622.json"
@@ -0,0 +1 @@
+{"metadata":{"gardian_id":"209e53b6d22a2892585bb89e453b5daf","source":"gardian_index","url":"https://cgspace.cgiar.org/rest/bitstreams/9eab2e1a-9917-401a-96a4-9b82d276473d/retrieve","description":"The end of the apartheid system in South Africa in 1994 brought with it the end of legally sanctioned racial segregation in schools. In practice, however, racial disparities in educational attainment continue, with white and Asian students outperforming the country's African majority in school. Such a gap in achievement has a significant impact on the ability of Africans to get jobs and escape poverty. Human Capital Formation: History, Expectations, and Challenges in South Africa investigates the causes of South Africa's persistent schooling imbalances, examining education laws and policies, as well as other influences on human capital investment. The study finds that inaccessibility of quality education, resulting from a lack of financial resources at both the local and household levels, is currently a significant constraint on educational attainment among the poor. This limitation will likely relax in the future as the government continues to subsidize schools, but the study also concludes that educational disparities cannot be overcome by direct attention to schools alone. For example, children require adequate nutrition at the pre-school stage in order to perform well in school. Furthermore, parental death or illness resulting from the HIV/AIDS epidemic can disrupt the education of adolescents, who might need to leave school to care for their parents or support their family. Steps to compensate for these problems, as well as improved access to schools, are necessary. These findings clarify the problems underlying inequalities in educational attainment, offering guidance to policymakers, development specialists, and others concerned with South Africa's welfare.","id":"-864998923"},"keywords":[],"sieverID":"b14e520a-fb63-426b-a8f9-27e0f6af1f0c","pagecount":"150","content":"The International Food Policy Research Institute (IFPRI ® ) was established in 1975 to identify and analyze alternative national and international strategies and policies for meeting food needs of the developing world on a sustainable basis, with particular emphasis on low-income countries and on the poorer groups in those countries. While the research effort is geared to the precise objective of contributing to the reduction of hunger and malnutrition, the factors involved are many and wide-ranging, requiring analysis of underlying processes and extending beyond a narrowly defined food sector. The Institute's research program reflects worldwide collaboration with governments and private and public institutions interested in increasing food production and improving the equity of its distribution. Research results are disseminated to policymakers, opinion formers, administrators, policy analysts, researchers, and others concerned with national and international food and agricultural policy.Foreword viii S outh Africa's apartheid system ended in 1994. This monumental event stopped legally sanctioned racial segregation but left the country with one of the highest rates of income inequality of any nation in the world. This research monograph investigates how to develop human capital-through the education system and other avenues-to bring income equality to South Africa by preparing low-income groups to better contribute to economic growth.This study by Futoshi Yamauchi investigates influences on human capital investment, the most prominent of which is the state of South African education. By examining education policies and household behavior, Yamauchi arrives at two major findings. First, access to quality education is still unequal across different segments of the society. Black Africans continue to have lower-quality schooling than other racial groups. For this reason, providing subsidies to black African households and predominantly black African schools to relieve financial constraints is an urgent priority.Second, although education is at the center of human capital formation, such formation must go beyond the classroom. To guarantee the outcomes expected from the government's efforts to develop public-school education, building human capital must start in early childhood-indeed, even before birth. South Africa needs policy interventions aimed at enhancing healthy child growth, such as the current child support grants. Another obstacle to public education improvements and human capital development is the HIV/AIDS epidemic. For many adolescents, educational progress stalls when HIV-infected parents cannot work, forcing students to leave school and enter the labor market to support their families. The government should provide these households with support, thereby protecting children from negative impacts that may affect human capital development.The formation of human capital will take time, but it is a more politically feasible method of promoting income equality than direct asset redistribution from the white African minority to the black African majority. Human capital formation can also be a longer-lasting and more sustainable way to promote income equality and overall economic growth. Investing in the human capital of a disadvantaged group builds new assets for that group, expanding the possibilities for South Africans struggling to emerge from the shadow of apartheid and for the country as a whole.Introduction H uman capital is a general notion of the knowledge and skills embodied in human beings, which plays an important role in determining their labor productivity (Schultz 1961;Becker 1962) and their ability to absorb new knowledge and master new technologies (Schultz 1975).Human capital formation takes different forms and passes through various stages in parallel with the human life cycle. The core of human capital formation is acquisition of new knowledge and skills. There is no doubt in modern societies that education plays an important role in this central activity. However, since knowledge and skills are embodied in human beings, it is also hard to separate them from human health, which also determines labor productivity (for example, Strauss and Thomas 1995). Moreover, the interaction of human beings also affects knowledge spillovers in a society (Romer 1986;Lucas 1988). Therefore, the way in which the society is structured is crucial to the formation of human capital.Formation starts before childbirth, when parents' decisions and behavior determine birth outcomes. Child growth affects outcomes during schooling (for example, Alderman et al. 2000), which subsequently influence labor market outcomes. It takes a long time to form human capital.The fact that human capital is embodied in human beings implies that it can directly improve the earning capability of the poor. Human capital can contribute to greater equity in a society by directly changing the earning capability of those who have little access to physical and financial resources. Publicly funded education is one of the most important means of building human capital among the poor, who tend to be credit constrained for private investments in education. In this sense, human capital investment through public education is essentially open to the majority in many societies.The accumulation of human capital is also a basis for economic growth. In many economies that have experienced successful economic growth, the accumulation of human capital either preceded the growth path, was an important component that explained growth, or both (Hayami and Godo 2005). Complementarities between human and physical capital in the economy lead to the acceleration of further investments in human and physical capital in the long run (Lucas 1988). The experiences in Asian economies show that human capital is one of many critical factors that can enhance economic growth with equity (World Bank 1993).This monograph describes challenges and possibilities in the formation of human capital in South Africa. The country experienced apartheid until the mid-1990s. Given the difficulty in directly redistributing assets and wealth from the historically advantaged minority to the majority, the broad formation of human capital in the population-especially access to quality education among the majority Africans-appears to be able to provide more opportunities to the majority of Africans and to have the potential to improve equity in the country. 1This monograph highlights the issues surrounding human capital formation in South Africa. Apartheid resulted in segregation and discrimination against the majority African population in almost all aspects of social life, such as education, employment, and residence. As part of apartheid, Africans were deprived of both physical and human assets, being forced to live on infertile land or in urban townships, and of the rights of citizenship and political participation, particularly the right to vote. 2  Since 1994 the South African government has promised to implement policies to help the majority of Africans to gain lost opportunities. However, since the redistribution of assets from the minority whites to the majority Africans has been politically challenging, the country has also taken more gradual measures to improve equity, without adversely affecting its economic performance.In the area of public education, the government of South Africa formally terminated school segregation laws, but realities for African learners did not improve greatly (see Chapters 2 and 3). The government introduced the South African School Act (Republic of South Africa 1996a, 1996b) and the Norms and Standards for School Funding (Republic of South Africa 1998) to provide guidance in implementing a nonsegregated education system. Under apartheid, segregated schools were governed by racially separate government agencies according to their classification both by population groups (white, colored, needs a rather long gestation period. For example, school education up to the completion of high school requires 12 years. This factor also implies that the expected improvement of equity and income distribution takes a rather long time.Human capital investment needs time input, which implies opportunity cost at the individual level. Individuals need to indirectly pay the cost, forgoing earning opportunities that are available during the period they invest in human capital.This section provides evidence on the returns to schooling in South African labor markets. Since I do not cover labor market issues in the main analysis, it is important to briefly discuss the functioning of labor markets and point out unique features of the markets in South Africa.High unemployment is one key feature of the South African labor markets. Kingdon and Knight (2006) estimate that, in the period between 2002 and 2005, the unemployment rate was holding at 40 percent.Several studies provide evidence on returns to schooling in the South African labor markets using micro data (see Bhorat et al. 2001). Though racial gaps in returns to schooling have been narrowed, lower returns among Africans are attributed to grade repetition and the lower quality of the schools available to them (for example, Anderson, Case, and Lam 2001).One interesting aspect of the data on schooling returns is their convexity. Education up to the high school level does not contribute to employment, but tertiary-level education significantly increases the probability of employment (Bhorat et al. 2001), thus creating convexity in the earning profile. This finding is similar to reports from the Philippines (Yamauchi 2005a).The effect of heterogeneity in school quality on labor market outcomes is important in South Africa. For example, Case and Yogo (1999) attempted to quantify the effect of school quality on labor market earnings in South Africa, merging data from population censuses and school censuses. 6 On this point, it is worth noting that the matriculation pass rate at high school graduation is generally low in South Africa, having decreased from 70 percent in 2004 to 65 percent in 2007. This means that even if children officially finish high school, they are not qualified based on measures of educational accomplishment. The problem is particularly severe in rural areas where schools are located in predominantly African communities. I estimate returns to schooling using both employment and log monthly wage equations. The employment equations show determinants of employment, while the wage equations estimate the effects of schooling on monthly earnings using the population of employed persons. Combining results from both equations, we can determine how schooling affects the likelihood of being employed and the salaries paid to the employed.As is well known and has been discussed in various studies (for example, Card 2001), ordinary least squares (OLS) estimates of returns to schooling could be biased because of correlations between years of schooling and unobserved fixed error components, omitted determinants, or both. Since individual ability (augmenting the wage) and schooling attainment are positively correlated, our estimates are likely to be biased upward. 7  Moreover, school quality and family background (neither of which is directly observable) affect schooling attainment and income levels. In the current context, they are both correlated with population groups: whites still have persistently better access to high-quality schools and enjoy more family resources than the other groups, particularly Africans. Therefore it is important to control for population groups to minimize potential bias in estimating returns to schooling.Columns 1-3 in Table 1.1 summarize estimation results for the employment equation. 8 Schooling significantly increases the probability of having work (column 1), but it only increases the probability if high school or a higher level is attained (column 2). Educational attainment up to grade 12 does not significantly contribute to the likelihood of being employed in South Africa.Column 3 includes the interaction of population group indicators and years of schooling. Whites have an advantage in getting employment, and Indian/ Asians have a disadvantage regardless of schooling, compared to Africans. However, returns to schooling are significantly higher among coloreds and Indian/Asians than Africans.Chapter 5 presents empirical evidence on transitions among adolescents from school to the labor market, highlighting the impacts of prime-age adult mortality. Many adolescents who quit school do not find jobs in the labor market (increasing the labor supply in the labor force).Similar findings are confirmed by the wage equations (columns 4-6). One interesting finding is that, compared to Africans, whites earn higher wages regardless of schooling, and Indian/Asians demonstrate higher returns to schooling. It is also confirmed that schooling up to completion of high school does not contribute to monthly wage.These findings demonstrated heterogeneity in returns to schooling in both the employment and wage equations, but they also pose two as yet unresolved INTRODUCTION  7   7 In the empirical setting and for the data used in the estimation of returns to schooling, it is difficult to find valid instruments or exogenous variations in years of schooling (schooling attainment). Schooling attainment increased in the recent cohorts, especially among Africans after 1994, but in the post-apartheid regime the change occurred simultaneously with changes in the labor markets, such as employment opportunities formerly restricted to whites being opened to Africans. 8 The employment indicator takes the value of one if an individual works in some capacity and zero otherwise. I dropped from the sample those who did not look for work within the past seven days because they are currently learners, housewives (homemakers), retired (prefer not to seek work), ill, disabled, too young, or too old. The indicator takes a value of one if the individual is guaranteed work in the near future. questions. First, what creates differences in the returns to schooling? One important missing factor is the quality of education, which may substantially differ across schools, possibly correlated with historical backgrounds and local school and community conditions. This issue is examined in depth in Chapters 2 and 3. Second, why do returns to schooling show convexity? Investments in schooling below the high school level do not help individuals find jobs, nor does increasing wages in the labor market. Yamauchi (2005a) shows similar findings from the Philippines. This issue is not directly addressed in this monograph.The approach taken in this monograph is twofold. First, we try to understand institutional factors that constrain the formation of human capital. In this respect, the school system is the most important factor within the context of South Africa, where the regime transformed from segregation under apartheid to democracy in the mid-1990s. Unless we understand which institutional and historical factors determine the constraints imposed on microeconomic behavior in human capital investment, it is difficult to correctly interpret findings from household-level analysis. Notes: Numbers in parentheses are absolute t-values, using robust standard errors with sample stratum clusters (province including both urban and rural areas). The employment indicator equals one if an individual worked in some capacity and zero otherwise. The employment indicator equals one if an individual worked in some capacity and zero otherwise. Those who did not look for work within the seven days preceding the survey because they were students or housewives (homemakers) or were retired (preferred not to seek work), ill, disabled, too young, or too old were dropped from the sample. The indicator has a value of one if the individual was guaranteed work in the near future. Data for college graduates and postgraduates were converted into years 16 and 18, respectively.Second, we analyze household behavior related to human capital investment. In this area, we focus on dynamic human capital production, from earlychildhood nutrition intake to schooling investments and attainment, and on adolescents' transition to the labor market. Even at this stage of the analysis, it is important to integrate institutional factors and observed household behavior to achieve a comprehensive understanding.In South Africa, access to quality education is still highly unequal across population groups and locations, although the situation has been changing. In the transition from apartheid to a democratic education system, two milestones stand out: the 1996 South African School Act and the 1998 Norms and Standards for School Funding. These two sets of rules provided basic guidelines to departments of education at the national and provincial levels. More recently, South Africa introduced a revised set of Norms and Standards to promote more progressive fiscal interventions in the school system.As described in Part 1, however, there still remain discrepancies in access to quality education across population groups and communities (where population groups remain locally clustered). To understand institutional constraints, the two key dimensions are introduced.The first dimension is population groups. Historically there have been four politically defined population groups in the country: white, colored, Indian/ Asian, and African. The majority of the population is African. Under apartheid, Africans had semiautonomous homelands where they governed and received almost no support from the white central government. Public schools were also governed by different education ministries that were separated by population group and homeland. It has been important to investigate the convergence in access to quality education across population groups.The second dimension is location. Since population groups were segregated in residential locations under apartheid, their schools were also segregated by location. For example, African schools were located in predominantly African communities. Therefore access to quality education has a spatial dimension. Since commuting long distances to school is not possible for many children owing to the substantial financial and time costs involved, accessibility to good schools depends on their spatial distribution (and on changes to that distribution, given past conditions).In Part 2 findings on household behavior using micro panel data are reported. In particular, this part examines the dynamic nature of human capital investments and formation, from early-childhood growth (nutrition intake), to schooling investments, to adolescents' transition to the labor market. Chapter 4 focuses on dynamic human capital production in which earlystage investment can have long-term implications for schooling outcomes (and therefore potentially labor market outcomes).In the analysis of school-to-market transition, it is important to note that South Africa has suffered from unusually high levels of unemployment (or underemployment), which makes the transition from school to labor markets quite difficult (see also results in Table 1.1). Individuals, while in the labor force, do not necessarily transition to employment. In Chapter 5 I identify the impacts of prime-adult mortality, prevalent in recent years because of HIV/AIDS, on adolescent labor supply behavior under circumstances in which completion of high school or higher-level schooling significantly increases the likelihood of being employed.This shock might have affected human capital formation by reducing parental inputs to child schooling and increasing the need for school-age individuals to look for work. Perceived returns to human capital investment (for example, schooling) may also be altered owing to an increase in prime-age adult mortality. Though spatial coverage is restricted to the province, these are unique panel data that started before the transition from apartheid to democracy. 10 KIDS involved three rounds: round 1 in 1993, round 2 in 1998, and round 3 in 2004. Compilation of the dataset, which is focused on the gathering of income and expenditure data via a detailed questionnaire, followed the national PSLSD, which was carried out across South Africa in 1993 and managed by the World Bank. As such, KIDS round 1 is a geographic subset of households from the national survey, with colored and white households excluded from follow-up.The sampling frame for the 1993 national survey was based on a two-stage self-weighting design. The 1991 Census Enumerator Subdistrict (or an equivalent unit such as a village or village group) was the first-stage unit; systematic sampling was then applied to households within that unit during the second sampling stage (see South African Labour and Development Research Unit 1994 for further details).In 2004 data were collected for 1,426 African and Indian/Asian households across 68 (rural and urban) \"clusters\" in KwaZulu-Natal; 867 households interviewed contained key decisionmakers in 1993 (see May et al. 2007 for further details on the survey). In round 3 information was collected on about 65 percent of the 1998 household members. 11  The 2004 survey in particular collected some retrospective information on child schooling, such as repetitions and school-start age. Combining these data with anthropometry measures for 1993 and 1998, I can analyze dynamic Institute of Urban and Regional Studies, and the South African Department of Social Development. In addition to support from these institutions, the following organizations provided financial support: the Department for International Development South Africa (DFID-SA), the United States Agency for International Development (USAID), the Andrew W. Mellon Foundation, and the National Research Foundation Norwegian Research Council, through a grant to the University of KwaZulu-Natal. 