Are Returns to Mothers' Human Capital Realized

Are Returns to Mothers’ Human Capital Realized in the Next Generation? The Impact of Mothers’ Intellectual Human Capital and Long-Run Nutritional Status on Child Human Capital in Guatemala

By Jere R. Behrman,a Alexis Murphy,b Agnes R. Quisumbing,b Usha Ramakrishnan,c and Kathyrn Yountc* a

Economics and Population Studies Center, University of Pennsylvania, Food Consumption and Nutrition Division, International Food Policy Research Institute, c Global Health, Emory University b

March 21, 2007

*The authors are listed alphabetically. This study was supported by NIH/Fogarty grant TW05598 on “Early Nutrition, Human Capital and Economic Productivity,” NSF/Economics grants SES 0136616 and SES 0211404 on “Collaborative Research: Nutritional Investments in Children, Adult Human Capital and Adult Productivities,” NIH grant HD046125 on “Education and Health over the Life Course in Guatemala,” R01 HD045627-01 grant on “Resource Flows Among Three Generations in Guatemala” NIH/NIA grant P30 AG12836 to PARC at the University of Pennsylvania, the Boettner Center for Pensions and Retirement Security at the University of Pennsylvania, and the Mellon Foundation grant to the Population Studies Center of the University of Pennsylvania. The authors thank their colleagues on the larger project of which this study is a part and Meng Wang for excellent research assistance in the preparation of the data for this paper. The corresponding author is Agnes R. Quisumbing [email protected].

Abstract There are many estimates in the literature of significant cross-sectional positive associations between maternal human capital – usually represented by schooling attainment and, in some cases, by height (as an indicator of long-run nutritional status) – and child human capital. But almost all of this literature ignores possible estimation biases that may result from maternal human capital being behaviorally determined in the presence of intergenerationally-correlated endowments. A small number of recent studies, primarily on more developed economies, report that treating maternal schooling attainment as behaviorally determined affects substantially (generally reduces) the estimated causal impact of maternal schooling on children’s outcomes. No prior studies consider what happens to estimates of the impact of maternal long-run nutritional status if this “biological” component of maternal human capital is treated as behaviorally determined. The contribution of this paper is to explore, for the first time, how estimates of the impact of maternal schooling attainment and maternal long-run nutritional status on child human capital are affected if both of these components of maternal human capital are treated as behaviorally determined, using an unusually rich longitudinal data set collected over 35 years in Guatemala. The estimates are provocative. They suggest that the common procedure of treating maternal human capital as exogenous may yield misleading coefficient estimates for the impacts of maternal human capital on children’s human capital in the Guatemalan context. For maternal schooling, results suggest that the OLS estimates may understate slightly the impact on child grades of schooling relative to the age-cohort mean: moreover, including maternal schooling as the only measure of maternal human capital appears to bias upward the impact of schooling because it proxies in part for mothers’ biological human capital. Prior studies also have used only schooling attainment to measure maternal intellectual human capital; our estimates suggest that other indicators of maternal intellectual human capital such as test scores yield somewhat different results than using schooling alone. For maternal height, the estimates suggest that, for all but one of the child outcomes considered, the OLS estimates understate, in some cases substantially, the causal impact of maternal long-run nutritional status on both anthropometric and schooling outcomes in children. The estimates imply, thus, not only that in a number of cases are the “standard” estimates likely to be misleading because maternal human capital is endogenous, but they are likely to understate the importance of long-run maternal nutritional status relative to maternal schooling attainment in determining child human capital.

ii

1. Introduction Investments in women’s human capital often are justified because of its presumed large positive causal effects on the next generation. Several influential papers and books argue that the effect of maternal human capital on child health and education is large and causal (e.g., Summers 1992, 1994; Stern 2001; World Bank 2001). Returns to investments in women through childbearing and childrearing could be realized through a number of pathways. First, better maternal nutritional status before and during pregnancy through “biological” pathways may lead to better nutrition in utero and higher birth weights of her children, which result in healthier offspring over their life cycles.1 Second, more maternal human capital may be associated with shifts from child quantity to child quality, in part because more maternal human capital raises the opportunity costs of women’s time for investing in child quantity versus child quality. Third, more maternal schooling may make women more aware of and more likely to seek information, and more likely to engage in behaviors that result in better-educated and healthier children. These behaviors could include those related to the nutrition and care of children, such as breastfeeding and proper diet, as well as behaviors that enhance the intellectual development and school performance of their children. Better health outcomes of children are also associated with the development of cognitive potential early in the children’s life cycle (e.g., Engle et al. 2006). Through all three of these pathways, improved maternal human capital is thought by many to result in improved health and schooling outcomes of children that in turn result in higher lifetime incomes of the next generation. Many studies, both for more and less developed countries, present positive associations between maternal human capital and child human capital that are consistent with the importance of these various pathways (Cleland and Van Ginneken, 1988; Brooks-Gunn, Klebanov, and Duncan, 1996). However, within a dynamic life-cycle framework, schooling and adult nutritional status reflect behavioural choices that depend on observed and unobserved individual and family background and other characteristics. Some individuals and some families may have unobserved characteristics, such as ability and motivation for schooling that are rewarded in labor markets, or better healthseeking behaviours and greater food availability, that lead to greater investments in schooling and better nutritional status. If the estimation methods used do not control for the behavioral determinants of schooling and adult nutritional status, the estimated associations of maternal schooling and nutritional status with outcomes in the child generation may largely be results of the impact of such unobserved factors on all of these behavioural outcomes. A small subset of articles, primarily recent studies for more developed countries, have investigated changes to estimates of the impact of maternal human capital – in particular, maternal schooling attainment – on child outcomes with controls for the behavioural determination of maternal human capital within a life-cycle framework that accounts for unobservables such as innate ability and health. For example, Behrman and Rosenzweig (2002, 2005) report that the positive association between mothers’ and child schooling in cross-sectional estimates disappears or possibly becomes negative once there is control for innate ability and marriage market matching using special data on identical twin mothers in the United States. One explanation for this shift is that increasing the 1

While height is our measure of mother’s “biological” human capital, it can also be determined by a range of genetic, behavioral, and biological antecedents. We use “biological” in our discussion, however, to distinguish this component of human capital from the measures of intellectual human capital that we discuss subsequently. We do not use the term “physical” capital because of its association with physical asset stocks in the economics literature.