10 More recent panel data from the country include: four rounds of the Cape Area Panel Study, a longitudinal study of the lives of youths and young adults in metropolitan Cape Town (see www.caps.uct.ac.za for details), two demographic surveillance sites providing annually collected panel data on 11,000 households in Limpopo province since the early 1990s (Agincourt Health and Population Unit), and twice-yearly panel data on 11,000 households in KwaZulu-Natal since 2000 (Africa Centre for Health and Population Studies). 11 May et al. (2007) describe the methodology of the KIDS survey and its history. They show that the follow-up rate was comparatively high for women (relative to men) and for those less than 20 years old in 1998, as these subgroups are comparatively less mobile than other groups. They note that there may have been problems of representativity from round 1. Nevertheless they show that the age distribution, sex ratio, and change by age in the sex ratio of household members (including the children of core household members who were tracked at times outside KwaZulu-Natal) in 2004 are generally representative of the trends for the province when set against the 2001 census data. Children 10-14 years old are overrepresented in KIDS, while younger adults are underrepresented; the authors provide detailed explanation for these situations. human capital formation from early-childhood to school stages (Chapter 4). In the analysis in Chapter 4, I also use some information from the 1993 survey.In the analysis of adolescent transition from school to the labor market (Chapter 5), I use prime-age adult deaths as external shocks to adolescents who change the timing of their entrance into the labor force. In the period from 1998 to 2004, South Africa (and KwaZulu-Natal in particular) experienced a high prime-age mortality rate. In the face of increased prime-age adult mortality, human behavior has been changed to mitigate its adverse impacts. 12 P A R T 1 Institutions P art 1 describes the school system in post-apartheid South Africa. The focus is on quality gaps across schools, which have been persistent since the time of apartheid. Chapter 2 analyzes differences in school quality, measured by the learner-educator ratio, across former white, colored, Indian/Asian, and African schools, using school census data. Chapter 3 examines school fees, which are also indicative of school quality in a system in which schools and communities must raise needed resources; the findings demonstrate that school quality is spatially correlated with residential patterns of different population groups. It is also shown that government subsidy can improve school quality and lower school fees. From both analyses, we conclude that financial constraints seem still to be binding at both the school and community levels in South Africa.Race, Equity, and Public Schools in the First Decade of Post-Apartheid South Africa I n the transition from apartheid to a democratic society in South Africa after the first democratic national election in 1994, the government promised to provide equal opportunities for education to all racial groups and regions (Republic of South Africa 1996a, 1996b). However, as reported in Education Atlas of South Africa (Bot, Wilson, and Dove 2000), there still exist wide variations in major indicators of educational quality across regions. Given the clustered spatial distribution of racial groups in the country, it is not difficult to infer variations in educational opportunities among children across different population groups. In this chapter, I use South African school census data from 1996 and 2000 to assess variations in educational quality of public schools across former population groups and investigate changes in post-apartheid South Africa.It has been recognized both inside and outside South Africa that under apartheid (over which the African population had no control) African schoolsfor example, those in the former homelands-were totally inferior to white schools in terms of funding (Kriege et al. 1994;Marais 1995;Crouch 1996). Differences in conditions between African and non-African schools led to corresponding differences in learner achievement, particularly examination scores in numeracy (Case and Deaton 1999). Unless the government actively strengthens its support of former African schools in terms of budget and personnel allocations, in order to narrow apartheid-created differences in educational quality, a vicious cycle of poverty and low-quality education will persist. Children who cannot receive a sufficiently high quality of education are less likely to be engaged in regular employment and are therefore more likely to remain in the low-income class (for example, Case and Yogo 1999). Since they cannot afford to live in well-off residential areas (in many cases, former white areas), which typically have high-quality schools, they are likely to stay in areas with inferior schools. When high residential rents prohibit access to better schools, this cycle will persist, potentially becoming a crucial determinant of the long-term poverty trap for Africans in the country.To study gaps in educational quality across population groups, I focus on the ratio of learners to educators (teachers and other staff)-the learnereducator ratio (LER)-from two school censuses, SRN 1996 and 2000. In 1995 the government reached an agreement that ratios of 40:1 and 35:1 were to be achieved for primary and secondary schools, respectively, in the next five years. Therefore, LER can provide a good indicator not only of the distribution of education quality but also of the government's policy interventions to achieve educational equity.Recent empirical work shows significant effects of LER and class size on learner achievement, although the literature as a whole contains some ambiguity (Hanushek 1998). The difficulty in identifying causality arises from potential endogeneity in the number of learners and unobserved fixed components specific to school and community, which is likely to be correlated with school inputs.1 For example, Lazear (2001) argues that the effect of LER on learner achievement could be empirically ambiguous because of (often unobserved) heterogeneity in learner quality, that is, discipline. In his model, the optimal size of a class (LER) increases if learner discipline improves, since the probability of disruption in a classroom decreases. To avoid such a correlation between LER and unobservables, recent studies use exogenous variations (changes) in LER and class size to identify the effect on learner achievement (for example, Angrist and Lavy 1999;Case and Deaton 1999;Krueger 1999;Hoxby 2000). In these studies with exogenous variations in LER, the effect is found to be significant. In the context of South Africa, Case and Deaton (1999) show that among Africans under apartheid who were not free to choose schools, LER has a significant effect on learner achievement, particularly in numeracy, while its effect is not significant among whites.Table 2.1 compares means of LERs by population groups in both 1996 and 2000. A striking fact evident from the table is that the gap between formerly African and white schools did not narrow during the period. Formerly white schools maintained their superior situation in the post-apartheid period. Though more detailed statistical analysis is provided in the section on empirical findings, the difference between formerly African and white schools seems persistent and stable.The LER gaps can have long-term implications. For instance, school quality matters in subsequent labor market outcomes (Card and Krueger 1996;Case and Yogo 1999;Dustman, Rajah, and Soest 2003). Based on Case and Yogo's estimates of the impact of LER on returns to schooling investments, for example, the marginal effect of LER on rate of return is around 0.002. The mean LER gap between formerly African and white primary schools, 10.060 in 1996 (Table 2.1), is equivalent to a reduction of 0.0201 in the rate of returns. The reduction is substantial because the average rate of return was 0.089-0.094 for men age 24-28 in 1996. Thus we can infer that inequality in educational opportunities (that is, gaps in LER) is transformed into inequality in labor market earning opportunities in South Africa. 2The organization of this chapter is as follows. The next section sets up a simple framework in which liquidity constraint is highlighted. The following section describes the data that I use in the analysis, SRN 1996 and 2000. The surveys focus particularly on school facility information, in addition to basic information such as the numbers of educators and learners. 3 To identify the former racial groups of those schools, SRN 2000 provides information on the departments that governed the schools under apartheid. Therefore, merging the two surveys, I can systematically track former apartheid departments. I exclude from the analysis of the sample the provinces of Gauteng, Mpumalanga, and Northern Cape since they changed school registration codes (Education Management Information Systems [EMIS] codes) after 1996, preventing an accurate merge of the 1994 and 2000 datasets. 4In the section on empirical findings we see, first, that the LER distribution for formerly African schools differs from that of formerly white, colored, Indian/Asian, and new schools in both 1996 and 2000. In particular, the difference between formerly African schools and white or Indian/Asian schools was found to be statistically significant. A large number of formerly African schools exhibit LERs above the targets set by the government: 40:1 and 35:1 for primary and secondary schools, respectively.To identify how the number of educators was adjusted in response to changes in the number of learners, the estimation strategy takes into account community-school-level unobserved fixed components, using specifications drawn directly from the model in the appendix. First, the adjustments of educators in responding to changes in the number of learners (with budget constraints) differ statistically across racial groups in primary schools, especially the adjustments of subsidized educators. Formerly African schools are more budget (liquidity) constrained than non-African (white, colored, and Indian/ Asian) schools when they employ educators. Second, among secondary schools, the gaps are smaller than those found in primary schools. Interestingly, formerly white secondary schools do not show any significant adjustments to changes in the number of learners during this period, probably because their condition was already optimal. Third, in combined schools (both primary and secondary levels), the gaps between formerly African and Indian/Asian or new schools are significant. This observation reflects the fact that combined schools are regionally concentrated in certain districts and that there are few formerly white schools of this type. Fourth, in the analysis restricted to nonsubsidized (privately employed) educators, the number of educators does not significantly respond to changes in the number of learners. In this sense, the liquidity constraint is more binding at the school level than at the government level.Setting I factor out possible reasons for changes in the number of learners. First, natural population growth contributes to cohort size, and therefore the number of school-age children in a community. Second, after the abolition of apartheid, households could freely migrate from formerly African areas to white areas. Third, parents can send their children to live with distant family members or foster in other children, or children can go to private schools that formerly belonged to different population groups, even though these schools are not located in their residential areas. 5In response to cross-sectional differences as well as changes in the number of learners in public schools, it is desirable to adjust the number of educators optimally to maintain efficiency in learning and equity among children. There are several scenarios. Consider a stationary environment in which the total number of learners does not change. If the provincial government coordinates the employment of teachers and allocates them among schools with no transaction costs, the optimal ratio of learners to educators can be smoothly maintained. The ratios will be equalized across schools.If schools have discretion over the employment of educators independently of the local government (for example, principals decide to employ teachers with the approval of school governing bodies consisting of community leaders, parents, and educators), the adjustment of educators depends on decisionmaking in each school and mostly on its financial condition. Currently in South Africa, many public schools receive insufficient financial support from the government. In this case, equalization of the ratios is not guaranteed. In other words, the equalization of LERs is a necessary condition for, among other things, unitary decisionmaking (or interventions) by the government. Even if the local government suffers budget constraints, unitary decisionmaking will lead to the equalization of LERs.In response to changes in the number of learners, budget constraints may matter at the school and government levels. Without population growth, under unitary decisionmaking by the government it is easy to transfer educators from one school to another to equalize ratios across schools. This is especially important under the post-apartheid regime, in which people are essentially free to migrate. With population growth, however, to maintain the current LER, the adjustment of educators (like the adjustment of capital stock) depends on the government's budget (liquidity) constraint, since the government needs new educators.When public schools receive little or no government subsidy, the situation is more serious. Schools with binding budget constraints that cannot collect enough school fees from learner households are likely to have great difficulty in hiring more educators. Unlike with unitary decisionmaking, there will be more variations in LER across schools in this case, since financial conditions are likely to be different between schools. As a result, for quasi-privatized and budget-constrained public schools, LERs could have wide variations in cross-section as well as time series. The appendix presents a simple model that formalizes this idea.In the empirical analysis, I estimate a response function of educators to learners taking into account school budget condition:where β* ≥ γ it (p), p denotes population group, and μ i is the fixed effect that reflects unobserved school-and community-specific components. I(y* > φ it ) means that the school is not budget constrained, while I(y* < φ it ) means that it is budget constrained. In the latter case, adjustment of educators is lower than the optimal. The derivation of γ it (p) is given in the appendix. Here local condition f is also represented by population group p. Since, in the analysis using the SRN, the information on subsidies and school fees is not available, I assume the patterns according to which these two variables are determined differ across population groups. I estimate γ it (p) as a reduced-form parameter in the estimation of (2.1).In equation (2.1), as in many cross-sectional studies, it is likely that the number of learners is correlated with the unobserved fixed component μ i , which will bias the OLS estimate of the slope. For example, in communities experiencing rapid urbanization, where teachers can easily commute from urban centers and learners can migrate to them, the numbers of learners and educators will increase simultaneously. In this case, OLS estimates are biased upwardly. Assuming that parameters do not change over the four years, after conditioning on cross-group differences, we difference them between two periods:where Δ is the differencing operator. The shocks are assumed to be ex post in each period.The parameter of interest represents the degree of liquidity constraint. As we will see in the section \"Distribution Comparison,\" the empirical distributions of LER motivate the analysis of determinants for the observed LER gaps across population groups. However, naive comparisons of LER distributions cannot identify school and government behavior-that is, how the number of educators changes in response to changes in the number of learners-and how likely liquidity constraint is to be binding in adjusting the number of educators. Changes in the number of learners represent fundamental changes in schools or the government that lead to adjustment in the number of educators. 6Since the main interest of this chapter is differences in school behavior across population groups, we group schools into five groups-African, white, colored, Indian/Asian, and others (new schools)-in equation (2.2). I use race-group dummies to approximate differences in patterns where liquidity and subsidy constraints bind decisionmaking regarding the employment of educators. In this framework, we cannot distinguish whether the liquiditycum-resource constraint is binding or the target ratio is different across the groups. I exclude the latter case here. In the estimation, I also use magisterial district dummies so that we capture variations across population groups within districts, in which schools and communities are more homogeneous than those in an entire province. By focusing on within-district cross-race differences, we can identify how differentially the liquidity constraint is binding the decision on adjusting the number of educators across population groups. In the null hypothesis that the entire budget is pooled over all population groups, the liquidity (budget) constraint should bind equally for all the groups.The estimation of equation (2.2) requires additional consideration. It is possible for the past shock in the number of educators (ε i1 ) to partly cause subsequent changes in the number of learners, E[ΔL i∈p ε i1 ] ≠ 0. Suppose that a positive shock to the number of educators increases the incentives for potential learners to attend the school. This positive correlation leads to a negative bias in the OLS estimator in equation (2.2). In this sense, the endogenous movement (decisionmaking) of learners influences the magnitude of the negative bias. Under this circumstance, it is likely that the true value of the slope is somewhere between a possibly upwardly biased estimate from the cross-sectional analysis in equation (2.1) and a possibly downwardly biased estimate from the panel analysis in equation (2.2). 7 SRN, with its main focus on the conditions of school facilities, was initially fielded in 1996. In that survey, trained fieldworkers attempted to visit all schools in the country and collected information from educators, mainly school principals. Although the survey's coverage was found to be imperfect because some schools were not accessible during the survey preparation stage, this was the first systematic school census in the country. Schools were identified by school codes provided by provincial departments of education (EMIS codes) and by province codes, and also by latitude and longitude using a global positioning system.Four years later, the National Department of Education conducted the second round of the survey. This time, however, data were collected through questionnaires distributed to school principals. This means of data collection alerts us to possible errors in the recorded answers, especially those concerning facility conditions: principals might want to attempt to get more funding by underreporting their school facilities, for example, building condition and the number of classrooms. To minimize this problem, the questionnaire was designed to elicit only changes from 1996 conditions, which were described on the distributed form.Yet even with potential measurement errors and bias in some questions, the 2000 survey accomplished almost-perfect coverage of schools in the country. In particular, fieldworkers visited those schools that were missed in SRN 1996. Unlike SRN 1996, the 2000 version does not include technical colleges and special schools, but it completely covers all primary, secondary, and combined schools. (For detailed discussions of SRN 1996and 2000, see EduAction 2001.) The data that I use here were provided by EduAction, Durban, and the National Department of Education, Pretoria. Table 2.2 shows summary statistics.For the purpose of constructing panel data, it is important to note that EMIS codes are also available in SRN 2000. However, some provincial departments of education changed their EMIS codes after 1996, and the details of the code changes are not transparent. Therefore I decided to include only provinces that used the same EMIS codes in 2000 as in 1996. As a result, Gauteng, Mpumalanga and Northern Cape are excluded from our sample for the analysis that follows.Another important feature of SRN 2000 for our purposes is that it asked about former departments that governed the schools under the apartheid regime. From this information, we can correctly identify the racial background of each school under the previous regime. The correspondence between former departments and population groups is as follows:   (1996,2000). Notes: Primary schools in 1996 include normal primary (grades 1-7), junior primary (grades 1-4), and senior primary schools (grades 5-8 • African-Venda Education Department • All races-New schools established after 1994, New Education Department Under the post-apartheid regime, children of any racial origin can attend any school. In our analysis, those schools established after the end of apartheid are grouped as \"new schools.