1

schooling of women in this context induces a shift from time spent on child care to time spent in the labor market.2 Likewise, Plug and Vijverberg (2003) and Plug (2004) use data on adoptees in the United States to investigate the intergenerational schooling relation and find that the positive association between maternal schooling and their children’s schooling in cross-sectional estimates disappears if maternal schooling is treated as endogenous using adoption as an instrument. Similarly, Black, Devereux and Salvanes (2005) use phased-in changes in compulsory schooling in Norway as instruments for endogenizing parental schooling; despite strong cross-sectional associations between parental and child schooling, they find little evidence of causal effects of parental schooling, though the impact of mothers’ schooling on sons’ schooling remains significant. We are aware of only two articles for developing countries that investigate changes in the estimated impact of mothers’ schooling on child schooling or health when mothers’ health also is treated as endogenous. Two studies from about two decades ago use data on adult sisters in Nicaragua to control for the common genetic environment and parental family environment shared by siblings (Behrman and Wolfe 1987a, b). Both of these studies find that most of the association between maternal education and child human capital in the form of health and schooling in Nicaragua reflects family background and genetic endowments, not causal effects of maternal schooling per se. We are not aware of studies that investigate what happens if maternal long-run nutritional status is also treated as endogenous in relations that determine child human capital in a parallel fashion. However, two studies using data from Guatemala find that when one’s own long-run nutritional status is treated as endogenously determined the estimated impact changes considerably – in comparison with OLS estimates. A person’s long-run nutritional status appears insignificant rather than strongly positively significant in wage production function estimates (Behrman et al. 2005), and appears much stronger and significant in adult cognitive achievement production function estimates (Behrman et al. 2006a). These results suggest that the treatment of maternal long-run nutritional status as endogenous in intergenerational relations for child human capital may also change the estimated effect substantially in comparison with OLS estimates. Thus, there are large literatures for developing and more developed economies that indicate strong positive cross-sectional associations between maternal human capital – both schooling attainment and long-run nutritional status – and child human capital. But the vast majority of these studies do not control for the endogenous determination of maternal human capital with persistent intergenerationally-correlated unobservables such as genetic endowments and home environments. A few studies for more developed countries and a smaller number for developing countries find that the estimated impact of maternal schooling on child human capital lessens or disappears when special data are used to control for the endogeneity of maternal schooling. No such studies of which we are aware address the impact of maternal nutritional status on children’s human capital. An additional drawback of most of the existing literature is the incomplete characterization of mothers’ intellectual human capital. In most of these studies, schooling attainment is used as the single measure of intellectual human capital. Schooling attainment, however, is a measure only of grades completed, not necessarily of ability or cognitive skills. A relatively small literature (e.g. Bossiere, Knight and Sabot 1985) attempts to use direct measures of ability (such as Raven’s Research in sociology (e.g. Bianchi 2000, 2005) suggests that “what gives first” when educated women enter the labor market is domestic labor, time with spouse, and leisure time spent alone.

2

2

Progressive Matrices) to examine whether human capital is associated with greater productivity as opposed to merely being a signal of having invested in education or training, which presumably results in (though does not guarantee) higher productivity. In most of these studies, there are no controls for the possible endogeneity of cognitive abilities, and thus estimates of their impact on child outcomes may also be erroneous. In this study, we investigate the impact of various measures of mothers’ intellectual human capital and long-run nutritional status at the life-cycle stage in which initial parenting decisions about a number of child human capital outcomes tend to be made in the society under consideration. We use unusually rich data collected over 35 years in a particular developing country context, Guatemala. Specifically, we examine the impact of maternal schooling attainment, Raven’s test scores, readingcomprehension test scores, and math test scores and long-run nutritional status on children’s anthropometry at birth (birth weight, birth length), nutritional status at 36 months (z-scores for length-for-age, weight-for-age, and weight- for- length), and schooling (deviation of grades completed from the age-cohort mean).3 We advance beyond almost all of the prior literature because we treat all measures of maternal intellectual capital as behaviourally determined and also treat long-run nutritional status as behaviourally determined. We organize this study by presenting first a conceptual framework (Section 2), then the data (Section 3), then our alternative estimates for each of a number of outcomes (Section 4), and finally our conclusions (Section 5). 2. Conceptual Framework Our conceptual framework considers the life cycle to have a series of stages. One of those stages is adolescence-young adulthood, during which time (for the society under consideration) most individuals initiate first unions, parenting, and child rearing. For ease of exposition, let us assume that three generations of individuals participate in, or are affected by, decisions regarding human capital investments. Denote the oldest parental generation as G1, the mothers’ generation as G2, and the youngest children’s generation (children of mothers in the G2 generation) as G3. Women (G2 mothers) have a vector of human capital stocks (K) that includes intellectual capital and longrun nutritional status and that determines the results of their union formation in terms of spousal characteristics and their G3 children’s human capital. We use “intellectual capital” to describe a variety of measures related to cognitive achievement, including completed schooling attainment, nonverbal cognitive skills (as measured by Raven’s test scores), and cognitive achievement (as measured by reading and mathematics test scores). These measures are described in greater detail in Section 3. Let Y be a vector of (G3) child human capital outcomes – e.g., health, nutrition, and schooling of children. The basic interest of this study is to estimate how Y depends on G2 maternal intellectual capital and long-run nutritional status (K), measured at the ages at which first parenting decisions are made in the society under study. We posit that there is a linear approximation to what determines Y, given G2 maternal human capital stocks (K), predetermined observed individual 3

We originally included whether the child ever attended school among our outcomes, but since almost all children in this cohort have ever attended school (88.6%), there is very little variation in this variable.