\" It should be emphasized here that, even though schools are sorted by former departments, the period covered by our analysis falls after apartheid. Therefore, all schools are theoretically racefree in both 1996 and 2000. However, the reality of the racial composition of learners did not change substantially until 2000. The majority of formerly African schools are still in communities that are predominantly African, so the learners in those schools are still mostly African. Some formerly white schools now accept children from African families that have relatively high incomes and reside within commuting distance. Therefore-although my focus on population groups is approximate, as it does not reflect the exact racial composition of each school-I can capture the essence of social distance across racial groups in South Africa, where most schools and communities are still racially homogeneous even after apartheid.8 Three types of empirical analyses are conducted here. First, I statistically characterize the distributions of LERs in 1996 and 2000 in different population groups. Cumulative distributions of LER are compared and Kolmogorov-Smirnov tests are used for statistical comparisons of LER distributions of formerly African schools with other schools. Second, I depict the relationship between changes in educators and learners for each population group. Third, I conduct a panel analysis that differences out fixed effects to estimate the response of the number of learners to the number of educators.Figures 2.1 and 2.2 show LER distributions in public primary and secondary schools for 1996 and 2000, respectively. Primary (grades 1-7), junior primary (grades 1-4), and senior primary (grades 5-7) are aggregated as primary schools, and secondary (grades 8-12), junior secondary (grades 8-10), and senior secondary (grade 11-12) are grouped as secondary schools. In these figures, distributions are shown for schools for different former population groups-African (African), white, colored, Indian/Asian-and for new schools.For formerly African and new schools, LER distributions have long upper tails. For the sake of display, values of LER larger than 200 were omitted in these graphs, though there are substantial numbers of formerly African and new schools in this range. On the other hand, the distributions are shown to be concentrated within a range of relatively small values for formerly white, colored, and Indian/Asian schools. This basic characterization of differences in LER distributions across former population groups is valid for all types of schools-primary and secondary. The main findings on cross-group differences are quite similar in both primary and secondary schools.To statistically characterize differences in the LER distribution between formerly African schools and the other schools, I use Kolmogorov-Smirnov tests (Tables 2.3A and 2.3B). Table 2.3A shows two basic findings. First, in the country as a whole, the LER distributions of African primary and secondary schools are statistically different from those of white, colored, and Indian/ Asian schools in 1996 and 2000. In particular, the test statistics show that the distance between African and white has not narrowed from 1996 to 2000.Table 2.3B shows provincial-level results for the Kolmogorov-Smirnov tests. At the provincial level I find that the results differ between provinces in 1996 and 2000. In 1996 the distance between African and white primary schools is found to be significant in many provinces, except Free State and North West, where the distances to colored, Indian/Asian, and other groups are also insignificant. In 2000, however, African and white primary schools are significantly different in all provinces. In this case, the difference remains quite robust between African and white in post-apartheid South Africa. Findings for secondary schools are stronger than those for primary schools. In Free State and North West, where African and white are not different in primary schools, the distance is statistically significant in both 1996 and 2000.The findings clearly confirm our prior perception that formerly African schools, at both the primary and secondary levels, have not improved relative to formerly white schools, even under the post-apartheid government. This finding does not directly imply that African children in the country suffer more severely from low quality of education than white children. In post-apartheid South Africa, all schools must not discriminate among children based on their origins, and children of any racial origin are selectively admitted. However, since most communities are still racially homogeneous, the    To cope with variations in the slope parameter across population groups and regions, and possibly at various levels of learner changes, I sort them by population groups, to the extent that the sample size of each group can permit analysis. In preliminary analyses, I found that if I used primary and secondary schools separately, sample sizes for non-African schools at the provincial levels became too small.9 Figure 2.3 depicts relationships between changes in primary-school educators and learners in 1996-2000 for all races and for different racial groups. The samples I use in this exercise are constructed as follows. Among schools that are successfully matched between SRN 1996 and 2000 by EMIS codes and province codes, I use only those classified by funding type as state or state-aided in 1996, those that show learner changes in the range of -1,000 to 1,000, and those that show educator changes in the range of -100 to 100. I dropped observations with missing values for the total number of educators in 1996 or 2000. Primary schools include normal primary (grades 1-7), junior primary (grades 1-4), and senior primary (grades 5-7) in the 1996 survey. Similarly, secondary schools include secondary (grades 8-12), junior secondary (grades 8-10), and senior secondary (grades 11-12) in 1996. If schools changed the range of grades offered during the period, they experienced large increases or decreases in learners.In Figure 2.3 the relationship is close to linear but shows a slightly convex shape. However, it is asymmetric between the point at which the number of learners increases and the point at which it decreases. The response of educators to increases in the number of learners is larger than that to Free State, KwaZulu-Natal, Northern Province (Limpopo), North West, and Western Cape), I used the same criteria used in Figures 2.1 and 2.2. In all provinces, changes in educators responded to those in learners positively. Though we find some variations in the slope across provinces, the magnitude is very small among African schools. Strong nonlinearity cannot be detected in these figures. However, it seems that while some provinces, such as Eastern Cape, Northern Province, and North West, did not experience large changes in learners at the school level, other provinces, such as Free State, KwaZulu-Natal, and Western Cape, have gone through large changes in number of learners.For African secondary schools by province, it is also found that changes in educators responded to those in learners positively in all provinces. However, except in KwaZulu-Natal, the variations in educator change seem to be larger in this case than those for primary schools. In this sense, the equity-improving interventions were larger in secondary schools, and thus worked to narrow the gaps across schools.   (1996,2000).   (1996,2000).   (1996,2000).decreases in the number of learners. In Figure 2.3B, in African schools, we have the same observations. However, for white, colored, and Indian/ Asian schools, nonlinearity becomes very strong (Figures 2.3C, 2.3D, and 2.3E). In white schools, while most observations show small changes in the number of learners, the overall shape is kinked with concavity (that is, there is slower adjustment when the number of learners increases). Among colored and Indian/Asian schools, however, the relationship is kinked and convex. Most observations in these groups also show small changes. In new schools that were established after 1994, the plot of results is nearly a straight line. Figure 2.4 depicts results for secondary schools. As in the case of primary schools, a nearly linear but slightly convex relationship is observed in all schools in the country (Figure 2.4A). The basic relationship holds among African schools (Figure 2.4B). Figure 2.4C shows white schools: it looks strikingly similar to the case of primary schools. Though observations are less concentrated in showing small learner changes than those for primary schools, the shape is kinked and concave. Strikingly, the number of educators does not respond significantly to large changes in the number of learners, but it does respond to small changes.One interesting observation from all these figures is that the cross-school variations in educator changes are quite large. The variations are large even with small changes in learners. One way to explain this finding is that government interventions narrow the initially existing differences in LER, and that LER does not directly respond to changes in the number of learners. Alternatively, even without government intervention, schools might have made efforts to weaken their liquidity (budget) constraints in order to adjust the number of educators. In either case, we expect that larger 1996 LERs induce larger subsequent increases in the number of educators.In this section I show the results from the estimation that incorporates community-school level unobservable fixed components. To deal with the fixed effects, I difference out those between 1996 and 2000, using changes in the number of educators and learners. Even in this differenced form, districtlevel dummies are included to control districtwise common changes in this period. Our focus in this exercise is on the differences across population groups in the response of the number of educators to changes in the number of learners. With district-level dummies, this procedure can essentially identify cross-group variations in the degree of educator adjustment within each district.  (1996,2000).   (1996,2000).   (1996,2000).This identification strategy calls for attention to the spatial residential pattern in South Africa, which is segregated by racial group. In the former homeland districts, for example, most of the communities are predominantly African, so that there exist few formerly white schools in such regions. This situation makes it difficult to identify gaps in school behavior between formerly African and white schools. However, provided that socioeconomic circumstances are diverse in different districts, it is more important to control the districtwise heterogeneity in terms of learners' movement and school decisionmaking.We also need to consider a possible correlation between past shocks to educators and subsequent changes in learners over time. If such a correlation exists, the OLS estimates in the differenced forms will provide downwardly biased estimates of the slopes. In this section I not only difference out the fixed effects but also use instrumental variables available from the 1996 data, so that consistent estimates of the slopes are obtained. The details of this process were discussed in the \"Framework\" section. The results are summarized in Tables 2.4A and 2.4B, respectively, for primary and secondary schools. I also decompose the educators into two categories: subsidized and nonsubsidized.Column 1 in Table 2.4A shows the response of the number of all educators to changes in the number of learners. First, the number of educators responds positively to an increase in the number of learners. Second, in this basic specification, differences from African schools are all significant. The number of educators increases more in white, colored, Indian/Asian, and new schools than in African schools, as the number of learners increases. Third, the interaction of learner changes with the indicator of an increase in learners shows some asymmetry in the educators' adjustment.In column 2, where I use only subsidized educators, the basic findings that I obtained for all educators hold. Column 3 shows the case of nonsubsidized educators. Contrary to the previous cases, the number of nonsubsidized educators increases more significantly when the number of learners increases than when it decreases.Table 2.4B displays the estimation results for secondary schools. The results are very different from those for primary schools. The benchmark response of African school educators is significant in all three cases. First, except for white schools, there are no significant differences in educator adjustment behavior from African schools. Second, and very interestingly, the response of the white school educators to changes in the number of learners is smaller than in the benchmark African school case. Adding the two estimates gives nearly zero response in white schools. This large difference from the primary school case suggests that at the secondary level the number  of educators has already been close to the optimal level among white schools, so that even in response to relatively small changes in the number of learners, schools do not adjust the number of educators significantly. Third, colored schools show stronger responses than African schools. There seem to be larger behavioral variations across different population groups in secondary schools than in primary schools. Fourth, except for the case of nonsubsidized educators, changes in the number of educators are larger when the number of learners increases than when it decreases.The empirical results show that opportunities for education in public schools are still unequal between African and white children in South Africa, even after the end of apartheid. The LERs in public primary and secondary schools differ statistically between African and white groups. During the period 1996-2000, overall differences in the distribution of LERs have not changed, and in some cases the gaps have been even reinforced for secondary schools. The resulting inequality in opportunities for education could lead to persistent inequality in labor markets and earning opportunities since the quantity and quality of education crucially determine labor market outcomes.The dynamics of school education also demonstrate strong inequity between population groups. The number of educators responds to changes in the number of learners in all population groups at the primary school level. However, the adjustment in the number of educators is significantly larger for formerly white, colored, Indian/Asian, and new schools than African schools. On the other hand, at the secondary school level, the results do not display significant apartheid-type inequity. In the case of white schools, the number of educators does not respond to changes in the number of learners, probably because these schools have retained their initial superiority.One possible reason why LERs have not converged even after the abolition of apartheid is that school fees charged at formerly white schools increased to prevent the entry of African children (Selod and Zenou 2003) (though the empirical analyses in this chapter do not address this proposition). This screening mechanism could partially explain changes in the number of learners and why LERs did not converge rapidly. It is also very difficult to obtain data on racial composition in each school.Our empirical results call for stronger policy support for African primary schools and schoolchildren, which can contribute to the human-capital-based reduction of the poverty and inequality that have resulted from apartheid in South Africa.School Quality, Clustering, and Government Subsidy G eography becomes critical when access to opportunities is distributed unevenly over space. For example, when good schools are concentrated in urban areas, one must live in these areas to have good educational opportunities and therefore good job prospects. In South Africa, which experienced nearly more than 40 years of apartheid, different population groups were segregated in separate residential areas with unequal access to education. 1 As a result, location was a critical factor. This chapter examines how spatial factors, highly correlated with historical factors, are determining school quality in post-apartheid South Africa.Two factors are relevant to the way in which school quality is determined. First, the legacy of apartheid imposes historical constraints on the spatial distribution of income and population groups. Good schools are located in selected areas. This has maintained interracial diversity in access to good education, as well as racial and socioeconomic homogeneity within neighborhoods. 2  Second, even if the mobility of populations was unrestricted after the abolition of apartheid, household-level financial constraints coupled with the imperfect credit market often prevent the poor from moving into those well-off areas that offer better educational opportunities. Thus the opportunity for better education is geographically correlated with land prices. 3 Even though African children can commute to formerly white schools, in so doing they incur additional transportation and time costs. Accordingly I explore the impact of apartheid on the spatial distribution of quality education under the post-apartheid regime, in which spatial mobility is legally unrestricted. 4This chapter asks how historical and location factors affect access to quality education in post-apartheid South Africa through the use of a unique database combining the 2002 school census and the Community Profile Database from the 2001 South African census. With the addition of GIS information, these data enable us to identify the location of a given school and to correlate that with local socioeconomic characteristics.Given the abovementioned spatial dependence, the role of government subsidy is expected to be significant in creating equitable and equal access to education. I assess to what extent government subsidy disconnects the linkage between local resources and school quality, given that school fee determines school quality. For this purpose, I use school finance data from the province of KwaZulu-Natal to analyze the dependence of school quality, measured by the LER, on school fee and government subsidy. Selod and Zenou (2003) examined the role of school fees in screening children from different backgrounds in a spatial model, showing that whites tend to overprice education in order to limit the numbers of African learners. It is likely that a high school fee supports high school quality even in South African public schools, as well as keeping the community and schools racially homogeneous. This chapter also provides some insight into this question.The chapter is organized as follows. The section \"Dependence of School Quality on Local Resources\" discusses how school quality (inputs) can depend on local resources in South Africa. The following section describes the empirical framework and data used in the analysis.Empirical results are summarized in the next section. First, some key spatial features of school fee distributions and population group compositions in South Africa are demonstrated and linked with the history of apartheid. School fees are significantly higher among formerly non-African schools and in predominantly white areas.Second, while local population-group composition and former apartheid departments of education still influence the way in which school fees (and thus school quality) are determined for local public schools, the role of local income opportunity is also significant, especially in large cities. Third, evidence from KwaZulu-Natal shows that school fees and per-learner government subsidies improve school quality, decreasing the LER and implying that more progressive allocation of subsidies can improve the quality of schools located in under-resourced communities. Policy implications are discussed in the final section.School quality is a function of school inputs, which in the context of South Africa are determined by local resource availability (through school fees) and government subsidy. Here \"school quality\" does not mean learning achievements or educational outcomes. We assume that a given outcome is a function of not only school quality (inputs)-including the availability of qualified teaching staff-but also of learners' family backgrounds and their own efforts and ability. This chapter focuses on the resources available to schools.Distinguishing between school inputs and educational outcomes is important. To analyze the determinants of educational outcomes, it is necessary to use some outcome measures such as test scores at the individual level or school averages. Qualified empirical analyses prove significant causal effects of school inputs on achievement (for example, Card and Krueger 1996;Angrist and Lavy 1999;Case and Deaton 1999;Krueger 1999;Hoxby 2000;Dustman, Rajah, and Soest 2003), though the literature has in general drawn mixed conclusions (Hanushek 1998), and causality seems to depend on subjects (Steele, Vignoles, and Jenkins 2007).In the context of South Africa, Case and Deaton (1999) show that school resources, measured by LER, can explain test scores using variations in the ratio from that under apartheid.More directly van der Berg (2007) used matriculation test pass rates to analyze the effects of school resources and socioeconomic factors on learners' learning performance. School fees, LER, and average teacher salary significantly influence the matriculation pass rate. Interestingly, former departments also have a significant effect on the rate. However, if the sample is restricted to formerly African schools, LER loses its statistical significance, implying that resource variations within this group are not relatively large. The level of school fees, correlated with local socioeconomic factors (as discussed later), significantly explains the matriculation pass rate even within formerly African schools. 5The mobility of learners and the dynamic nature of human capital formation raise other concerns. Since learners, especially at the secondary level under the post-apartheid regime, can potentially choose their schools more freely, endogenous school choice (mobility across communities) can be an important factor in determining educational outcomes at the school level. 6  Similarly, since prior investments in human capital affect educational outcomes at later stages, schooling inputs and outcomes at the primary school level are expected to influence those at the secondary level. 7 While mobility of learners can potentially weaken the spatial correlation between local factors and school outcomes, the dynamic production of human capital can strengthen the correlation.To understand the linkage between school quality and local resources, we need to know the roles of school governing bodies (SGBs). It is the SGB-a group consisting of the principal, teachers, community leaders, parents, and in some secondary schools, learners themselves-that sets school fees. Accordingly, the school fees charged represent the community's ability to pay for education. 8 SGBs are playing an even greater role now; under recently implemented funding reforms, provincial governments allocate school subsidies according to local poverty measures. To assess the quality of education, information on school fees charged by local public schools is used. In South Africa, school fees determine not only school quality but also the likelihood that residents will be able to afford investments in schooling.Until recently, government educational subsidies in South Africa have been limited, so financing of schools relies heavily on the collection of school fees-in effect a user charge-from parents. As mentioned in paragraph 46 of the 1998 Norms and Standards for School Funding, Ironically, given the emphasis on redress and equity, the funding provisions of the Act appear to have worked thus far to the advantage of public schools patronized by middle-class and wealthy parents. The under the post-apartheid regime. We argue that the ability to hire more qualified teachers depends on the community's income level (school fees), which is spatially clustered in today's South Africa. 6 Van der Berg (2007) reports that learners do not systematically move to better-quality schools probably because of lack of information on school performance, and that the amount of movement to private schools is minor. However, this observation was based on Western Cape province, so it is difficult to generalize. 7 In a slightly different context, but one highly relevant to this issue, Yamauchi (2008) showed significant effects of preschool nutrition intake (forming early-childhood human capital) on schooling outcomes. 8 See the 1998 Norms and Standards for School Funding (Republic of South Africa 1998), which was announced in response to the South African School Act (Republic of South Africa 1996b).apartheid regime favored such communities with high-quality facilities, equipment and resources. Vigorous fund-raising by parent bodies, including commercial sponsorships and fee income, have enabled many such schools to add to their facilities, equipment and learning resources, and expand their range of cultural and sporting activities. Since 1995, when such schools have been required to down-size their staff establishments, many have been able to recruit additional staff on governing body contracts, paid from the school fund.As discussed in the introduction to this chapter, local resource availability is determined by historical and spatial factors, which are correlated in the current empirical context, given limited government subsidy. Choice of residential area is limited even now, so schools that are locally available to African children are largely formerly African institutions, many of which were historically disadvantaged and remain so. Schools in well-off areas can charge higher school fees, which not only finance school inputs but also allow them to avoid the enrollment of children from low-income families.A school fee represents the community's capability to finance local public education. Yamauchi and Nishiyama (2005) analyzed the effect of local income distribution on the determination of school fees, showing that inequality decreases the level of school fee. Thus low-income groups in a community pull down school fees, an outcome that decreases school quality for all children in the community. 9If school inputs depend on local resources, to what extent does government subsidy disconnect the linkage between local resources and school quality? How effectively can progressive subsidy change the linkage between school quality and historically constrained local resource availability? In this analysis, I use LER as a measure of school quality (resource) to explore how local resources, approximated by school fees, and government subsidy can jointly determine school quality.There is a potential substitution (trade-off) between local and government resources, both of which determine school inputs. If government subsidy completely equalizes unequal local endowments, school quality no longer depends on local resources. From a policy perspective, we are interested in knowing how differentially elastically our measure of school quality (LER) can change in response to changes in school fees versus government subsidy. In the following sections, we discuss the empirical framework, data, and results.To assess the effects of historical and spatial factors on school quality, we estimate the following equation in which the log of school fee represents school quality:where ln p jkt is the log of the school fee at school j in location (subplace) k at year t; x kt-s is location factors such as local population composition and economic conditions at s years prior to t; z jk is historical factors at school j, such as the former department; and ε jkt is an error term. Officially, a subplace is defined as the smallest geographic unit available from the census, by which we can identify the location as well as its characteristics. The novel feature of this approach is that location factors are discovered from merging school data and geographic database by GIS. 10The data come from two different sources. Local characteristics are taken from the Census 2001 Community Profile Database (Statistics South Africa). This database provides distributions of socioeconomic characteristics in the 2001 census at the subplace level for the whole country. It covers, for example, education, labor force, migration, settlement types, and population group compositions.GIS data available in school censuses can help identify in which subplace a school is located. 11 The school identification codes, EMIS, enable us to merge the Census 2001 subplace data and school censuses. School fees in 2001 are captured in the Annual School Survey 2002 (National Department of Education). The information on former education departments is available in the SRN 2000 (National Department of Education).To answer the question of how the government can improve school quality and support the poor with spatially targeted interventions, we estimate the following school production functions:andwhere Δ is the difference operator, Y jk is the number of educators, L jk is the number of learners, y jk is the LER, and g jkt is the per-learner subsidy from the government. Here z jk includes indicators of former departments. In both specifications, we take the first difference between two periods to eliminate school-and location-specific unobserved fixed effects.The LER is used as a measure of school quality. However, we also admit that this measure can only partially capture overall school quality, which is determined by such other measures as teaching facilities (classroom conditions) and quality of school administration. I constructed the LER from two school censuses in 1996 and 2000, which focus on school facilities. 12 Since the government subsidy allocation had in principle not changed before 2000, we assume that the subsidy reported for 2000 was basically applied to the period before 2000.To supplement the limited number of subsidized educators, community members can collect school fees and employ educators privately. I therefore also test whether a change in the number of learners induces a change in the number of educators who are privately employed in the community.If the government allocates subsidy more to disadvantaged schools (that is, a smaller number of educators relative to the number of learners), potential bias in γ 2 would be upward since differenced ξ jt are positively correlated with per-learner subsidy g jkt . On the other hand, if government subsidy allocation increases inequality in the number of educators, we expect a downward bias in the estimate. However, since fixed unobservables are already differenced out, the systematic component of endogenous subsidy allocation has no impact on our estimates.Finally, the determination of per-learner subsidy is also of interest in the empirical analysis. Though one possible way to eliminate the bias mentioned earlier is to use instruments for g jkt , we lack identifying instruments in the available data. Therefore I simply examine the effects of school fees, the initial LER, former departments, and school type and location fixed effects.For this analysis, I use school and community information from the province of KwaZulu-Natal. School information comes from the Annual School Survey 1999 (Department of Education) and the KwaZulu-Natal Department of Education's Norms and Standards database. The information on school fees in 1999 and 2000 is also from the KwaZulu-Natal Department of Education. In the province of KwaZulu-Natal, therefore, we can track dynamic changes in school fees to check the robustness of the principal findings.To assess school quality, I use SRN 1996 and 2000 (National Department of Education), which focus on school facility information, making them suitable for computing changes in LERs and number of educators from 1996 to 2000. 13Data on government subsidy are from the KwaZulu-Natal Department of Education (Norms and Standards database). Current funding reforms in the South African public education system attempt to allocate more funding to poor schools and communities on the basis of a poverty ranking of schools and areas within each province. I use the information on actual funding during the period January-March 2000, before the implementation of the funding reforms, so that we can assume that it represents the status quo in the period prior to 2000.This section clarifies some features of the public education system in South Africa, using school fees as a proxy for school quality. We need to be aware of the history of modern South Africa, and of two factors in particular. The first is the segregation policy adopted in education under apartheid, by which population groups were separated from each other in various dimensions. In public education, different departments were responsible for different population groups, and children from different population groups were segregated in separate schools. The second factor is the spatial distribution of residential areas and school locations. Because of the segregation policy under apartheid, different population groups were not allowed to live in the same area. Thus formerly white schools are located in formerly white areas.Figure 3.1 depicts the distribution of annual school fees charged for public schools in 2001 (the weight being number of learners). The mean school fee is 431.72 rand, while the median is 50 rand, which implies that the distribution is highly skewed. Interestingly the graph exhibits a clear bimodal distribution, showing that a group of public schools charges higher fees than the majority. It is also possible that their locations have certain characteristics in common.Figure 3.2 depicts school fee distributions for population groups defined by the former education departments, to illustrate the impact of apartheid on school fee distribution. In South Africa before 1994, the Department of Education and Culture's House of Assembly (HOA), House of Representatives (HOR), and House of Delegates (HOD) governed white, colored, and Indian/ Asian schools, respectively. The Transvaal Education Department (TED) represented white schools in Gauteng province. Schools established after 1994 are categorized as a new group. These figures clearly show the importance of the historical influence of the former regime. Schools formerly under the control of HOA, HOR, HOD, and TED charged higher school fees than schools for other groups. The finding suggests that, given that school fees are positively correlated with school quality, formerly white, colored, and Indian/ Asian schools provide higher-quality education than the majority of formerly African schools. Next, the relationship between former departments and population group composition in school neighborhoods is demonstrated. Table 3.1 shows the proportions of African, white, colored, and Indian/Asian populations in the census subplace of school location. Note that the population group compositions are computed from the Census 2001 Community Profile Database, whereas former departments are those under the pre-1994 apartheid regime.It is interesting to confirm that formerly white schools are located in subplaces where the white population is still the majority. Similarly, formerly Indian/Asian schools are in subplaces where the majority population is Indian/Asian. Formerly colored schools are in colored and white-dominated The proportion of each population group is used in the census subplace where a school is located.areas, respectively. Schools under the other former departments for the African population are located in predominantly African residential areas.To disentangle the spatial relationship between school fees and population composition, Figure 3.3A shows kernel regression line linking school fees to the proportion of whites in a given subplace. Given that the movement of the African population to formerly white residential areas was prohibited under the apartheid regime and is still limited today for financial reasons, the proportion of whites in the population tells us whether a particular school is located in a formerly white area.Interestingly, in Figure 3.3B the distribution falls into two groups (concentrations). Higher school fees are likely to be charged in the areas where the majority population is white. 14   3.1) demonstrates not only the systematic segregation policy in the education system under the apartheid regime, but also that location factors and spatial segregation of different socioeconomic groups (correlated with population groups) are important in determining opportunities for quality education in the next generation.Table 3.2 shows two sets of results, for South Africa as a whole and for its metropolitan areas (Johannesburg, Cape Town, and Durban) where population inflow has been significant since 1994. Each specification includes district fixed effects.The points observed in the previous section are confirmed, namely that former education departments and the proportion of whites in the population in a given subplace influence the ability to pay for education quality. In addition, the implications listed in \"Empirical Framework and Data\" are tested here. Income opportunities are measured by average household income, the average years of schooling in the population aged 20-64, and the unemployment rate. To characterize the economic values of residential areas, the distribution of settlement types and population density from the 2001 census are used.Column 1 lists factors that represent apartheid regime and type of residential area. First, the proportion of Africans and whites in the population has significant negative and positive effects on school fees, respectively. Colored and Indian/Asian cases have been omitted. It is clear that the spatial segregation of population groups significantly affects school fees.   Second, schools formerly under HOA, HOD, and TED charge significantly higher school fees. The omitted case here is schools established after 1994 under the New Education Department. Combined with previous segregation in residential locations, apartheid still influences school quality.Third, the distribution of residents among urban, informal, industrial, institutional, or hostel significantly affects school fees. Omitted cases include sparse, tribal, farm, or smallholding settlements. Therefore, schools in urban areas are likely to charge higher school fees, leading to higher education quality. The effect of population density is, however, insignificant.Column 2 considers metropolitan areas. Although qualitatively similar results were obtained, the magnitude of the parameter estimates for the proportions of Africans and whites in the population is greater than that for the results in column 1. In this sense, population group compositions at the subplace level seem to be more influential in the large cities. Similarly, the effects of HOA, HOD, and TED are larger than in column 1. Hence, it appears that in general the former apartheid regime affects school fees more significantly in these metropolitan areas than in the rest of the country. Population composition, however, is highly correlated with income and level of education.Columns 3 and 4 focus on factors that represent income opportunities. These variables are expected to be significant if the credit market is imperfect. In the country as a whole, mean household income and average years of schooling (ages 20-64) significantly increase school fees, while the unemployment rate significantly decreases school fees. These results are consistent with the predictions of the simple model described in the section \"Dependence of School Quality on Local Resources.\"In column 4 the sample is restricted to Johannesburg, Cape Town, and Durban. Mean household income, average years of schooling, and the unemployment rate significantly affect school fees. The income effect is greater here than that in the country as a whole. Consistent with the previous findings on population composition in metropolitan areas, the income gap correlated with population composition matters more in metropolitan areas than nationally. In contrast, the effects of settlement type become weaker in metropolitan areas.Finally, columns 5 and 6 include apartheid-regime and income opportunity factors. Column 5 shows that both factors matter significantly. The magnitude of the impact, however, differs between the two. While population group composition remains as influential as in columns 1 and 2, the effects of mean household income, average years of schooling, and the unemployment rate become much smaller than in column 3. That is, even though financial and labor-market constraints under the current regime seem to be significant, historical factors originating under the apartheid system (partly correlated with income opportunities) are more significant in the way they constrain the ability to pay for school quality and the quality of schooling investments in the next generation.In metropolitan areas (column 6), however, the effect of the proportion of Africans in the population decreases by nearly half (from 0.412 to 0.205) and becomes insignificant, while the effect of average income increases from 0.031 to 0.213 and is thus significant. Socioeconomic factors matter more in these large cities than in the country as a whole.This section summarizes estimation results on school quality determination. School quality is measured by LER and the sensitivity of the number of educators to changes in the number of learners, which I construct from SRN 1996 and 2000. An increase in LER implies a decrease in school quality. 15In the education function that I estimate, inputs are (log transformed) school fee and per-learner funding from the government. As discussed in previous sections, the school fee for 1998 is taken from the Annual School Survey for 1999. School funding information comes from the KwaZulu-Natal Department of Education.Table 3.3 shows our empirical results. Columns 1-3 use school fees in different years. The dependent variables are changes in LER from 1996 to 2000. Former population group, school type, and circuit indicators are controlled. Parameters of interest are school fee and per-learner funding. In those columns, the effects of these revenue conditions are significant and negative. Thus a better school financial situation improves school quality. In a preliminary analysis, the log of the 2000 school fee was included, but its effect on dynamic change in the LER from 1996 to 2000 was insignificant. Column 3 uses per-learner total revenue (excluding government funding), which also has a significant and negative effect on LER.In columns 4 and 5, I test how school financing can change the number of privately employed educators (nonsubsidized educators), controlling changes in the number of learners. First, an increase in the number of learners increases the number of those educators. Second, the log of the 1998 school learners and unobserved fixed components specific to school and community, which are likely to be correlated with school inputs. For example, Lazear (2001) argues that the effect of LER on learner achievement could be empirically ambiguous because of (often unobserved) heterogeneity in learners' quality, that is, discipline. In his model, the optimal size of a class (that is, LER) increases if learners' discipline improves, since the probability of disruption in a classroom decreases. To avoid such a correlation between LER and unobservables, recent studies use exogenous variations (changes) in LER and class size to identify the effect on learner achievement. fee increases the number of nonsubsidized educators, while government funding decreases the number. Third, and most interestingly, the interaction term of log school fee and change in the number of learners shows a significant positive effect, which implies that with a higher school fee (that is, a greater ability to pay for schooling in the community), an increase in the number of learners can be accommodated by an increase in privately paid educators. These results are consistent with the prediction that communities that are capable of paying for schooling investments will increase the quality of education for the next generation with their own resources.In the last column, per-learner funding is regressed on the 1998 school fee and 1996 LER with fixed effects of former population group, school type, and circuit. The estimate shows that in 2000 those schools (and areas) that were initially less well endowed were likely to receive more funding from the government.Neighborhood factors matter, as agents with similar socioeconomic backgrounds are likely to be clustered in the same space. This happens partly because apartheid created inequality in income opportunities (correlated with population groups) and introduced spatial segregation by population group, and partly because even after the abolition of apartheid, financial constraints remained important in choice of residential location, which in turn determines access to income and educational opportunities.This chapter examined historical and spatial factors that determine education quality and a community's capacity to finance education in postapartheid South Africa. Our findings show that both historical and financial constraints matter in terms of access to quality education. First, population group compositions created by apartheid (especially proportions of Africans and whites) at the subplace level and by the former departments of education significantly affect school fees and therefore quality of education. Higher school fees are charged in residential areas with a large proportion of whites in the population. Second, average income, schooling, and unemployment rate at the subplace level also influence the determination of school fees, a finding that implies the existence of an imperfect credit market.Migration to cities became unrestricted when legal constraints were lifted after the abolishment of apartheid, and thus income mobility is now more dynamic in urban areas. As a result financial constraints are more important and population composition is less important in large cities. Wealthier households can move to well-off (that is, formerly white) residential areas to send their children to better schools, a practice that was formerly prohibited. This is happening in the areas surrounding large cities. This chapter also demonstrated that government subsidy can improve the quality of schooling, despite the fact that school quality largely depends on local resource availability. Government subsidies, if progressively allocated to lower-quality schools with a poorer resource base, can potentially disconnect the linkage between local resources and school quality. To narrow the current imbalance, the government should increase financial and personnel support to disadvantaged locales and schools by targeting specific areas-as its progressive subsidy allocation has recently begun to do.If this direct option is limited owing to government budget constraints, alternative strategies may require the redistribution of fees from richer to poorer schools or the implementation of programs that more explicitly move certain kinds of children to better schools-including busing and school vouchers-as already adopted in developed countries.In this chapter I did not account for more qualitative factors, such as the quality of teachers and school management. It is puzzling why learner achievement in South Africa still differs so widely across population groups despite the equalization (albeit very gradual) of school inputs, such as those measured by LER. There is a need to pay equal or even greater attention to qualitative inputs within the school system, in addition to the highly progressive subsidy allocations.Household Behavior P art 2 examines household behavior in human capital formation using micro panel data. Chapter 4 analyzes the formation of human capital from early-childhood nutrition intake to schooling investments and outcomes. Chapter 5 examines adolescents' transition from school to the labor market, with a particular focus on the effect of prime-age adult mortality shocks on their behavior. This chapter also provides evidence of increased prime-age mortality, which implies that the expected returns to human capital investments are being altered in South Africa.Early-Childhood Nutrition, Schooling, and Sibling Inequality H uman capital takes a long time to accumulate; it passes through several phases from early childhood through higher education. While nutritional intake in early childhood forms the basis for a child's health capital, which in turn provides a foundation for subsequent child development, investments in schooling augment the child's knowledge capital, which is directly rewarded in advanced labor markets and in dealing with advanced production technologies. 1 In this chapter I examine the effects of nutritional status and health capital in early childhood on schooling investments and outcomes, using recently available panel data from South Africa.The dynamic process of human capital development creates the possibility that investments in early childhood will influence the optimal amount and effectiveness of investments at later stages (Cunha et al. 2004). Several studies have attempted to identify the link between early-childhood nutritional status and schooling outcomes (for example, Glewwe, Jacoby, and King 2001;Alderman, Hoddinott, and Kinsey 2006) and adult outcomes (for example, Smith 2005;Behrman et al. 2006;Maluccio et al. 2009). 2 In general, the way in which early-stage human capital investments determine the subsequent path of human capital accumulation and future income depends on (1) whether investments in different stages are mutual complements or substitutes and (2) the extent to which early investments and outcomes alter the environment, information, and preferences of children and parents in ways that affect investment decisions at later stages.To answer the question of whether early investments in children affect future outcomes requires an understanding of the complex interactions of market forces and household behavior. First, there is an input-output relationship between health and schooling in the human capital production function. For example, if health capital is an input in the schooling production function, enabling children to attend classes every day, whether health capital augments the productivity of schooling investments or substitutes for schooling inputs, it affects the optimal level of schooling investments. 