3

characteristics (I) of the G3 child (such as whether a twin and gender), and unobserved inherent endowments (E0) that are correlated across generations (such as innate ability and health), and a vector of stochastic disturbance terms, one each for each different outcome (V): (1) Y = a0 + a1 K + a2 I + a3 E0 + V, where the ai are matrices of coefficients to be estimated. Note that this relation pertains to the total effects of maternal human capital on child human capital, some of which work through mate selection and others of which directly affect child human capital (conditional on mate characteristics). The questions that are posed in the introduction pertain to obtaining good (consistent) estimates of the coefficients of maternal intellectual capital and long-run nutrition in relation (1), which are components of K, for each of the components in Y. But estimation of relation (1) is a challenge because parenting behaviors that result, for example, in better nutritional status or schooling outcomes of children all reflect previous behavioral choices. As a result, OLS or similar estimates of relation (1) are likely be inconsistent because of the endogeneity of maternal human capital, particularly if there are intergenerationally-correlated endowments, such as genes. To deal with these possible estimation problems, we first assume that individuals (G2 women) and their parental (G1) families make investments in prior life-cycle stages that determine the components of K. These investments are made within a dynamic, reduced-form demand context, given initial conditions (including G1 parental family background F0, initial community prices and policies C0, genetic and other endowments E0, and the G2 mother’s individual characteristics I0 such as age) and changes that occurred over time such as those in markets, policies and other conditions ΔC (conditional on the G2 individual’s birth year and subsequent age): (2) K = K(F0, C0, E0, I0, ΔC, W), where Δ refers to observed changes or shocks from the initial conditions to critical ages for the determination of K, and W refers to unobserved idiosyncratic influences. This expression encapsulates the results of many decisions that the G1 parents, and then increasingly the G2 children themselves, make over the adolescent/young adult’s life cycle, given initial conditions and time-varying factors outside of the control of the family. The elements in relation (2) are in general vectors of individual and community opportunities and constraints to which the family responds. One example is the genetic endowments (E0), a vector that includes innate “ability” endowments related to learning and “physical” endowments related to physical growth.4

4

These various endowments may be significantly correlated but they need not be positively correlated. A recent study for the United States, for example, finds that the endowments related to schooling and earnings on one hand and those related to physical health on the other, are negatively correlated (Behrman and Rosenzweig 2004). If that were the case in this study, the failure to control for genetic endowments could result, for example, in an overestimate of the effect of maternal schooling on child schooling, but an underestimate of the effect of maternal nutritional status on child schooling.

4

Consideration of relations (1) and (2) illustrates the two important aspects of the problems of estimates in the literature of relation (1) that are described in the introduction. First, mothers’ human capital (intellectual capital and long-run nutritional status) is determined in part by genetic and other endowments (E0) that also are posited to have direct effects on the outcomes of interest in relation (1) directly or because there are intergenerational correlations between maternal and child endowments. To obtain consistent estimates of the impacts of maternal intellectual capital and long-run nutritional status on the child outcomes of interest, some combination of data and estimation method must be used to avoid the bias that otherwise would result from correlations between intellectual capital, long-run nutritional status, and the expanded compound error term in relation (1) that includes unobserved genetic and other endowments (E0) in addition to the idiosyncratic error (V). This problem can be addressed in principle by direct control for genetic and other endowments (E0), by random assignment of maternal human capital (K) or by instrumental variable (IV) methods in which the components of maternal human capital (K) in relation (1) are replaced by their predicted values from relation (2), which are not correlated with the unobserved E0 and therefore are not correlated with the compound disturbance in relation (1). In the present (and almost all) data sets, only the last of these is an option, which we will explore.5 Second, the components of K are determined in part by the same initial conditions (F0, C0, E0, I0) and some common observed community changes ΔC and unobserved influences W. As a consequence, maternal intellectual capital and long-run nutritional status are likely to be correlated, and so estimates that do not control for both of these components of K—as in most of the literature—are likely to be subject to omitted variable bias because of the failure to control for the other component of K. 3. Data The data demands are considerable for estimating the relations posited in Section 2. We use an unusually rich, longitudinal data set collected over a 35-year period with parenting histories, child health and schooling outcomes, alternative measures of own human capital, family background, and shocks from an experimental nutrition intervention as well as market and policy changes. We first provide a brief general description of the data6 and then focus on the particular variables that we use in the analysis. Section 3.1 General Description of the Data In the early and mid-1960s, protein deficiency was seen as the most important nutritional problem facing the poor in developing countries, and there was considerable interest in the possibility that this deficiency affected children’s ability to learn. The Institute of Nutrition for Central America and Panama (INCAP), based in Guatemala, became the locus of a series of preliminary studies on this subject in the latter half of the 1960s (see Habicht and Martorell 1992, Martorell, Habicht and

5

Within-sibling estimates could be used to control for the average family genetic endowments as in Behrman and Wolfe (1987a,b), but previous studies suggest that the individual-specific deviations from those family averages have important impacts on human capital investments (Behrman, Rosenzweig and Taubman 1994, 1996). 6 For more extensive discussion, see Grajeda et al. (2005), Maluccio et al. (2005), Maluccio, Murphy and Yount (2005), Martorell et al. (2005), Quisumbing et al. (2005), Ramakrishnan et al. (2005) and Stein et al. (2005).

5

Rivera 1995 and, especially, Read and Habicht 1992). These preliminary studies informed the development of a larger scale supplementation trial that began in 1969. The data used in this study are from that larger supplementation trial, collected for individuals who were 0-7 years old during 1969-77 in four villages in Eastern Guatemala. Three villages—San Juan, Conacaste, and Santo Domingo—are located in mountainous areas with shallow soils, whereas the agricultural potential of Espíritu Santo, located in a river valley, is somewhat higher. All four villages are located relatively close to the Atlantic Highway, connecting Guatemala City to Guatemala’s Caribbean coast—from 36km to 102km from Guatemala City. From January 1969 to February 1977, INCAP implemented a nutritional supplementation trial in these four villages, together with data collection on child growth and development.7 The data collection focused on all village children aged seven years or less and all pregnant and lactating women. Cohorts of newborns were included until September 1977. Data collection for individual children ceased when they reached seven years of age. The birth years of children included in the 1969-77 longitudinal data collection thus range from 1962 to 1977, so their ages ranged from 0 to 15 years when the intervention ended.8 Therefore, the length and timing of exposure to the nutritional interventions (described below) for particular children depended on their respective birth dates. For example, only children born after mid-1968 and before October 1974 were exposed to the nutritional intervention for all of the time they were from 6 to 36 months of age, which often is posited to be a critical time period for child growth in the nutritional literature (see Maluccio et al. 2006, Martorell, Habicht and Rivera 1995 and Martorell et al. 2005 and the references therein) The principal hypothesis underlying the intervention was that improved pre-school nutrition accelerates physical growth and mental development. To test this hypothesis, 300 villages were screened to identify those of appropriate size, compactness (so as to facilitate access to feeding stations, health centers and psychological testing sites, see below), ethnicity, diet, schooling levels, demographic characteristics, nutritional status and degree of physical isolation. From this screening, village pairs similar in these characteristics were determined: Conacaste and Santo Domingo (relatively populous villages) and San Juan and Espíritu Santo (relatively less populous villages). Two villages, Conacaste and San Juan, were randomly assigned to receive a high protein-energy drink, Atole as a dietary supplement. Atole contained Incaparina (a vegetable protein mixture developed by INCAP), dry skim milk, and sugar and had 163 kcal and 11.5 g of protein per 180 ml cup. This design reflected the prevailing view of the 1960's that protein was the critically limiting nutrient in most developing countries. Atole, the Guatemalan name for hot maize gruel, was pale gray-green and slightly gritty, but with a sweet taste. In designing the data collection, there was considerable concern that the social stimulation associated with attending feeding centers—such as the observation of children’s nutritional status, the monitoring of their intakes of Atole and so on—also might affect child nutritional outcomes, 7

The intervention began in the larger villages, Santo Domingo and Conacaste, in February 1969, and in the smaller villages, Espíritu Santo and San Juan, in May 1969. 8 Though the number of villages is small, the number of village-birth-year cohorts is 64, which provides a reasonable number of clusters for estimation of standard errors that account for clustering. See Maluccio et al. (2006) and Behrman et al. (2006).

6

thus confounding efforts to understand the impact of the supplement per se. To address this concern, in the villages of Santo Domingo and Espíritu Santo, an alternative drink, Fresco, was provided. Fresco was a cool, clear-colored, fruit-flavored drink. It contained no protein and only sufficient sugar and flavoring agents for palatability. It contained fewer calories per cup (59 kcal/180 ml) than Atole. Several micronutrients were added to the Atole and Fresco in amounts that achieved equal concentrations per unit volume. This addition was made to sharpen the contrast between the drinks to protein; the energy content differed, of course, but this was not recognized to be of central importance at the time. The nutritional supplements (i.e., Atole or Fresco) were distributed in supplementation centers and were available daily, on a voluntary basis, to all members of the community during times that were convenient to mothers and children but that did not interfere with usual meal times.9 For this study, with the differential “intent to treat” exposure to these nutritional supplements as first-stage instruments to estimate relation (2), a critical question is to what extent the intervention design resulted in differences in access to calories, proteins and other nutrients. In addressing this question, we can exploit the intensive nature of the survey and observational work associated with the intervention. Averaging over all children in the Atole villages (i.e., both those that consumed any supplement and those who never consumed any), children 0-12 months consumed approximately 40-60 kcal per day, children 12-24 months consumed 60-100 kcal daily and children 24-36 months consumed 100-120 kcal per day as supplement (Schroeder, Kaplowitz and Martorell, 1992, Figure 4). Children in the Fresco villages, in contrast, consumed virtually no Fresco between the ages of 0-24 months (averaging at most 20 kcal per day) with this figure rising to approximately 30 kcal daily by age 36 months (Schroeder, Kaplowitz and Martorell, 1992, Figure 4). Micronutrient intakes from the supplements were also larger for Atole than Fresco villages. This population has been studied intensively in the years since the original data collection, with particular emphasis on the impact of the nutritional intervention (Martorell et al. 2005 provide references to many of these studies). Multidisciplinary research teams conducted several follow-up studies on participants in the 1969-77 study as well as on their children. The first conducted in 1987-88 targeted the same individuals born between 1962 and 1977 who had been participants in the INCAP longitudinal study and were 11 to 26 years of age in 1988. It was feasible to include those who remained in the original study villages and only those migrants who moved to Guatemala City and to the provincial capital of the study area. This study focused on the impact of nutritional improvements in the critical period of gestation and the first three years of subsequent human capital formation as measured by body size, working capacity, maturation, intellectual functioning, school achievement, as well as family formation and occupation (Martorell et al., 1995). Of the 2392 individuals 0-15 years old in the original 1969-77 data collection, 224 had died by 1987. Of the 2168 surviving subjects, 1574 participated in the 1987-88 study, a coverage rate of approximately 73%. When disaggregated by gender, the rate of coverage was slightly higher for women (~76%) than for men (69%) (Rivera et al. 1993). Between 1991 and 1996, investigators conducted a surveillance of births (offspring of the original subjects) in the four study villages (outmigrants were not studied). In 1996, the study was expanded 9

A program of primary medical care was provided free of charge throughout the period of data collection. Periodic preventive health services, such as immunization and deworming campaigns, were conducted in all villages.