3 In the former case, I predict a cumulative process of widening inequality among siblings, given differences in nutritional status and health capital in early childhood, since healthy children tend to have better schooling outcomes. If parents are averse to sibling inequality in future earnings, however, they will make greater schooling investments in unhealthy children.Second, parents learn about potential returns to schooling investments from the outcomes of early-stage investments (in nutritional status and health, in our context) and make decisions regarding optimal investments at later stages. In these decisions, parents' preferences concerning sibling inequality in human capital and future income matter. If parents are averse to inequality among their children, they may increase investments in the schooling of their less well-endowed children to equalize the children's future incomes (Quisumbing, Estudillo, and Otsuka 2003). In the context of human capital production, since the outcomes of early-childhood investments signal the expected outcomes of investment at a later stage, parents can react to those signals by changing late-stage investments to maximize their objectives.Third, health capital, as well as schooling investments, generates positive economic returns, especially in the developing-country context (Strauss 1986;Haddad and Bouis 1991;Thomas and Strauss 1997). Therefore, health capital may increase opportunity costs for schooling investments (that is, higher wages), and it may affect intertemporal decisionmaking, creating heterogeneity in the effect of health capital on schooling investments.In empirically assessing this issue, we encounter challenging problems even with longitudinal data for children. The first problem arises from the potential endogeneity of nutritional status; fixed household-specific unobserved factors may affect both child health capital and schooling decisions, creating a positive correlation between them. To eliminate this problem, our approach requires household fixed effects, which base inference on (often small) sibling variations. 4The next section describes the model. Human capital accumulation is modeled as a sequential process in which health is formed at an early stage and schooling investment is undertaken depending on health outcomes. Both health and knowledge (education) capital determine earnings in the labor market. The section on data discusses econometric issues, focusing on specification and identification strategy.Data and variables are described in the section \"Outcomes at the Early Stage of Schooling.\" To measure schooling outcomes, I use the 2004 KwaZulu-Natal Income Dynamics Study (round 3), which collected individual-level information such as enrollment, age schooling started, grade completed, grades repeated, and expenditures from children aged 7-20. 5 To supplement the main analysis, the survey also used the results of simple mathematics tests given to children aged 7-9 to measure their learning performance. Therefore, combined with the information on nutrition and health outcomes for children aged 1-5 available in the 1998 survey, I can investigate the effect of earlychildhood nutrition on schooling investments and outcomes of children aged 6 years and above. One advantage of focusing on the early stages of schooling is the high enrollment rate at the primary school level, which minimizes a selectivity problem arising from time allocation decisions for children at later stages of schooling.The section \"Outcomes at the Later Stage of Schooling\" summarizes empirical results. First, data for siblings showing the effects of nutrition (as indicated by height) on the age schooling started and the grade completed demonstrate that for the majority, children of normal height (as measured by tification strategy is based on findings that under credit-constrained circumstances, income shocks (such as drought and flood) change consumption, which affects child growth (Foster 1995;Hoddinott and Kinsey 2001). Glewwe, Jacoby, and King (2001) take a similar approach to sibling estimation, using longitudinal data on Filipino children, but their identification strategy uses information on older siblings when the child was younger than 3 years. Alderman et al. (2000) use price data, interacted with parents' education and child gender, as an instrument for child height growth in Pakistan. This chapter uses information on the availability of healthcare personnel in sample communities in 1993, right before most children in the 1998 sample were born. During South Africa's transition to a democratic nation, the government implemented propoor interventions in the health sector by building healthcare facilities and increasing the numbers of healthcare personnel. Our identification strategy uses these post-apartheid dynamics, captured by age (birth year), which differ across differently endowed communities. 5 The survey is representative in the province of KwaZulu-Natal, which has the largest population in the country. We do not have any reason for attributing our findings specifically to education systems in the province and the Zulu families in particular. For political reasons, however, this province experienced more violent turmoil than other provinces during the transition from apartheid to democracy, which delayed the implementation of the first national democratic election to 1995. This situation might have more adversely affected schooling behavior among children who entered school during the transition period in KwaZulu-Natal than in other provinces.height-for-age z-score) start school earlier, complete more grades, and repeat fewer grades. The analysis also identifies some outlying observations among taller children (who make up less than 5 percent of the sample) that show a negative effect of the height z-score on schooling outcomes. However, it also remains highly possible that ages for these children in 1998 were underreported, so their height z-scores were overestimated. I also find that, although better nutrition and health status in early childhood improve primary school outcomes, this positive effect diminishes over time as children age. The smaller effect observed among older siblings may also reflect the fact that the effect of nutrition on height is large among children aged less than 3 (in 1998), and height may rebound afterward. 6Second, the analysis of mathematics test results, using the sample of children aged 7-9, shows that health capital, measured by height in early childhood, has a significantly positive effect, implying that early-childhood nutrition affects learning performance at the early stage of transition to schooling.This section describes the empirical framework used to assess the effects of early-childhood health capital on schooling decisions and outcomes at a subsequent stage. The schooling equation iswhere i, j, and t denote household, child, and time, respectively, and q ijt is schooling inputs or outcomes; h ijt-1 is health capital, which is measured by the height-for-age z-score (formed at t -1); a ijt is the age of the child; x ijt is a set of control variables; μ i is a household-specific fixed effect; φ j is child-specific fixed unobservables; and ν ijt is an error term. First, it is important to control the heterogeneity that arises from the current ages. For example, cumulative years of grades repeated increases (but weakly) as children spend more time in school, that is, as their age increases. The score on the numerical tests also changes by age (and grade completed). In the analysis that follows, I assume that age structure in the sample of chil-dren is exogenous; it is uncorrelated with shocks in schooling decisions and outcomes, an assumption which justifies the inclusion of age fixed effects. 7  Second, since it is highly likely that household-specific unobservables μ i are correlated with h ijt-1 , OLS estimates of β 1 are biased. This makes it necessary to eliminate this component from the errors. For this purpose, I include household fixed effects to control μ i . Therefore, the estimation is based on variations across siblings in the household (that is, within-sibling estimates).In the context of panel analysis, the inclusion of household fixed effects has another advantage regarding the attrition bias. Since we look only at within-household variations, given household observations in the two rounds, we do not have to control for household-level attrition problems. Individuallevel attritions are investigated in the section \"Data Sources\" (see Table 4.3).With household fixed effects, however, we use only within-household variations from the sample of multiple-child households. Dropping observations from single-child households reduces the size of our sample, which potentially decreases the precision of parameter estimates in our analysis.Third, even with household fixed effects, we still encounter a potential problem of bias that may arise from a correlation between φ j and h ijt . To eliminate this correlation, it is necessary to use a set of instruments that explains the variations in h ijt-1 but is uncorrelated with either φ j or shocks in schooling investments and outcomes ε ijt . However, the necessity depends on the magnitude of covariations in differences among siblings in the z-score and schooling endowments.For this purpose, we use the information on whether each community (cluster) had different types of healthcare personnel-doctor, nurse, pharmacist, trained midwife, family planning worker, community healthcare worker, or traditional birth attendant-in t -2 (1993 in our setting), interacted by child age. 8 The 1993 initial condition of healthcare personnel availability should EARLY-CHILDHOOD NUTRITION, SCHOOLING, AND SIBLING INEQUALITY 717 Since we only use preschool children in the sample, we can exclude the possibility that parents observe the schooling outcomes of older children before making reproductive (childbirth) decisions. However, as discussed, it is still possible that the health outcomes of older children in the preschool period affect reproductive decisions. 8 The data include information on the distance from the community to the nearest personnel if the community has no personnel. The data also have information on types of healthcare facilities in the community, number of healthcare facilities, and if they do not exist, distance to the nearest one. In this chapter, however, we use only the indicator of whether communities had those personnel. In an early version, an instrument was constructed as follows. First, define the indicator, which has the value of one if children were less than 3 years old (inclusive) between the beginning of 1994 and the end of 1995. The period before the age of 3 is regarded as that when a child's growth is most sensitive to nutritional intake, which reflects economic conditions. This indicator is interacted with cluster fixed effects to capture possible heterogeneity in the impacts of the 1994-95 disturbances on child growth, I(Age 3 in year = 1994 or 1995) × cluster have affected subsequent changes in the numbers of community-level healthcare personnel after 1994, given the fact that South Africa has attempted to build healthcare facilities and increase the number of healthcare personnel in the post-apartheid period. However, it is hard to show quantitative evidence for this conjecture from our sample.In the following estimation, I interact child age with the initial number of healthcare personnel to capture dynamic changes in the personnel specific to a given community. For example, if a nurse was not in the community in 1993, it is likely that the community would have nurses in the subsequent period. Therefore, the interaction between the initial availability of healthcare personnel and child age (cohort) captures community-specific changes in healthcare conditions. 9 Another advantage of interacting the healthcare information with child age comes from the fact that nutrition input is most important in children under 3 years of age, an outcome measured by height in our context. The impact of healthcare on height differs by child age. The individual-level variations, created by age, make it feasible to examine intrahousehold (within-sibling) variations in health and schooling outcomes in the household fixed-effect model.However, we also notice a possibility that dynamic changes in healthcare and education facilities are correlated, though both sets of changes were slow to occur at the beginning of the post-apartheid government. In the first stage, we are concerned with the impact of potential changes in health infrastructure on the health outcomes of the sample children during the period 1993-98, while their schooling outcomes were directly affected by changes in education infrastructure in 1998-2004. This time gap may justify the use of the 1993 initial health infrastructure data (personnel availability) as a potential source of intercommunity variations. However, to the extent that (changes in) healthcare facilities and schools are correlated, the proposed instrument is invalid.indicators. This period also corresponds to the abolishment of apartheid, so new economic opportunities were presented to the African population. On the other hand, unrest associated with the transition was particularly violent in the province of KwaZulu-Natal. Thus there could have been positive impacts as well as negative ones. In addition, to capture the heterogeneity in the impacts related to the initial income level, the indicator is also interacted with total monthly household income in 1993. Although an F-test supports the joint significance of these instruments in explaining variations among siblings in height-for-age z-scores, a Hausman-Wu test rejects the relevance of these instruments. We observed some differences between withinsibling OLS and within-sibling instrumental-variable estimates, but the magnitude did not alter the qualitative nature of our results. 9 Age distribution is potentially endogenous, correlated with private information on policy changes that parents might have had when the society moved into the post-apartheid regime. Parents might have changed reproductive behavior and their fertility might have changed; these factors affect birth timing and therefore the initial health conditions for their children.In the analysis, I use the information on nurses, community healthcare workers, and traditional birth attendants since a preliminary examination suggested that they are particularly important (that is, statistically significant). Since I also include age fixed effects when estimating schooling equations in the household fixed-effect model, the average cohort effects are controlled.I also use the weight-for-age z-score as an instrument to eliminate measurement errors in the height z-score. Results will be compared between the two different sets of instruments.The first-stage regression results are shown in Table 4.1, where the identifying instruments are jointly significant. Interestingly, the result (controlling for age fixed effects) implies that in communities with some nurses in 1993, younger children (those who were born more recently) have gained height relative to older children. On the other hand, older children have gained height (relative to younger children) in communities which had community healthcare workers and traditional birth attendants in 1993. The difference may imply some endogeneity of the initial allocation of healthcare personnel. However, the fact that I examine intrahousehold variations in child health outcome makes this issue less problematic. On the other hand, the weight-for-age z-score has a significant positive effect on the height-for-age z-score.The analysis requires information from different points in time for the same individuals. In this chapter, I use data from the KIDS of 1993KIDS of , 1998KIDS of , and 2004 (see Chapter 1). The sample was population self-weighted in the first round in 1993, based on the 1991 population census, and enumeration-based weights were introduced in 1998. The 1993 and 1998 surveys provide information on anthropometric measures and health outcomes of children, enabling us to construct age-standardized z-scores for height. The 2004 survey provides some detailed information on schooling decisions and outcomes. Our analysis combines the nutritional status of pre-primary-school-age children in 1998 and 1993 and their schooling inputs and outcomes until 2004. 10In the principal analysis, I use as schooling variables (1) age schooling started, (2) grade completed (conditional on current age), (3) number of grades repeated, and (4) mathematics test results. For age schooling started, the 2004 survey asks for the calendar year in which the child started primary school. That year, compared with the current age in 2004, tells us the age at which the child started attending primary school.11 Table 4.2 reports the descriptive statistics of schooling outcome variables: age started school, the highest grade completed, and the cumulative number of grades repeated. First, the age started school increases as the current age increases, which suggests that younger cohorts enter school at an earlier age. Second, the highest grade completed and the cumulative number of grades repeated also increase with the current age.In the mathematical tests, the team implemented four types of numerical tests for children aged 7-9: addition, subtraction, multiplication, and division. The four questions were 3 + 5 (addition), 7 -3 (subtraction), 2 × 6 (multiplication), and 12 ÷ 4 (division). Table 4.3 reports the number of observations with correct and incorrect answers. Note that the sample size for each age group is nearly the same. First, the likelihood of giving a correct answer increases as age increases for all four questions. Second, the difficulty increases as we move from addition to division. Table 4.4 reports the determinants of attrition from the 1998 to the 2004 round and from the 1993 to the 2004 round (see Fitzgerald, Gottschalk, andMoffitt 1998a, 1998b;Thomas, Frankenberg, and Smith 2001). Since our principal analysis focuses on variations among siblings, controlling for household fixed effects, attritions at the individual level are of interest. Given observations in the 1998 round, our concern here is to determine whether the probability of being observed in the 2004 round depends on explanatory variables used in the schooling investment and outcome equations. The sample is restricted to children from households found in 2004 who were between the ages of 1 and 5 in 1998 (0-5 in 1993) with height-for-age z-score values between -6 and 6.In the attrition analysis, I use explanatory variables from the schooling equations: the height-for-age z-score, age indicators, and gender dummy, which are taken from either the 1993 or the 1998 round. To the extent that these predetermined variables are not correlated with attrition, we would not expect bias in estimates in the schooling equations owing to the attrition process. However, a possibility of attrition on unobservables still remains. For example, because of a correlation between shocks (unobservables) in the attrition and schooling equations, attrition on unobservables may cause additional bias in the schooling equations.12   Madison (1998Madison ( , 2004)). Notes: Dependent variable equals one if observed in 2004 and zero otherwise. Estimation uses the linear probability model. Numbers in parentheses are absolute t-values. Sample consists of children with height-for-age z-scores between -6 and 6 from households observed in both 1993 or 1998 and 2004. Estimation uses children from households with multiple siblings. There were 44 attritions out of 513 children and 145 attritions out of 689 children in the periods 1998-2004 and 1993-2004, respectively. Age is defined as years old in the initial year, 1998 for columns 1-3 and 1993 for columns 4 and 5, respectively.households were split and young people formed new households. The analysis in the next section uses the 2004 household definition, since decisions regarding child schooling are supposed to be made in current household units. 13 In both cases, age in 1998 affects the likelihood of being observed in 2004. This suggests that child mortality and mobility depend on the age of the child.Column 4 shows the 1993-2004 attrition rates. Age-specific attrition rates are higher than those in the 1998-2004 case, because we cover a longer period. We do not observe a clear relationship between attrition rate and initial age. In both cases, boys show a higher attrition rate than girls. Column 5 reports determinants of attrition from 1993 to 2004. Contrary to the 1998-2004 case, initial age is not significant.In the analysis of nutrition-height effects on schooling, we screen out observations of children that show ages inconsistent with the 2004 round. For example, age 8 in 2004 corresponds to age 1, 2, or 3 in 1998. Discrepancies arise from the timing of the surveys and birthdays. 14 Mostly for this reason, sample size differs between the attrition analysis and the schooling analysis. Screening out observations of inconsistent ages between 1998 and 2004 with reference to ages reported in the 2004 survey, in particular, excludes large height-for-age z-scores as a result of understated ages in 1998. 15 The same procedure is applied to age matching between 1993 and 2004. In the analysis of schooling outcomes, I include over-age and under-age indicators to control for the birthday effect, in addition to age fixed effects.In fixed-effect models, I use only observations of children from fixed-effect units with multiple children. For example, with household fixed effects, I use households with multiple siblings. This procedure excludes spurious observations from the estimation process.Age Schooling Started Table 4.5 shows the effect of the height-for-age z-score in 1998 on the age at which school was started. The sample consists of children aged 1-4 in 1998 78 CHAPTER 413 Preliminary analysis shows that there are no significant differences even if we use the 1998 units. 14 The 1998 survey asked for \"approximate\" current age, wording that I believe caused variations in reported age. 15 We find some negative effects of large height-for-age z-scores on schooling outcomes. One possibility is that ages were underreported in a previous round, a situation that causes overestimation of height-for-age z-scores for those children. However, by screening out observations with ages inconsistent between rounds, this possibility can be minimized. and aged 7 or above in 2004. 16 Column 1 controls only cluster-level fixed effects, while Columns 2 and 3 report within-sibling estimates. The specifications include current age indicators to control cohort effects. In column 1, greater child height is found to significantly lower the age at which the child started school, though this estimate is likely to be biased owing to a correlation between household-level factors and child height. Column 2 confirms this finding, showing an even greater effect of height on age schooling started. The upward bias suggests that household-specific endowment (which increases the child's age) is positively correlated with the height-for-age z-score. Column 3 shows the instrumental variables estimation result, which implies upward bias owing to the correlation between individual-level endowment and the height-for-age z-score. It is also likely that the difference between columns 2 and 3 might have captured measurement errors in the reported age schooling started. The marginal impact of the height-for-age z-score on the age schooling started is even larger in column 3. To eliminate measurement errors, the weight-for-age z-score is used as an instrument in column 4. The parameter estimate is comparable to that in column 2, which suggests that measurement errors are not a large issue here. These results imply that early-childhood malnutrition delays the age at which a child starts school. 17Current age does not influence age schooling started (not shown in the table), which suggests that the decision to start primary schooling did not change between 1998 and 2004. 