7

to include a surveillance of pregnancies and to carry out a longitudinal study of these offspring. Between 1996 and 1999 in what is called the Generational Effects study, all confirmed pregnancies in the four villages were identified and followed through intensive surveillance. Information was collected on these pregnancies, all children who were born during 1996–1999, and children who were born before the study’s launch and were less than 3 years of age by the onset of the study. All of these children were followed to age 3 years or study closeout, whichever came first. Data included birth weight and length, weight and length at periodic intervals, morbidity and healthseeking behaviors, breastfeeding, consumption of complementary foods, mothers’ functional competence and intellectual functioning, and mother-child interactions (Ramakrishnan et al. 1999). Between 1991 and 1999, 698 children of 392 women who were original study participants were measured at birth. From 1996-1999, 570 children of 363 original study participants were routinely followed for anthropometry measurements through age 3 years. Finally, a multidisciplinary team of investigators, including the authors of this paper, undertook follow-up data collection in 2002-4 on all participants in the 1969-77 data collection. In 2002-4, sample members ranged from 25 to 42 years of age. Figure 1 shows what happened to the 1162 women 0-15 years old in the original 1969-77 sample by the time of our 2002-4 data collection: 919 (79%) were alive and known to be living in Guatemala (10% had died, 6% had migrated abroad, 5% were not traceable). Of these 919, 521 lived in the original villages, 95 lived in nearby villages, 222 lived in or near to Guatemala City, and 81 lived elsewhere in Guatemala. For the 919 traceable sample members living in Guatemala, 649 (71%) finished the complete battery of applicable interviews and measurements and 818 (89%) completed at least one interview during the 2002-4 data collection (Grajeda et al. 2005). During each major study, a census of the four villages was conducted: in 1967, 1975, 1987, 1996, and 2002. These censuses contain valuable information on the family backgrounds of the original subjects from the 1969-1977 study, including their parents’ age, schooling attainment, and assetholdings. The censuses in more recent years are additional sources of information about their children, including attendance and completed grades of schooling. Using the 1996 and 2002 censuses, we are able to identify 1318 children over age 7 whose 558 mothers were participants in the original study. Following the generational terminology proposed in our conceptual framework, from this point forward, we refer to the original study subjects, who were age 0-7 in 1969-1977, as our middle generation, or G2. We refer to their parents as the first generation, or G1, and to their children as the third generation, or G3. We draw upon information on all three generations in the analyses. Although this decision greatly enriches our analyses, it also increases the chance of missing data, including attrition from the study. Of the 919 potential female G2s in 2002-4, 679 (74%) have information on schooling and fertility from the 2002-4 study and have had at least one live birth.10 However, only 630 (69%) of the G2s additionally have late adolescent height to represent long-run nutritional status, calculated from information collected during either the 1988 or 2002-4 studies, depending on age.11 These requirements for G2 inclusion into the analyses, coupled with those for 10

685 (75%) have an adult Raven’s score, fertility data, and at least 1 live birth. 683 (74%) have an adult reading and math score, fertility data, and at least 1 live birth. 11 634 (69%) of the G2s have an adult Raven’s score, fertility data, at least 1 live birth, and late adolescent height. 632 (69%) have an adult reading and math score, fertility data, at least 1 live birth, and late adolescent height.

8

G3 inclusion reduce our sample sizes available for analysis. For example, G2 mothers in our analyses on the impact of maternal schooling and height on G3 birth weight had to have completed the fertility and schooling questionnaires during the 2002-4 study, to have height information from either the 2002-4 or 1987-8 studies, and to have had at least one live birth between 1991 and 1999 while the G2 mother was living in one of the four study villages. These various requirements reduce the analysis sample to 573 G3 children of 325 G2 mothers (35% of the 919 potential G2 female subjects in 2002-4 and 52% of the 630 women who had data on their schooling, height and fertility, and at least one live birth).12 Likewise, our analyses of the determinants of G3 nutritional status at 36-months include 294 mothers and 456 of their children.13 Our analyses of G3 schooling include 479 mothers and 1162 of their children.14 The high rates of attrition not only reduce sample size, but also may be non-random, especially considering that subjects who do appear in our analyses are a select group of G2 mothers who were generally non-migrants: present in the four communities during the intervention and in the 1990’s, and who were accessible (though not necessarily in the communities) from 2002-4. We will consider possible attrition bias in the next revision of this paper, using methods parallel to those in Fitzgerald, Gottschak and Moffitt (1998). We note that such explorations in other studies on these data (e.g., Behrman et al. 2006a, Maluccio et al. 2006) do not find large impacts of attrition on the estimated coefficients. Finally, a number of these G2 individuals are siblings or half-siblings, so we control for mother cluster effects in the estimation of the standard errors that are reported in Tables 2-4 below. [Figure 1 about here] Section 3.2 Central Variables for the Analysis Tables 1a and 1b present means and standard deviations (SD) for the outcome variables, explanatory variables, and instruments, respectively. Table 1a includes G3 outcome summaries for the regressions with maternal schooling, both including and excluding height as a right-side variable. Table 1b includes summaries of maternal intellectual human capital for regressions including and excluding height on the right side. The data summaries for maternal height, other controls, and first-stage instruments are all based on the sample used in regressions with maternal schooling and height. [Tables 1a & b about here] Dependent Variables: G3 Children Outcomes (Y): (1) Birth Anthropometry: birth length (cm) and weight (kg) were collected during 19911999. The mean birth weight of G3s in this study is 3.0 kg, which is above the standard 12

574 G3s of 326 G2 mothers are included in the analysis of the impact of mothers’ Raven’s score and height on G3 birth weights. 572 G3 children of 325 G2 mothers are included in the analysis of the impact of mothers’ reading or math scores and height on G3 birth weights. 13 This sample is for the analysis with endogenous maternal schooling and height. The analysis with endogenous maternal Raven’s scores and height includes 295 G2 mothers and 459 of their children. The analysis with endogenous reading or math and height include 294 G2 mothers and 457 of their children. 14 This sample is for the analysis with endogenous maternal schooling and height. The analysis with endogenous maternal Raven’s scores and height includes 481 G2 mothers and 1165 of their children. The analysis with endogenous reading or math and height includes 479 G2 mothers and 1160 of their children.

9

cutoff for low birth weight of 2.5 kg. However, there is a fair amount of variance in this estimate, and 13% of the births were below 2.5 kg. (2) Nutritional status at 36-months: represented by length-for–age (LAZ), weight-for–age (WAZ), and weight-for–length (WLZ) z-scores from the 1996-1999 study.15 Not all G3 children were measured at exactly 36 months. Dummy variables for age at measurement were regressed upon all z-scores in order to generate correction coefficients. These corrections were then made to the z-scores closest to when the child was 36 months. According to these scores, this population not surprisingly is malnourished relative to the reference population, particularly with regard to stunting (LAZ), which generally is considered to be an indicator of the long-run impact of early childhood nutrition on subsequent development. Indeed, 43% of the children have LAZ values below -2.0, which is the standard cutoff used in the literature for stunting. (3) Schooling of children by 2002-4: represented by the difference in grades of schooling completed from the age-cohort mean for all children over age 7 years, taken from the 1996 and 2002-4 censuses. The difference from the mean measure of schooling is positive when a child has more schooling than the cohort mean. One tenth of the G3 children aged 7-23 years who were linked successfully with their G2 mother via the community censuses had not entered school at the time of the survey. Right-side Endogenous Variables: G2 Intellectual Capital Stocks At or Prior to First Parenting (K): 1) Maternal Schooling: Completed schooling attainment, as measured in 2002-4. Since in our population most individuals complete their schooling during adolescence, we use the 2002-4 adult measure as our best approximation of schooling before first parenting. Of all 2002-4 study participants, two-thirds completed their schooling by age 13 and over four-fifths by age 15. Only 4% initiated their first union (which for most signals impending first parenthood) before completing their schooling. The mean schooling for all women interviewed in 2002-4 (n=786) is 4.3 grades, with a SD of 3.3. The distribution has a mode at six grades (24% of women), completion of primary school. There also are secondary modes at zero grades (15%) and three grades (14%). The average grades completed for women included in our analyses are 3.4 (SD=2.6), 3.5 (SD=2.7), and 3.4 (SD=2.7) for G3 anthropometry at birth, G3 36-month z-scores, and G3 schooling respectively.16 2) Maternal Nonverbal Cognitive Skills: In 1988-9 and 2002–4, all respondents were administered Raven’s Progressive Matrices (Raven, Court and Raven 1984), a widelyused nonverbal measure of interpretative cognitive skills that consists of a set of shapes 15

Z-scores give the number of standard deviations from the median of the distribution for a reference population. As in most other studies, we use the NHS-CDC standards. 16 These averages are from G2 mothers in those analyses including maternal schooling and height on the right side. This includes 325 mothers of 573 children in regressions on G3 anthropometry at birth (birth weight), 294 mothers of 456 children in regressions on G3 36-month z-scores, and 479 mothers of 1162 children in regressions on G3 schooling.

10

and patterns, with the respondent asked to supply the ‘missing piece.’ In both studies, we administered three of the five scales (A, B and C with 12 questions each for a maximum possible score of 36) because pilot data suggested and subsequent survey data confirmed that few respondents were able to progress beyond the third scale. Within each study, this estimated score varies considerably, and nonverbal cognitive skills in our sample also appear to have improved on average over time. Women scored a mean of 10.4 points in 1988-9, with a standard deviation of 4.2 (n=683), while in 2002-4 they scored a mean of 16.3 points with a SD of 5.4 points (n=779). The Raven’s test exhibits adequate test-retest reliability (correlation of 0.87) and internal consistency in this population in the 1988-9 study (Pollitt et al. 1993). The correlation between the tests conducted in 1988-9 and 2002-4 in our sample of 551 women who took both tests is 0.55 (p=0.000). Since many of our subjects were still of schooling age during the 1988-9 study (ages 1125), and since the 2002-4 study took place well after initial child-bearing for many subjects (ages 25-42), we devised a composite measure of these two tests to approximate an “adult” Raven’s score closest to when the subject started to have children.17 First, we predict the 1988 Raven’s score (using the 2002-4 Raven’s score, age in 1988, gender, age and gender both interacted with the 2002-4 score) for the 228 female subjects who have a 2002-4 score but who did not participate in 1988-9. For subjects who were older than 18 years at the time of the 1988-9 Raven’s test, we use their actual Raven’s score from 1988-9 (or predicted score, for those not present in 1988-9). For subjects who were 18 years old or younger at the time of the 1988-9 Raven’s test, we correct their actual (or predicted) 1988-9 Raven’s score to what we predict their score would have been had the individual been over 18 at the time. The correction used is a sum of the coefficients on age and age interacted with the 2002-4 score from the prediction regression described above, multiplied by the number of years away from “adulthood” (age 19) they were in 1988. The mean “adult” Raven’s score for women is 9.8 with standard deviations (SD) of 4.0 (n=911). The correlation between this “adult” score and the 1988-9 and 2002-4 tests are 0.98 (p=0.000) and 0.56 (p=0.000) respectively. The average score for women included in our analyses are 9.7 (SD=3.6), 9.7 (SD=3.8), and 10.3 (SD=3.7) for G3 anthropometry at birth, G3 36-month z-scores, and G3 schooling respectively.18 3) Maternal Reading-Comprehension Cognitive Skills: In 1988-9 and 2002–4, the vocabulary and reading-comprehension modules of the Inter-American Reading and Comprehension Tests (IARC, see Manuel 1967) were administered to study participants. In 2002-4, the vocabulary portion had 45 questions and the reading comprehension portion had 40 questions, yielding a maximum possible score of 85 points.19 During 17

Women’s average age at first union is 19.8 years, and their average age at first parenting is 20.3 years (Behrman et al. 2006b). 18 These averages are from G2 mothers in those analyses including maternal Raven’s score and height on the right side. This includes 326 mothers of 574 children in regressions on G3 anthropometry at birth (birth weight), 295 mothers of 459 children in regressions on G3 36-month z-scores, and 481 mothers of 1165 children in regressions on G3 schooling. 19 In 1988-9, the vocabulary portion had 40 questions and the reading comprehension portion had 40 questions, yielding a maximum possible score of 80 points.

11

both studies, study participants were given a pre-literacy screening test before taking the reading-comprehension test.20 In 2002-4, 632 of the 777 women (81%) who participated in the pre-literacy screen passed the screen and took the IARC. In 1988-9, 507 women took the IARC (74% of 683 women who participated in the pre-literacy screen). The 19% and 26% of the sample who did not pass the pre-literacy screen in 2002-4 and 1988-9, respectively, are assigned a value of zero for the reading-comprehension tests. Including those we score a zero, the mean score is 34.4 in 2002-4 and 31.6 in 1988-9, with standard deviations (SD) of 21.9 and 21.5, respectively, indicating substantial sample variation. The IARC tests demonstrated adequate within year test-retest reliability (correlation coefficients of 0.87 and 0.85 for vocabulary and reading), internal consistency and validity during the 1988-9 study (Pollitt et al. 1991, 1993). The correlation between the tests conducted in 1988-9 and 2002-4 in our sample of 551 women who took both tests is 0.82 (p=0.000). “Adult” reading comprehension scores closest to when the subject started to have children were constructed in an identical manner to that described above for Raven’s scores. Again including those we score at zero, the mean “adult” reading comprehension score for women is 31.3 with standard deviations (SD) of 21.2 (n=909). The correlation between this “adult” score and the 1988-9 and 2002-4 tests are 0.99 (p=0.000) and 0.86 (p=0.000) respectively. The average score for women included in our analyses are 27.9 (SD=20.1), 27.7 (SD=21.4), and 29.7 (SD=21.6) for G3 anthropometry at birth, G3 36month z-scores, and G3 schooling respectively.21 4) Maternal Numeric Ability: In the 1988-9 study, respondents’ abilities to read numbers and prices, order prices, and perform simple arithmetic calculations involving prices, salaries, and distances were assessed. The range of possible scores is between 0 and 41. Women scored an average of 31.3 points with a SD of 8.7 (n=683). No equivalent assessment of numeric ability was undertaken during the 2002-4 study. However, we generate a similar “adult” numeric ability closest to age at initial parenting, predicting missing 1988 numeric ability scores using the 2002-4 IARC, and also using the same method for predicting “adult” numeric ability scores of those subjects 18 or younger in 1988. The correlation between the actual numeric ability test conducted in 1988-9 and the IARC conducted in 2002-4 in our sample of 551 women who took both tests is 0.72 (p=0.000). The mean “adult” numeric ability score for women is 32.9 with standard deviations (SD) of 8.2. The correlation between this “adult” score and the 19889 numeric ability test and 2002-4 IARC tests are 0.97 (p=0.000) and 0.78 (p=0.000), respectively. The average scores for women included in our analyses are 31.7 (SD=8.4), 20