18 Thus there is no systematic change in the behavior of entering school across cohorts during the period.Table 4.6 reports the effects of the height-for-age z-score on years of schooling. As children have not completed schooling, specifications include age dummies which standardize years of schooling. Columns 1 and 2 compare estimates with cluster and household fixed effects. Though a positive significant effect is found with cluster fixed effects, the effect is negative and insignificant with household fixed effects. In both cases, girls are more likely to advance grades than boys.To see slope differences, column 3 interacts the height-for-age z-score with an indicator which takes the value of one if the height-for-age z-score is above 2 and zero otherwise. In this case, an improvement in the height-for-age z-score increases years of schooling completed, but the slope turns out to be negative at large values of the z-score (2-6). In column 4, I also include age interactions to investigate how the height effect changes as the child ages.The height-for-age z-score improves schooling while the effect becomes negative when the score is large. We also find that the effect diminishes as the child ages, particularly after age 9.To interpret these findings, we must take into account two possibilities. First, if age was underreported in 1998 for some reason, the expected years of schooling completed in 2004 would be smaller, which makes it more likely to find a negative estimate of the height effect among those large children. Second, we cannot deny the possibility that large children have a high opportunity cost of schooling as returns to their health capital are positive in the labor market or, more generally, in nonschool activities (though this point is not proven in this chapter). Note that the opportunity costs for those healthy children do not arise solely from strictly defined labor market activities, but also from nonschool activities, including informal work (usually not captured in statistics), household work, and crime.The concavity found previously also suggests that parents are averse to inequality among siblings. Given that healthy (taller) children earn more in the labor market, parents may invest more in less healthy (shorter) children to decrease the inequality in human capital and future earnings. Recall that the aversion to inequality may be traced to the concavity of the parents' utility function.The observation that the height effect diminishes as the child ages suggests another possibility. If it is height-for-age before age 3 which matters most and height may also rebound afterward, height at an older age (still ≤5 years) explains fewer of the latter outcomes than height at earlier ages. Alternatively, if returns to height (health capital) increase as the child ages, the incentive to work (to study) increases (decreases).In columns 5-7, I compare within-sibling and within-sibling instrumentalvariable estimates using the sample of children with z-scores below 2. Column 6 uses as instruments the initial availability of healthcare personnel interacted with age, whereas column 7 uses the weight-for-age z-score. It is found in columns 5 and 6 that the instrumental-variable estimate for the height-for-age z-score effect is greater than the within-sibling effect, which suggests downward bias. Higher endowment in academic performance is negatively correlated with early-childhood health capital. 19 Though the difference might have captured measurement errors in reported grades completed, this possibility is less likely than in the case of age schooling started.Column 7 shows a parameter estimate between those in columns 5 and 6, which implies that part of the downward bias is the result of measurement error attenuation, though the estimate is insignificant.Table 4.7 summarizes the results on grade repetition. The dependent variable is the cumulative number of grades repeated. As in the previous section, all specifications include age fixed effects, which standardize years of repetition. Columns 1 and 2 show estimates with cluster and household fixed effects, respectively. The effect of the height z-score is insignificant in both cases. Consistent with the previous finding on grades completed, girls experience a smaller number of repetitions than boys.Column 3 includes slope differences of the height-for-age z-score (as defined previously) to capture possible changes in the slope. These are found to be insignificant. In column 4 I include age interactions in addition to the slope differences. Interestingly, we find positive effects on grades repeated at ages 9 and 10 and height-for-age z-scores above 2. Though these results are only marginally significant, they are consistent with previous findings regarding grade completed.In this chapter we cannot identify possible explanations for these findings. First, it is likely that for large children, ages were underreported in 1998, which leads to a positive estimate of the height effect among them. Second, we still cannot reject a hypothesis that the opportunity cost of schooling is high among large children because market and nonmarket returns to their health capital may be high.Table 4.8 reports the effects of height-for-age on mathematics test scores. In the 2004 survey we implemented four types of basic mathematics tests for children aged 7-9. For each question, an indicator is constructed to take the value of one if the answer is correct and zero otherwise (the descriptive statistics are given in Table 4.1).Table 4.8 shows probit results with cluster fixed effects in columns 1-4.20 Significantly positive effects are found in addition and subtraction-relatively easy computations. The point estimate decreases as the difficulty of calculation advances. In addition, age has a significant positive effect on the probability of answering correctly, which implies that schooling improves learning performance.Column 5 lists the number of total correct answers. Consistent with previous results, an increase in the height-for-age z-score as well as age improves learning performance. However, these estimates are potentially biased owing to omitted household factors, which are correlated with the height-for-age z-scores.This section summarizes our findings on the effects of children's height-for-age z-score in 1993 on schooling outcomes in 2004. Before discussing the results, we note that South African education was in transition from apartheid to democ-  racy during the period 1993-98. The South African School Act and the Norms and Standards for School Funding were announced in 1996 and 1998, respectively; these introduced compulsory nonsegregated education throughout the system (although the reforms would take some time to have an effect).It is also important to note that by 2004, this group of children was in transition from primary to secondary education. This may create heterogeneity by age in the effects of the height-for-age z-score on schooling outcomes. Therefore, it is important to examine possible variations in the height effect by age. The role of positive returns to health is expected to be greater among older children than younger children.Table 4.9 shows the effects of the height-for-age z-score on grades completed and repeated. All the specifications control household fixed effects (using sibling variation within a household). Column 1 has only the height z-score, which shows its insignificance. Age heterogeneity is controlled in column 2. It is interesting to note again that an improvement in the height-forage z-score marginally increases years of schooling completed, but it is likely to decrease grades completed as children age (conditional on age). Greater health capital may discourage further schooling from the primary stage to the secondary stage, given the positive returns to health in the labor market. It is also possible that height at ages older than 3 years may not have a positive direct effect on schooling attainment. To confirm the robustness of this finding, column 3 includes slope differences on the height-for-age z-score. Contrary to the previous findings from younger children (and cohorts), nonlinearity is not found. We also find heterogeneity by age.Columns 4-6 report on grade repetition. In column 5, similar to the results on grade completion, the height-for-age z-score has a negative effect among the youngest group (age 11), but the marginal effect becomes positive among older groups. This finding suggests that (conditional on age) greater health capital may discourage further investments in schooling at the transition from primary to secondary school. Column 6 investigates potential nonlinearity by introducing slope differences, which again show insignificant slope heterogeneity in grade repetition. 21In both grades completed and repeated, girls perform better than boys. However, a preliminary analysis shows that gender does not matter in the effect of the height-for-age z-score on these schooling outcomes (that is, the interaction is insignificant).This chapter examined the effect of early-childhood health capital on schooling investments and outcomes, using panel data from South Africa. Good nutrition and health in early childhood are thought to be a precondition for child development and school learning at subsequent stages.Nutrition intake and health capital in early childhood, measured by the height-for-age z-score of pre-primary-school-age children, enhance schooling investments and improve the outcomes. That is, children who are well nourished and in good health start school at an earlier age, progress further, and repeat fewer grades.We also found that some taller children (z-scores above 2) perform worse than shorter children, but since the number of observations for this segment in our sample is very small and ages might have been underreported, it is difficult to generalize this nonlinearity. It is also important to note that different cohorts in our sample experienced certain historical changes in the South African education system, which might account for the age heterogeneity in the height effect.Impacts of Prime-Age Adult Mortality on Adolescents' Labor Supply I t is widely recognized that prime-age adult mortality has increased dramatically in many African countries. This drastic demographic change is largely attributed to the HIV/AIDS epidemic (for example, Epstein 2004). Excess mortality is concentrated among women between the ages of 25 and 39 and among men between the ages of 30 and 44 (Timaeus and Jasseh 2004). The micro foundations for such increases in adult mortality can be found in studies of the sexual activities and marriage decisions of young adults (Munshi and Myaux 2002;Hallman 2004;Oster 2005;Yamauchi 2007b; Ueyama and Yamauchi 2009).Households can respond to an increase in mortality among prime-age adults in many ways. They can use government grants and formal insurance to smooth their income; they can engage in ex ante and ex post risk-coping or -mitigating strategies (for example, borrowing or relying on remittances) to buffer shocks; they can develop foster-care arrangements or income-diversification strategies). There are, however, problems with these approaches. If these strategies are imperfect in smoothing consumption, prime-age adult death can decrease child schooling investments and increase the labor supply, at least in the short run. Moreover, prime-age adult death also reduces the expected future earnings for the household. This in turn reduces investments in child schooling, given that the period over which capital is formed is long and the loan market is imperfect. Typically, growth in the number of orphans in a society is taxing on both families and the society (see Kelly 2000 quoted in Bennell 2005, 473), and an increase in mortality among prime-age adults, unlike mortality in other age groups, directly reduces the capability of households to secure income.Motivated by recent increases in prime-age adult mortality in South Africa, we use recently available panel data to assess the impacts of such mortality on labor supply behavior in nonagricultural settings by examining the transition of adolescents from school to the labor market and examining female labor supply decisions.The issue at hand in this chapter is of increasing importance to contemporary Africa. The death of adults in their productive years raises serious concerns about the pervasive impact of the HIV/AIDS epidemic on household behavior and on human capital development, particularly through education and labor supply decisions. Since the onset of HIV/AIDS more than two decades ago, mortality rates in many Sub-Saharan African countries have escalated dramatically. In South Africa, according to Statistics South Africa ( 2005), the number of annual recorded deaths in the 20-45 age group more than doubled between 1997 and 2002, from a little less than 90,000 to more than 190,000. Though explicit reports of HIV/AIDS as a cause of death are comparatively low, the number of HIV/AIDS deaths increases sharply when the underlying causes of the disease are taken into account: HIV/AIDS would account for nearly half of all deaths in South Africa (Groenewald et al. 2005). Over 70 percent of the deaths among those 15-49 years old can be attributed to HIV/ AIDS, according to one model (Dorrington et al. 2006, ii, 21).The next section, which deals with mortality changes, presents evidence of excess mortality among prime-age adults in the province of KwaZulu-Natal, using a particular dataset. The panel data I use throughout this chapter have been collected primarily from households in this province.The main focus of our study is on adolescents' transition from school to the labor market and on changes in the time allocation between household production and labor-market activities, potentially as a response to prime-age adult mortality and as part of households' optimal risk-mitigation strategies. While I stress that responses to prime-age adult mortality are not only ex post but could also be ex ante (that is, before the adult's death), adult mortality would have a series of effects.1 It has long-term implications for human capital formation if it causes an acceleration of adolescents' transition to the labor market (see, for example, Young 2005); in turn, to the extent that prime-age adult mortality produces children (for example, orphans) who cannot receive enough education and therefore participate in the labor market, HIV/AIDS creates inequalities in human capital (and earnings) between those affected and those unaffected.On a shorter time horizon, the transition affects the unemployment rate among the young if, as a consequence of their exit from the school system, they are insufficiently educated. Another effect is on adult household members, who survive the crisis and accommodate the mortality shocks by changing their time allocation. For example, household members may need to look for earning opportunities in the labor market or may move to household work to care for the ill. 2 In the analysis that follows, I investigate these issues in detail, acknowledging the possibility that the behavioral response to adult mortality may differ by gender.A number of recent studies attempt to identify the impacts of prime-age adult mortality on child schooling and labor supply (for example, Ainsworth, Beegle, and Koda 2005;Yamano and Jayne 2005). Though I subsequently detail the main relevant findings of these studies, and note that they vary greatly in their methodologies, the point is that they demonstrate the importance of prime-age adult deaths in determining child school enrollment and attendance. In the literature, however, the impact on labor supply is less visible than that on child schooling (Beegle 2005). Because we identify labor supply effects, our contribution to the literature is important in this regard. Also, whereas most of these studies share motivations similar to ours, they deal with agricultural settings. In contrast, our sample comes from semiindustrialized settings in South Africa, where the dominant income source for households is wage employment (see, for example, Dieden 2004). These differences provide a set of risk-coping and -mitigating strategies distinct from those in other rural contexts in Sub-Saharan Africa. They also justify our focus on labor supply and schooling decisions.This chapter is structured as follows. The section \"Mortality Change\" provides evidence of excess mortality in the province of KwaZulu-Natal. The data used for this chapter come from KIDS rounds 2 and 3, conducted in 1998 and 2004, respectively. \"Impact of Prime-Age Adult Mortality on Schooling and Labor Supply\" sets the empirical framework for our analysis. The next two sections respectively describe activity transitions in the sample and present our empirical results.Several findings emerge from the analysis. I find first that deaths of primeage working adults significantly increase the labor supply of both male and female adolescents, stopping their schooling. Deaths of prime-age adults in the future decrease female school enrollment, suggesting that girls shift activity, possibly staying at home to take care of the sick or of the household more generally. Second, since the enrollment of male adolescents is decreased prior to the death of a working adult, their response is different, intended to compensate for an income loss. These findings imply that excess mortality among prime-age adults disrupts human capital formation in the society.The 2004 KIDS data allow us to link death with specific changes among household members, because the data contain retrospective information on those who died between 1998 and 2004. More specifically, the data allow an identification of the individuals who died and when the death occurred. (However, to minimize emotional distress, respondents were not asked about deaths when these had occurred up to three months prior to the 2004 survey.) Combining this information with the roster information on individuals in 1998, I can identify the age at death. This information is critical for our analysis as it enables us to identify changes over the period 1998-2004. This corresponds to the time when South Africa experienced substantial increases in prime-age adult mortality (for example, Groenewald et al. 2005;Statistics South Africa 2005). In other words, the period covered is quite appropriate to identify the impacts of prime-age adult mortality on the time allocation decisions of children and housewives.Figures 5.1 and 5.2 show the changes in mortality rates between 1998 and 2004 for the populations of males and of females, respectively. One of the two series plotted on the figures presents information gathered from the surveyed households. The changes are reported across age groups. Each figure illustrates the non-AIDS mortality rate for South Africa as one of the two series. These data are from the 2002 model estimates of the Actuarial Society of South Africa (ASSA) for 1996 and for all racial groups. Data for 1996 were chosen to provide a credible non-AIDS mortality rate relative to the period spanned by the KIDS data. Age-specific non-AIDS mortality rates for men and women are the benchmark rates against which the KIDS mortality ratesthe second data series in Figures 5.1 and 5.2-are considered in this section (see Table 5.1). Since the sample period I consider spans six years, the mortality rate of the KIDS household members over the period is converted into annual terms assuming a constant mortality rate throughout the period. The two figures show significant increases in adult mortality for those aged 20-44 across gender-the gap between the two lines in the two figures reflects excess mortality in the KIDS sample at the end of the 1990s for each age group. The figures also show that excess mortality is higher for males than for females. 3Table 5.1 reports the probit results on individual-level mortality, controlling for the age-specific non-AIDS mortality rates. The non-AIDS mortality rate is that from the ASSA 2002 model. This variable is included to control for the pre-AIDS level of mortality rate. The various results incorporate cluster fixed effects (columns 1-4), allow for household random effects (columns 5-7), and finally incorporate household fixed effects (columns 8-10).Column 1 reports the results using only the non-AIDS mortality rate and a gender dummy. The results confirm that the benchmark non-AIDS level of mortality significantly explains mortality over the period 1998-2004, though the gender dummy is insignificant. Column 2 introduces age-group indicators to capture age-specific changes in mortality during the period. With those aged 15-19 as the omitted group, we find significant increases in mortality among those aged 20-44. Households faced some exogenous changes in prime-age adult mortality among those aged 20-44 over the period. To distinguish agespecific mortality changes between men and women, columns 3 and 4 report the results of separate estimations by gender. The results show that there are larger increases in mortality among men than among women, particularly among men aged 30 and older. However, during the period under study, those aged 20-44 experienced significant increases in mortality across both groups.Given that these findings might have been generated by unobservable household-specific factors correlated with the household demographic structure, I alter the model specifications. Columns 5-7 report the results of probit regressions for the within-household incidence of mortality with household random effects. Columns 8-10 use household fixed effects in linear models. Both the random-effect and fixed-effect estimates show qualitatively similar effects of the age indicators and individual characteristics. Yet the significance of parameter estimates in the fixed-effect estimates is slightly lower than that for the random-effect probit results.These results are important: they confirm that those between 20 and 44 years of age in 1998 were more likely to die than others in each household. Increases in the incidence of mortality among the prime-age adults should cause a reallocation of resources within the household, including changes in time allocation among household members. This effect arises as household members seek to mitigate the negative impacts of adult mortality on household welfare.    