Subjects who reported completing six or more grades of schooling were assumed to be literate. Respondents who reported having completed fewer than three grades of schooling, and those who reported three to five grades of schooling but could not read correctly the headline of a local newspaper article, were given a pre-literacy test that began with reading out letters. They were considered literate if they passed the test with fewer than five errors out of 35 questions, the most difficult of which was reading a five word sentence aloud.

21

These averages are from G2 mothers in those analyses including maternal reading comprehension and height on the right side. Estimates are based on sample sizes of 325 G2 mothers of 572 children in regressions on G3 anthropometry at birth (birth weight), 294 G2 mothers of 457 children in regressions on G3 36-month z-scores, and 479 G2 mothers of 1160 children in regressions on G3 schooling.

12

31.4 (SD=8.7), and 32.3 (SD=8.1) for G3 anthropometry at birth, G3 36-month z-scores, and G3 schooling respectively.22 G2 Biological Capital Stocks At or Prior to First Parenting (K): 1) Maternal Long-Run Nutritional Status: We use maternal height (cm) at age 18, by which age most females have attained their adult height. We use a combination of the 1988 and 2002-4 data to construct this variable, taking the 1988 measure for those who were older than 18 in 1988, and the 2002-4 measures for those aged 11-18 in 1988. We also use the 2002-4 measure of height for those who were older than 18 in 1988 but who were not measured at that time. Right-side G3 Individual Characteristics (I): 1)

Gender (male=1)

2)

Twin

Initial Conditions (F0, C0, E0, I0) for IV Estimates: G2 parental characteristics and family background (F0): We include the G1 mothers’ and fathers’ schooling attainments, and a constructed socioeconomic status score that is the first principal component of both the assets owned and the housing characteristics of the G1 household in 1975 (Maluccio, Murphy and Yount 2005). We also include a dummy variable for whether either G1 parent died before the index G2 was age 15 years. Community characteristics during G2’s childhood (C0): Communities differ significantly in their learning environments, in part because of different experiences of prior generations regarding schooling and occupational structure (Bergeron 1992; Maluccio et al. 2005). We control for the permanent, or fixed, dimensions of these differences by including village fixed effects in the firststage regressions. Genetic endowments (E0): We do not have direct observations on genetic endowments beyond the sex of the individual G2s and G3s, and in the analyses in this paper we use only G2 mothers and their children. Observed individual G3 characteristics (I0): These include age at the time of the 2002-4 interview and whether the individual was a twin, which may have longer-run implications associated with the generally lower birth weight of twins (e.g., Behrman and Rosenzweig 2004). Observed shocks and events (ΔC): 22

These averages are from G2 mothers in those analyses including maternal numeric ability and height on the right side. This includes 325 mothers of 572 children in regressions on G3 anthropometry at birth (birth weight), 294 mothers of 457 children in regressions on G3 36-month z-scores, and 479 mothers of 1160 children in regressions on G3 schooling.

13

Natural, market or policy events (ΔC): We construct variables at the community level that relate as closely as possible to the timing of key decisions in G2 mothers’ human capital development. For example, using information reported in earlier work about infrastructure, markets and services in the villages (Pivaral 1972; Bergeron 1992), complemented with a retrospective study in 2002 (Estudio 1360, 2002), and prior literature suggesting that measures of community school quality employment opportunities are important determinants of schooling achievements, we construct variables such as student-teacher ratios (a proxy for school quality) when the mothers most likely started their schooling, age 7 years, as well as work in local markets when the mothers most likely were making the decision to continue schooling or to join the work force, age 15 years. The variable reflecting work availability in local markets is equal to one if a “boom” was occurring in any local market: yuquilla production in San Juan, vegetable cooperatives in Conacaste, or intensive hiring of community members at a cement factory near to Conacaste and Santo Domingo. Thus, while reflecting community level characteristics, these variables vary by single-year age cohorts within each village, as well as across villages. Since this measure more closely relates the availability and longevity of schools and markets to the period in a woman’s life when critical decisions (e.g., attending school, working in the labor market) were being made, it is an improvement over the more typical approach of including indicators about such factors in a given year for a population with different ages at that point in time. Experimental nutritional shocks (ΔC): The set of observed nutritional shocks that we consider relate to the nutritional interventions underlying the original data collection (see Section 3.1). We construct two measures of exposure to the interventions based entirely on the birth year of the G2 mother, the dates of operation of the interventions, and where the G2 mother lived as a child. The first is a dummy variable control for cohort effects and the second is the exposure to the Atole treatment for that cohort. For each G2 mother, we determine whether she was exposed to either intervention for the entire period from birth to 36 months of age. The Atole intervention indicator is then calculated by multiplying the cohort measure by a dummy variable indicator of whether or not the G2 mother as a child lived in one of the two Atole villages. We include these two measures separately. 4. Results Tables 2a, 2b, 3a, 3b, and 4 present OLS and IV estimates of the impact of mothers’ intellectual human capital and height on a number of next-generation outcomes: anthropometry at birth, zscores at 36 months, and schooling. Appendix 1 presents first-stage results for endogenous rightside variables, mothers’ schooling and height, in selected G3 outcome regressions, weight at birth, length-for-age z-score at 36 months, and the child’s deviation in schooling attainment from the agecohort mean schooling attainment (other stage-one results are available upon request); Appendices 2, 3, and 4, respectively, present analogous first-stage results for endogenous G2 Nonverbal Cognitive Skills (Raven’s score) and height, endogenous reading (IARC) scores and height, and endogenous math scores and height. Instruments that are excluded from the second stage include dummy variables for the G2 respondent’s exposure to the nutritional intervention at critical ages during infancy, village dummies, birth year, the student-teacher ratio when the G2 mother was age 7