Madison (1998Madison ( , 2004)).Notes: The dependent variable is height-for-age z-score. Prime-Age Adult Mortality and Labor Supply A handful of studies examine the impact of prime-age adult mortality on labor supply. With the HIV/AIDS epidemic striking most men and women at their prime age, an investigation of this phenomenon remains a necessity in developing countries generally and in South Africa in particular. Some studies focus on this impact in countries where the labor supply is largely within the agricultural sector. In contrast, very few studies examine this impact within the nonagricultural sector. The key focus of this chapter is the examination of changes in the time allocation of adolescents who experience prime-age adult death or illness in their households. In order to smooth household consumption and income, adolescents exit school and enter the labor market (whether employed or unemployed), and female household members alter their labor force participation.A few studies empirically examine the impact of AIDS mortality on labor supply. An analysis of two LFSs (1990-91 and2000-01) in Tanzania undertaken by Wobst and Arndt (2004) revealed an important trend in this regard; the data suggest a dramatic increase in labor force participation rates for children aged 10-14, from 23 percent to 46 percent, implying a tendency to exit primary schools over the decade. This chapter is similar to their paper in that I focus on adolescents' time allocation. However, I explore this issue using micro-longitudinal data instead of the aggregate measures that are used in their analysis.There are two studies on this topic from dominantly agrarian settings: Beegle (2005) and Donovan et al. (2003). Using panel data from Tanzania, Beegle (2005) explores how prime-age adult mortality affects the time allocation of surviving household members and the portfolio of household farming activities. The author analyzes hours spent farming and on household chores across demographic groups and finds small and insignificant changes in the labor supply of individuals in households that experience a prime-age adult death. While some farm activities are temporarily scaled back and wage employment falls after a male death, households did not shift cultivation away from subsistence farming. Moreover, they did not appear to reduce their diversification of income sources more than six months after a death. Donovan et al. (2003) analyze the effects of prime-age adult morbidity and mortality in rural Rwanda using recent data from household surveys with overlapping samples and retrospective information on deaths within the household. They show that loss of agricultural labor was most pronounced for tasks involving cropping and animal husbandry relative to other potential effects (for example, diet effects, effects on children, and other incomegenerating effects).Our empirical setting is different from that of these two studies in that most of the sample households supply labor in nonagricultural sectors. That is, intrahousehold decisions on labor allocation for self-employment, such as own farming, are not the issue. With a focus on trade-offs between human capital investments and labor supply, I analyze labor force participation decisions among adolescents without distinguishing between the agricultural and nonagricultural sectors.It is widely believed that the death of a parent and prime-age household member has serious repercussions on the well-being and future of childrentheir schooling in particular. However, the literature on this topic does not necessarily present similar results, partly because of differences in methodologies across studies.First, the literature is divided into two groups of studies: those using panel data and those based on cross-sectional data. The studies that used panel data can potentially identify ex ante and ex post effects of prime-age adult mortality on child schooling.Second, the focus of studies may be parental deaths (leading to orphanhood) or prime-age adult deaths in general. Since the two are not exclusive, studies could basically analyze both impacts, but this portion of the literature is generally divided into two groups, with relative emphasis on either parental deaths or prime-age deaths. This chapter sheds light on both issues using panel data.Approaches and findings from the literature are summarized in Table 5.2. Five studies use panel data: Ainsworth, Beegle, and Koda (2005); Yamano and Jayne (2005); Beagle, De Weerdt, and Dercon (2006); Case and Ardington (2006); and Evans and Miguel (2007). All of these studies, except Beegle, De Weerdt, and Dercon (2006), used standard methods of panel analysis with children in similar age ranges. Beegle, De Weerdt, and Dercon (2006) used children aged 0-15 who had not experienced parental deaths in the baseline to track their completed years of schooling and anthropometric measures later in their lives, under the assumption that parental deaths were not predictable and also not correlated with the health and schooling of children in the initial round. In this respect, although they used panel data, it is not straightforward to compare their results with others.As Table 5.2 shows, with the exception of Yamano and Jayne (2005), all studies analyzed the effects of parental deaths. However, Ainsworth, Beegle, and Koda (2005) and Yamano and Jayne (2005) are the only studies that directly  analyzed the effects of prime-age adult deaths. In all studies, both ex ante and ex post impacts are studied.Of the two studies that highlight the effects of prime-age adult mortality, both report that the ex ante effect of prime-age adult deaths-hours at school in Ainsworth, Beegle, and Koda (2005) and school attendance in Yamano and Jayne (2005)-is significantly negative. As discussed subsequently, our results also show significant ex ante effects on adolescents' transition from school to labor markets (that is, changes in enrollment status).The other studies which focus on parental deaths (Case and Ardington 2006;Evans and Miguel 2007) show that a mother's death has a more significant and strongly negative impact on child schooling than a father's death. This chapter, however, shows that both mothers' and fathers' deaths have almost equal impacts on adolescents' decisions regarding the continuation of schooling.One caveat in these comparisons arises from differences in the child age range between the present study and the other studies in the literature (see the third column in Table 5.2). In our sample from South Africa, enrollment status does not show any cross-sectional variations among children below age 13 at the primary-school level, partly because of the tendency of children to remain enrolled even if they participate in some other activities. In this sense, more sophisticated measures of schooling, such as attendance, should be used, but I decided not to pursue this line of investigation since the 1998 data do not include such information.Three studies used cross-sectional data: Case, Paxson, and Ableidinger (2004), Ainsworth and Filmer (2006), and Ardington and Leibbrandt (2010). Case, Paxson, and Ableidinger (2004) and Ardington and Leibbrandt (2010) used within-household variations to address the relative status of orphans who lost their parent(s). The former study shows that orphans' enrollment is lower than that of non-orphans within the same household, though residence arrangement is an endogenous choice. If orphans become better off by moving into other households (for example, those of relatives who have not experienced parental death) after losing their parent(s), their status is overestimated in this method -which, however, only strengthens the findings. Using within-household variations, Ardington and Leibbrandt (2010) found that maternal death has a larger effect on years of schooling and enrollment than paternal death-a finding in line with those from other studies using panel data. Ainsworth and Filmer's (2006) results are not clear. Using data from 51 countries, they report that orphans' enrollment status is not significantly lower than that of non-orphans in many countries-but as they acknowledge, this could be attributed to the failure in their methodology to account for socioeconomic factors that determine orphan status.Our approach belongs to the first group of studies, those using panel data. Owing to data constraints (explained earlier), I focus on the enrollment status of adolescents aged 14-19 in the baseline and their transitions to labor markets thereafter. Though our results in general conform to those in the literature, one difference is the finding of almost equal magnitudes of ex post effects of maternal and paternal deaths, partly as a result of difference in the age ranges used. Because of small incidences of parental deaths in a short period, it was not possible to estimate the ex ante effects.This section describes our empirical methodology. The behavioral equation of interest to us is as follows:where, for individual i in household j at time t, our observable activity indicator y ijt = 1 (enrolled in school or engaged in housework) if latent variable y* ijt ≥ 0 and y ijt = 0 (in labor market) otherwise, x jt is a vector of householdlevel factors including the demographic composition of the household, s it0 is the highest grade completed in the initial period t 0 , a it0 is age at the initial period t 0 , μ ij is individual i's fixed effect in household j, and ε ijt follows the standard normal cumulative distribution function (probit) or the logistic cumulative distribution function (logit). As I discuss subsequently, I use conditional logit estimation to eliminate time-invariant factors in the analysis of dynamic activity transition. 4  Although schooling level and age in the initial period are controlled for in equation (5.1), these time-invariant variables in addition to unobserved fixed effects do not contribute to the estimation of the transition equation. Therefore, in the estimations carried out next I include initial age and schooling fixed effects only when estimating the prime-age adult mortality effect in crosssectional models to control for potential unobserved heterogeneity of labor supply behavior attributable to predetermined schooling attainment and age. 5 100 CHAPTER 54 Although multinomial logit estimation is also applicable in this context to capture participation in (1) school (adolescents) or housework (adult females), (2) the labor market, and (3) other activities, I decided to focus on (1) and ( 2) since the number of observations for (3) is very small in our sample of individuals in specific age ranges. The inclusion of observations for other activities destabilizes the parameter estimates in maximum likelihood estimation. 5 Including both initial schooling and age fixed effects indirectly controls for the effect of past grade repetitions on the labor supply.Correlations between x jt and μ i bias estimates of β. To solve this problem, I focus on changes in activity (for example, transitions from school to the labor market) over time. More specifically, I adopt the conditional (fixedeffect) logit model to eliminate μ ij (Chamberlain 1984). 6 The process involves first differencing the right-hand-side variables, which produces Δx j . This term captures changes in households' demographic composition; in our setting it represents the death of prime-age adult members.The endogeneity of adult mortality is not perfectly resolved, as we may Beegle 2005 for a detailed discussion of potential endogeneity problems of adult deaths.) To illustrate the problem, we can take the example of a drop in household income that is caused by a death and which causes dropouts from school in a context of credit constraints. The income drop at the initial stage could worsen subsequent living conditions. The worsened living conditions can, in turn, contribute to adult mortality. Therefore, the income drop can cause adult mortality and child dropouts from school.In the case of HIV/AIDS-related mortality, it is possible that foreseeing the death of AIDS-infected family members causes other members to adjust to the negative impacts ex ante, for instance by increasing the labor supply to secure the household's income. The question is who in the household changes his or her time allocation along with the shocks. For example, if girls are more likely than boys to care for the sick at home, girls' enrollment rate should be lower than that of boys even before the death of the ill family member(s).7 I use a sample of individuals aged 14-19 in 1998 irrespective of their enrollment status to check for ex ante responses.Although it is not possible in this data to identify whether a transition out of school is temporary, most adolescent transitions are a one-way process in our empirical setting. Though those who were recorded as participating in labor-force activities do not move back to school after the six-year interval covered by our panel data, the fact that I ignore the heterogeneity in the initial state causes a selectivity bias in the estimates. Therefore, I do not restrict the sample to those who were enrolled in school but also incorporate those already in the labor market in 1998. I also investigate how future mortality affects the initial state. If household members anticipate the death, in the future, of a household member, they may adjust their behavior ex ante to accommodate for the negative shocks that arise ex post. 8 Finally, in the analysis of adult labor supply, I focus on women's time allocation between in-house and labor-market activities.The KIDS surveys provide information about the activities of household members by employment status; labor force status consists of three categories of employment (regular employment, casual or temporary employment, selfemployment) as well as unemployment. The KIDS data also specify whether an individual is unemployed, a \"housewife / involved with child care,\" in school (including at a university), at a crèche or preprimary school, retired, or \"other.\" The specifications are obtained from the respondents themselves. Although it is not possible to verify in the dataset whether the respondents are actively seeking employment, school enrollment and unemployment are mutually exclusive among the questionnaire options. Combining the two survey rounds, it is possible to establish the transitions in activity from 1998 to 2004, although details of activities within the intervening period are not available.Table 5.3 shows the activity transitions from 1998 to 2004 among adolescents, that is, for those aged 14-19 in 1998.9 Two main trends emerge from the data. First, it is interesting to find that among those who transitioned to the labor market, the majority were unemployed in 2004. As discussed in  Wisconsin-Madison (1998, 2004).Note: Ages are as of 1998.Chapter 1, this tendency reflects current labor-market conditions in South Africa, where the unemployment rate is high, particularly among the young. It is important to note that making the transition from school to the labor markets does not necessarily mean that individuals are employed in those markets.In the analysis that follows, I examine adolescents' transitions from education to the labor force, including both employment and unemployment. Second, the transition from school to the labor force occurs in a relatively similar manner for both men and women. In South Africa gender gaps in schooling investment and in labor markets are relatively unimportant.The 1998 data contain seven adolescents whose activities were not classified as either labor market or school. Since the number of observations in this group is small, I decided not to show them in the table. The analysis in the next section also excludes this group, as inclusion of them in a multinomial logit analysis makes the estimation unstable. In the next section I provide empirical results on the impact of prime-age adult death on adolescents' decisions to continue with schooling or to enter the labor market.I use prime-age adult mortality as a measure of AIDS-induced mortality shocks to households in the analysis of adolescents' transitions from school to labor markets. An alternative method is to use parental death. These choices are not exclusive, but they could shed light on different aspects of AIDS-related mortality.In our sample, I had 408 cases of prime-age adult mortality (defined for ages 15-64). Among the children aged 14-19 who were residing in the sample households in 1998, 43 children lost their mothers during the interval 1998-2004, and 60 lost their fathers. 10 The number of children who had already lost their parent(s) before 1998 is not included in these figures, so the number of orphans is greater. A total of 362 children experienced prime-age adult mortality in their households. 11In this chapter, though I focus on the impacts of prime-age adult mortality, it is also interesting to examine the impacts of parental deaths and compare the results. As discussed in the previous section, the literature seems to confirm that a mother's death has a significantly greater adverse impact on child schooling than a father's death. This finding does not contradict the hypothesis that the death of a prime-age adult (especially one who is working, and thus contributing to household income) has a negative effect on child schooling. In fact I also confirm in this study that a mother's death has an equal and significant negative impact on child schooling for boys and girls.In this section I detail the main findings on the impact of prime-age adult mortality on the time allocation of adolescents. In the analysis that follows, I use two different definitions of prime-age adults: those between the ages of 20 and 64, and those between the ages of 20 and 44. The former corresponds mainly to the age range for labor force participation, while the latter focuses on the age range in which, as shown in the section \"Mortality Change,\" I found substantial increases in mortality rate.12 I consider a variety of measures of prime-age adult mortality-the total number of deaths in the household, prime-age adult deaths (ages 20-64 or 20-44), and deaths of working members and of prime-age working adult members, differentiated by gender and occurring between 1998 and 2004-as explanatory variables. For the data, I do not capture deaths of members who moved out of the households. 13 Instead, I include occurrences of death of new members who moved into the households, which have a relatively smaller effect on time allocation decisions. Mortality estimates could accordingly be biased in either direction. 14Table 5.4 shows how prime-age adult mortality in the very near future, 1998-2001, affects school enrollment in 1998, that is, ex ante. If future mortality among prime-age adults in the household is preceded by the need for home care, additional incomes, or both, it may change the time allocation of adolescents. 15 Since this analysis is cross-sectional, I include initial grade and cluster (community) dummies and age to control observable heterogeneity, possibly determining schooling progression and transition to the labor market. Since in our sample all children under age 13 were enrolled in schools, I do not include those children.Column 1 includes deaths of prime-age (20-64) members and prime-age working members. In the preliminary analysis, I included mother's and father's deaths in 1998-2001, but the number of observations is so few that it was not feasible to estimate their effects. Adding both values to arrive at total parental deaths did not help estimate their effects.The prime-age (20-64) deaths on their own do not affect enrollment significantly, but those of working members significantly decrease enrollment.  -4, 5-9, 10-14, 15-19, 20-24, 25-29, 30-34, 35-39, 40-44, 45-49, 50-54, 55-59, 60-64, and 65 years or above who have lived more than 15 days in the household in the previous month; and both grade (1998) and cluster fixed effects. Except for cluster fixed effects, these variables are interacted with a male dummy. The specifications include parental deaths in 1998-2001, but the effect was not estimable because of the small incidences of parental deaths in the three-year period 1998-2001.Given a significant ex ante negative effect on enrollment of the future death of a working adult of prime age, I can conclude that male adolescents' labor supply is likely to increase when the household income drops as a result of the death of a working adult. Column 2 adds the interaction of working prime-age deaths at ages 20-64 and a male adolescent dummy. However, I do not find any significant effects in this specification.In column 3 I use the prime-age definition of 20-44 years to account for the fact that this group has experienced significant excess mortality (see Table 5.1 and Figures 5.1 and 5.2). This measure is highly correlated with the death of working prime-age (20-64) members. The results in columns 1 and 2 indicate that the deaths of working prime-age adults active at ages 20-44, which cause a reduction of household income, have larger and more significant impacts on adolescents' transition to labor markets.Column 4 disaggregates the prime-age (20-44) death effects by adolescents' gender. First, though the prime-age (20-44) adult deaths did not affect enrollment in general, we find that they significantly decreased males' enrollment. However, the effect is significantly negative for male adolescents. This finding suggests that male adolescents assume greater responsibility for supporting their households by moving from school to labor markets once a prime-age adult dies.In column 5 I consider deaths of male and female prime-age (20-44) household members separately. Interestingly, the effect of female deaths is significantly negative while that of male deaths is insignificant. I further decompose the effect by distinguishing between male and female adolescents (column 6). The deaths of male prime-age members significantly decrease male adolescents' schooling, but not that of female adolescents. However, the deaths of female prime-age members decrease both male and female adolescents' schooling (thus promoting their transition to labor markets).Since the analysis is cross-sectional, we cannot distinguish the preceding interpretation from the case in which these females live in households with certain characteristics that are more likely to result in an adult death (that is, poor households). To account for this, I included the value of durable goods (assets) in 1998 in the regressions.In this analysis I also tried to expand the sample by including younger children (age below 13) as part of the robustness checks, but since almost all children in this age range were enrolled in schools, this change does not improve the precision of the estimates. 16Case, Paxson, and Ableidinger ( 2004) and Case and Ardington (2006) indicate that male death is likely to occur in poor households. If so, large impacts of the death of working males have to be interpreted carefully. It is still possible that we pick up the effect of poverty on school dropouts.One important cause of girls' decisions to drop out of school in South Africa is pregnancy. The question then arises of whether the probability of becoming pregnant during schooling is correlated with prime-age adult mortality in the household. For example, it is possible that poor families are more likely to have pregnant teenagers and face higher prime-age adult deaths than wealthy families. Again this factor was controlled for by including the value of durable goods in the household. However, it is possible for some households to have members with high biological fecundity (propensity to reproduce), which increases both the likelihood of teenage pregnancy and adult sexual activity (including extramarital sex), possibly leading to a higher incidence of HIV/ AIDS-related deaths. Another potential problem arises from a positive correlation between a pregnancy shock (implying a negative shock to enrollment) and a prime-age adult mortality shock in the household. In the preceding cross-sectional analysis, data limitations make it hard to control the resulting potential downward bias. I was thus forced to set this problem aside.Next I analyze the transition of adolescents from school to the labor force using panel data for 1998-2004. Table 5.5 shows the relationship between age in 1998 and activity in 2004-whether or not an individual is in school (employed and unemployed). Interestingly, the transition to the labor market starts above ages 12-14 in 1998. In the preliminary analysis, I used a sample aged 7-13, but it was found that children aged 7-12 (in 1998) did not contribute to the estimation owing to a very small number of transitions. Therefore-as long as enrollment is used as a measure of schooling status, as opposed to more sophisticated measures, such as attendance, grade progression, and test scores-it is not feasible to identify the transition from school to nonschool status using younger children. Therefore I focus on adolescents (defined as those aged 14-19) in this study.Table 5.6 gives the estimation results for the conditional (fixed-effect) logit model. 17 In this analysis I ignore other activities as an endogenous choice variable since the number of observations in this group is only seven for 1998 (rendering the estimation unstable). In all specifications, I also include dummies for maternal and paternal deaths. 18  Column 1 includes the number of deaths among prime-age (20-64) household members, maternal deaths, and paternal deaths during the period 1998-2004. First, deaths of prime-age members significantly increase adolescents' transitions from school to labor markets. Second, deaths of a mother and a father both significantly increase such transitions.Column 2 includes the number of deaths among working prime-age household members during the period 1998-2004. Since the death of working primeage members is a subset of prime-age deaths, we look at the additional effect that the death of working members may have. There is no additional effect on adolescents' transition from school to labor markets.In column 3 I interact the male adolescent indicator with the aboveprime-age adult mortality measures to check for gender differences in transition into the labor force. The column shows that an increase in the total number of working prime-age household members promotes a shift out of school into the labor force among adolescent males. The effect of prime-age deaths is significant without showing a gender difference. Mother's death and father's death continue to be highly significant. Columns 4 and 5 add the number of prime-age deaths using the age definition 20-44 years. The results show that while the death of prime-age members using ages 20-64 is insignificant, deaths among members aged 20-44 significantly increases the transition of adolescents from school to labor markets. Interestingly, male adolescents are more likely than females to stop attending school when more economically active household members die (that is, those aged 20-44 and/or working and of prime age).  In columns 6 and 7 I decompose prime-age deaths by gender. Both male and female prime-age deaths significantly increase the transition into the labor force. The effect is slightly larger with male prime-age deaths. The death of male prime-age members has a statistically larger effect on male adolescents. In summary, our results imply that it is more likely for male adolescents to enter the labor force when economically more active household members die-that is, males classified as of prime age (20-64), of prime age (20-44), or working and of prime age. This means that there is a substitution between male adolescents and economically active household members in their potential contribution to household income. Second, deaths of mothers and female prime-age household members affect both genders equally. This finding is consistent with the hypothesis posed earlier.Third, in all specifications, I found that mothers' and fathers' deaths both significantly increase adolescents' entry into the labor markets. Quantitatively the impacts are much larger than that of prime-age adult deaths. The findings are largely consistent with evidence provided in the literature, but these results show that the effects of mothers' and fathers' deaths are nearly equal, while some previous studies report a larger effect of mothers' deaths.The results in Tables 5.4 and 5.6 support the hypothesis that excess prime-age adult mortality in a society decreases the investment in human capital in the young population. I find decreased male and female schooling (increasing their entrance into the labor markets) before and after the death of prime-age adults and, more importantly, the death of parent(s).My empirical results show that prime-age adult mortality has had an impact on adolescents' transition from school to labor markets and on female labor supply decisions during the period 1998-2004. This six-year period coincided with an increase in prime-age adult mortality in South Africa.A few clear findings emerged from this research. The death of a primeage adult in the household, parent(s), or both significantly promotes adolescent transition to the labor markets. Adolescents leave school in order to compensate for a possible loss of or reduction in household income, to smooth consumption, or to look after the household and its members. Male adolescent labor supply increases more when the death of working adults is anticipated in the very near future.These results suggest that excess mortality among prime-age adult causes a reduction of human capital investments at school as well as at home. Thus the AIDS epidemic has negative effects on the formation of human capital for the next generation, not only by disrupting child schooling, as many studies have shown, but also by discouraging the continuation of schooling investment among adolescents.Generally, in light of constrained household income owing to medical and funeral costs and limited access to formal arrangements (for example, credit) and to foster-care grants (see Meintjes et al. 2003 for South Africa), the wellbeing of surviving household members, including orphans, remains a challenge. As the economy confronts an increased number of young people experiencing prime-age adult mortality in their households, this challenge will need to be addressed by the South African government and the society as a whole.History versus Expectations T he formation of human capital in the majority population of South Africa promises the alleviation of poverty and equitable economic growth. With the burden of its apartheid legacy, the country faces enormous challenges in transforming itself into an equitable society. Given the political difficulty of redistributing assets from the historically advantaged minority to the disadvantaged majority, the formation of human capital in the latter group can be a long-term solution to these challenges, as it builds new assets the poor can own and direct. However, the transformation of a society through the broad-based formation of human capital, especially among Africans, requires a relatively long time span.This monograph delivers two basic messages. First, though education is at the center of human capital formation, the dynamic aspects of human capital formation in individuals call for a wider perspective. Part 2 demonstrated that human capital formation is dynamic in the sense that it starts in early childhood-even before childbirth. Malnutrition during early childhood owing to poverty reduces achievements at the schooling stage. Policy interventions aimed at enhancing healthy child growth (such as the current child support grants) are needed to guarantee the outcomes expected from the government's effort to develop public school education. In South Africa the HIV/AIDS epidemic adversely affects schooling progress among adolescents, increasing their entry into the labor markets. Therefore, it is important to support households that are vulnerable to adverse shocks and protect children from the negative impacts of such shocks.Second, as Part 1 showed, access to opportunities is still unequal across different segments of the society. Given the historical path the country has taken, it is important to guarantee equal opportunities for the formation of human capital to the African majority. In particular, relaxing financial constraints in formerly African schools and households is an urgent priority.Chapter 1 posed two questions regarding returns to schooling in South Africa: Why do returns to schooling differ across population groups? And why do those returns show convexity-low returns up to high school, with significantly positive returns at the completion of high school or at higher levels?It appears to me that an answer to the first question is related to differences in the quality of education available to different groups of the society. Part 1 showed that local resource constraints, reflected in school fees, potentially limit the quality of education available in certain communities, though I also confirmed that government subsidies had relaxed those constraints. In this sense, it is interesting to observe that recent returns to schooling have been higher among the Indian/Asian population than among others. The labor markets still demonstrate whites' advantage in accessing jobs with higher wages.The convexity question also seems to be related to the quality of education as well as the high unemployment rate in the South African labor markets. It is possible that labor demands that are biased in favor of a skilled labor force create rigidity in the demand-supply mismatch, concentrating unemployment among the less educated. The relatively low pass rate in matriculation at grade 12 exacerbates this problem.As our finding in Chapter 3 suggests, however, the progressive allocation of government budgets to schools, supported by the Norms and Standards for School Funding, is expected to improve the situation faced by historically disadvantaged schools and communities. educators is the same in this framework, the two budget constraints can be added together. Assume that the government decides the per-learner amount of subsidy, g i , so G i = g i L i and g i ≥ 0 for all i (the government does not impose a tax).The school fee is bounded by some limit, q i --(f), determined by the socioeconomic circumstances, f, of the school. In particular, the fee is determined by income level and distribution. 3 In fact, school fees are determined by SGBs consisting of educators and community leaders, such that most parents can afford to pay them. Unless the government subsidy g i offsets q i (f), local con-L dition f affects --.H School maximizes the per-learner education output subject to the budget constraint:x,q H i s.t.Rearranging the budget constraint, we define φ(q i (f); w, g i ):w L i φ(q i ; w, g i ) ≡ -------≤ --.(A.1)The LER is constrained below by the ratio of educator wage (per-educator cost) to the sum of the school fee and the per-learner subsidy (per-learner revenue). When the school decides on the school fee and employs educators, the determination of the school fee is simple: q i *(f) = q i (f), that is, collect the highest school fees. 4 In this model I assume that, at least in the short run, the number of learners does not change immediately in response to an increase in school fees (that is, inelastic enrollment), but also that the fees are determined such that most parents can afford to pay them.5 Suppose now that budget constraints are not binding. Then the optimal solution is L i x i * ≡ --= y* > φ(q - i (f), g i ; w). H i * 1 In this case H i = β*L i , where β* = --. Next consider the case in which the budget y* constraint is binding: y* < φ(q - i (f), g i ; w). In this case:q - i (f) + g i where w < β*(q - i (f) + g i ). The second term is an efficiency loss in terms of the number of educators. The government will allocate the subsidies to those with binding budget constraints. Next consider the government's allocation of school subsidies. Assume that the government maximizes the total educational output Σi e i l i subject to its budget constraint but does not allocate any subsidy to those schools that are able to attain optimal ratios: w max Σ [ 1 -(y* --------) 2 ] L i { g i } i i|y*<φ(q -i,0;w) q i (f) + g i s.t.i|y*<φ(q -i,0;w)Without the government budget constratint, the necessary condition is g i * = wy*q i --(f). In general we have 2[y* -φ i (g i )][-φ′(g i )] = λ, where λ is the Lagrangian multiplier. From this we also know that when q i --(f) decreases, g i increases to compensate for gaps in the capability of collecting school fees (community endowment). In other words, LERs are equalized under the benevolent government's unitary decision. So, without government intervention, LERs are determined by school-level liquidity (budget) constraints, provided that the best ratio is identical in all schools no matter to which racial group they belong. However, we expect that with active government interventions, the ratios will be equalized across all schools. In particular, the subsidy is allocated more to those schools in less favorable socioeconomic circumstances, that is, those with larger initial LERs.Sections that are relevant to this monograph are the following: 45. The SASA [South African School Act] imposes a responsibility on all public school governing bodies to do their utmost to improve the quality of education in their schools by raising additional resources to supplement those which the state provides from public funds (section 36). All parents, but particularly those who are less poor or who have good incomes, are thereby encouraged to increase their own direct financial and other contributions to the quality of their children's education in public schools. The act does not interfere unreasonably with parents' discretion under the law as to how to spend their own resources on their children's education. 46. Ironically, given the emphasis on redress and equity, the funding provisions of the Act appear to have worked thus far to the advantage of public schools patronized by middle-class and wealthy parents. The apartheid regime favored such communities with high-quality facilities, equipment and resources. Vigorous fund-raising by parent bodies, including commercial sponsorships and fee income, have enabled many such schools to add to their facilities, equipment and learning resources, and expand their range of cultural and sporting activities. Since 1995, when such schools have been required to down-size their staff establishments, many have been able to recruit additional staff on governing body contracts, paid from the school fund.47. Poor people, on the other hand, especially in former homeland areas, have contributed a disproportionate share of their incomes over many decades to their building, upkeep and improvement of schools, through school funds and other contributions, including physical labor. All too many schools in poor rural and urban working-class communities still suffer the legacy of large classes, deplorable physical conditions, and absence of learning resources, despite a major RDP National School Building Programme, and many other projects paid directly from provincial budgets. Yet the educators and learners in poor schools are expected to achieve the same levels of learning and teaching as their compatriots.48. Such contractions within the same public school system reflect past discriminatory investment in schooling, and vast current disparities in the personal income of parents. The present document addresses these inequalities by establishing a sharply progressive state funding policy for ordinary public schools, which favors poor communities.This section introduces a simple model in which parents decide how much to invest in child health and schooling, resulting in returns in the labor market. For simplicity, we treat the age distribution of children as exogenous and assume that children enter the labor market in the final stage. Health is formed in the first stage, while schooling investment is undertaken in the next stage. 6In the pre-primary-school stage, per capita consumption determines health capital h j for child j, h j = f(c 1 , x) + ε j1 , where c 1 is per capita consumption in the household; x is predetermined community and household characteristics, such as the availability of healthcare facilities and personnel and parents' schooling; and ε j1 is an idiosyncratic health shock. For simplicity, health capital accumulates only until age a*, when a child enters the schooling stage. The investment component f(c 1 , x) is characterized by the properties ∂f ∂ 2 f ---> 0 and -----> 0. For simplicity, I assume that c 1 = y. Given that h is child ∂c 1 ∂c 1 ∂x height, c is specifically intended to capture nutritional intake.At the second stage, knowledge capital k j accumulates with schooling investments s j . The knowledge production function is given as k j = g(s j , h j , x) + ε j2 , where investment g(s j2 , h j , x) has health capital as its argument. Complemen-∂ 2 g tarity between schooling and health investments is captured by ---> 0. An ∂s∂h implication of such complementarity (or substitutability) is that parents want to observe the outcome of health capital among their children in order to optimally decide schooling investments in the children. Owing to the sequential nature of human capital investment, parents can learn about the total outcomes of child human capital and their labor-market returns from the outcomes of early-stage nutrition and health investments. 7  The household budget constraint in the second stage is c 2 + p Σ s j = y + Σ w(h j )[T -s j ] + b, j jwhere w(h j ) is the opportunity cost of schooling (returns to health capital), T is the time endowment for the child, p is the school fee, b is savings and loans, and y is exogenous household income. It is assumed that the opportunity costs increase with health capital, that is, w′(h) ≥ 0. 8 Assume that the child cannot work at the pre-primary-school stage, and can work in the labor market only when he or she enters school. 9 human capital. They focus on cognitive and noncognitive development. Their analysis does not directly include health and nutritional status as part of human capital in child development.The exclusion of health capital from the analysis results in a framework in which they can focus on human capital production function and the complementarity and substitutability of different inputs (for example, at the early-childhood and schooling stages). In this appendix, children are also considered as working in the labor market or participating in other activities where health capital has economic returns. This institutional setting creates implications that offset the health-schooling complementarity effect. 8 It is also important to note that the income opportunity in the child wage w(h) is not necessarily related to labor markets. It may also capture activities such as child care and self-employment. 9 Several qualifications follow. First, I assume that income from siblings, parents, and credit is pooled in the household budget constraint and is therefore perfectly substitutable. Second, to describe the income process, the model does not assume a production function in which adult and child members supply labor inputs that are not perfectly substitutable. This assumption is suitable in our empirical setting of South Africa, where wage employment (including formal and informal jobs) is a major source of income. Third, the utility function does not include leisure, which is imperfectly substitutable between household members (for example, Pitt and Rosenzweig 1990).Parents maximize the objective function max u(c 2 ) + βE Σ V(W(k j , h j ) -(1 + r)b j ), { w(h j ) + p } u′(c 2 *) = βV′(c j ) ---(s j , h j , x)R k ′, ∂s j where λ* denotes the Lagrange multiplier associated with the stage 2 budget constraint. These conditions provide the schooling function k*(y, h j , x). At the first stage, the problem is trivial, since exogenous income and shocks determine investment in health capital. Therefore, with a perfect loan market, the effect of health on schooling depends on ∂s j * ∂ 2 g(s j , h j ) ---> 0 ⇔ ---------R k > w′(h j )(1 + r). ∂h j ∂ s j ∂h jPreference and household income do not enter the condition. In this case, parents compare returns and opportunity costs for schooling. The important point is that child health capital can alter both returns and costs. If the opportunity cost does not increase with health capital (that is, w′(h) = 0), an increase in health capital will raise the optimal level of schooling if health and knowledge capital are complementary.In short, better health increases schooling, but it may also increase the probability that children are taken out of school to work either at home or in the market.Next consider the case in which b = 0, where credit opportunity is closed. Given the second-order condition, the effect of h j on s j depends on T he end of the apartheid system in South Africa in 1994 brought with it the end of legally sanctioned racial segregation in schools. In practice, however, racial disparities in educational attainment continue, with white and Asian students outperforming the country's African majority in school. Such a gap in achievement has a significant impact on the ability of Africans to get jobs and escape poverty. Human Capital Formation: History, Expectations, and Challenges in South Africa investigates the causes of South Africa's persistent schooling imbalances, examining education laws and policies, as well as other influences on human capital investment. The study finds that inaccessibility of quality education, resulting from a lack of financial resources at both the local and household levels, is currently a significant constraint on educational attainment among the poor. This limitation will likely relax in the future as the government continues to subsidize schools, but the study also concludes that educational disparities cannot be overcome by direct attention to schools alone. For example, children require adequate nutrition at the pre-school stage in order to perform well in school. Furthermore, parental death or illness resulting from the HIV/AIDS epidemic can disrupt the education of adolescents, who might need to leave school to care for their parents or support their family. Steps to compensate for these problems, as well as improved access to schools, are necessary. These findings clarify the problems underlying inequalities in educational attainment, offering guidance to policymakers, development specialists, and others concerned with South Africa's welfare.","tokenCount":"28689","images":["-864998923_1_5.png","-864998923_150_1.png"],"tables":["-864998923_1_1.json","-864998923_2_1.json","-864998923_3_1.json","-864998923_4_1.json","-864998923_5_1.json","-864998923_6_1.json","-864998923_7_1.json","-864998923_8_1.json","-864998923_9_1.json","-864998923_10_1.json","-864998923_11_1.json","-864998923_12_1.json","-864998923_13_1.json","-864998923_14_1.json","-864998923_15_1.json","-864998923_16_1.json","-864998923_17_1.json","-864998923_18_1.json","-864998923_19_1.json","-864998923_20_1.json","-864998923_21_1.json","-864998923_22_1.json","-864998923_23_1.json","-864998923_24_1.json","-864998923_25_1.json","-864998923_26_1.json","-864998923_27_1.json","-864998923_28_1.json","-864998923_29_1.json","-864998923_30_1.json","-864998923_31_1.json","-864998923_32_1.json","-864998923_33_1.json","-864998923_34_1.json","-864998923_35_1.json","-864998923_36_1.json","-864998923_37_1.json","-864998923_38_1.json","-864998923_39_1.json","-864998923_40_1.json","-864998923_41_1.json","-864998923_42_1.json","-864998923_43_1.json","-864998923_44_1.json","-864998923_45_1.json","-864998923_46_1.json","-864998923_47_1.json","-864998923_48_1.json","-864998923_49_1.json","-864998923_50_1.json","-864998923_51_1.json","-864998923_52_1.json","-864998923_53_1.json","-864998923_54_1.json","-864998923_55_1.json","-864998923_56_1.json","-864998923_57_1.json","-864998923_58_1.json","-864998923_59_1.json","-864998923_60_1.json","-864998923_61_1.json","-864998923_62_1.json","-864998923_63_1.json","-864998923_64_1.json","-864998923_65_1.json","-864998923_66_1.json","-864998923_67_1.json","-864998923_68_1.json","-864998923_69_1.json","-864998923_70_1.json","-864998923_71_1.json","-864998923_72_1.json","-864998923_73_1.json","-864998923_74_1.json","-864998923_75_1.json","-864998923_76_1.json","-864998923_77_1.json","-864998923_78_1.json","-864998923_79_1.json","-864998923_80_1.json","-864998923_81_1.json","-864998923_82_1.json","-864998923_83_1.json","-864998923_84_1.json","-864998923_85_1.json","-864998923_86_1.json","-864998923_87_1.json","-864998923_88_1.json","-864998923_89_1.json","-864998923_90_1.json","-864998923_91_1.json","-864998923_92_1.json","-864998923_93_1.json","-864998923_94_1.json","-864998923_95_1.json","-864998923_96_1.json","-864998923_97_1.json","-864998923_98_1.json","-864998923_99_1.json","-864998923_100_1.json","-864998923_101_1.json","-864998923_102_1.json","-864998923_103_1.json","-864998923_104_1.json","-864998923_105_1.json","-864998923_106_1.json","-864998923_107_1.json","-864998923_108_1.json","-864998923_109_1.json","-864998923_110_1.json","-864998923_111_1.json","-864998923_112_1.json","-864998923_113_1.json","-864998923_114_1.json","-864998923_115_1.json","-864998923_116_1.json","-864998923_117_1.json","-864998923_118_1.json","-864998923_119_1.json","-864998923_120_1.json","-864998923_121_1.json","-864998923_122_1.json","-864998923_123_1.json","-864998923_124_1.json","-864998923_125_1.json","-864998923_126_1.json","-864998923_127_1.json","-864998923_128_1.json","-864998923_129_1.json","-864998923_130_1.json","-864998923_131_1.json","-864998923_132_1.json","-864998923_133_1.json","-864998923_134_1.json","-864998923_135_1.json","-864998923_136_1.json","-864998923_137_1.json","-864998923_138_1.json","-864998923_139_1.json","-864998923_140_1.json","-864998923_141_1.json","-864998923_142_1.json","-864998923_143_1.json","-864998923_144_1.json","-864998923_145_1.json","-864998923_146_1.json","-864998923_147_1.json","-864998923_148_1.json","-864998923_149_1.json","-864998923_150_1.json"]}
\ No newline at end of file