14

years, and the availability of local markets when the G2 mother was age 15 year. Controls for family background include her mother’s schooling, her father’s schooling, her natal household’s wealth index in 1975, and dummy variables indicating whether each of the above is missing, whether any parent of the G2 mother died before she reached 15, and whether the G2 mother was a twin. In all of the IV regressions, the F-test of the instruments excluded from the second stage indicate that the instrument set is jointly significant at least at the 0.000 level in predicting the endogenous right-side regressor, although F-statistics are always greater for the measure of intellectual human capital than for height. The Cragg-Donald F-test for weak instruments exceeds the critical value of 4.45 (for two endogenous regressors and 17 excluded instruments) in estimates using grades of schooling, Raven’s scores, and math scores, which implies a bias relative to OLS of less than 0.30 (Stock and Yogo 2002) for the estimates on G3 anthropometry at birth (Table 2a) and G3 schooling (Table 4), except for the IARC score in the former. However, the apparent bias is somewhat greater in the estimates in Tables 3a and 3b for anthropometry at 36 months. The p values for the Hansen J statistic for overidentification do not reject the null hypothesis that the instruments are independent of the second-stage disturbance term at the usual 0.05 significance level in any of the regressions with both maternal intellectual capital stocks and maternal height. Thus these diagnostics suggest that our IV estimates, which we prefer to OLS on a priori grounds because they attempt to deal with the behavioral determination of maternal human capital, generally are fairly satisfactory. Though we ideally would like estimates that performed better on the weak instrument test, nevertheless, if, despite a 0.30 bias relative to IV they suggest different impacts of maternal human capital on child human capital than do the standard OLS estimates, this result would suggest some reason for being concerned about standard estimates and how well they represent the true causal impact of maternal human capital on child human capital. Comparisons of OLS and IV coefficient estimates for maternal schooling and height are of central interest to this paper because they reveal to what extent treating multiple dimensions of maternal human capital as endogenous affects the coefficients of the impacts of maternal human capital on child human capital. Most of the existing literature on the impact of maternal human capital also typically uses schooling attainment, so comparisons of our results with those of existing studies also can illustrate the potential value of treating maternal human capital—proxied by schooling—as endogenous. In the birth weight equation (Table 2a), the coefficient estimate for mothers’ schooling is not significant in any of the specifications. The coefficient estimate for mothers’ height remains roughly the same as the OLS estimate when schooling and height are endogenized, and is significant at the standard 0.05 level for both the OLS and IV estimates. In the birth length equation (Table 2), however, the coefficient estimate for mothers’ height drops from 0.106 to 0.067 when both maternal human capital measures are treated as endogenous and is significant only at the 0.10 level in the IV estimates. The coefficient estimate for mothers’ schooling in the birth length equation is not significant in either the OLS or IV regressions. Do nonverbal cognitive skills affect anthropometrics at birth? OLS estimates in Table 2a indicate a positive and significant impact of nonverbal skills on birth weight, which increases when both Raven’s scores and height are endogenized. The coefficient on the Raven’s score increases from 0.013 to 0.029 from the OLS to the IV specification when both Raven’s and height are endogenized. However, similar to the results for schooling, the coefficient estimate of the Raven’s score in the

15

birth length equation is not significant in either OLS or IV regressions. Maternal height has a positive and significant impact on birth weight in both OLS and IV regressions, with only a marginal increase in the size of the coefficient. Although the OLS coefficient of maternal height is significant in the birth length equation, it is no longer significant once the endogeneity of Raven’s scores and maternal height is considered. We next examine whether skills in specific areas of competence—reading and math—have similar effects on children’s birth outcomes in Table 2b. In general, test scores in reading or math do not have a significant impact on anthropometric outcomes at birth. Maternal height continues to be a positive and significant predictor of birth outcomes, but the coefficient estimate decreases in magnitude and becomes less significant once both test scores and height are endogenized. For example, the OLS coefficient on maternal height in the birth length equation drops from 0.102 to 0.055, in the specification with IARC scores. The coefficient on maternal height does not change perceptibly between OLS and IV estimates in the birth weight equation using both IARC and math scores; both coefficients remain statistically significant at the 0.10 level, but the coefficient on maternal height also decreases from the OLS to the IV regression and loses significance in the birth length equation. Table 3a presents OLS and IV estimates of length-for-age (LAZ), weight-for-age (WAZ) and weight-for-length (WLZ) z-scores at 36 months. Mothers’ schooling is significant only when height is excluded as a regressor, with the exception that maternal schooling is significant only at the 10% level in the WLZ equation with both schooling and height endogenous. Mothers’ heights, however, exert a positive and significant impact on all LAZ and WAZ (but not WLZ). The coefficient estimates for mothers’ height for LAZ and WAZ increase in magnitude if maternal human capital is treated as endogenous. The coefficient in the LAZ regression increases from 0.074 to 0.133; in the WAZ regression from 0.060 to 0.120. These increases suggest that failure to account for the endogeneity of mothers’ human capital stocks may underestimate the returns to mothers’ long-run nutritional status, at least in terms of child anthropometric outcomes. Results for Raven’s test scores are similar. Raven’s scores are not significant in any of the equations for anthropometric outcomes, whereas the coefficient on mothers’ height is positive and significant in LAZ and WAZ for OLS, and for all three outcomes using IV. Similar to the results when schooling and height are endogenized, the coefficients on mothers’ height increase in size when both Raven’s scores and height are endogenized. The coefficient on mothers’ height increases from 0.079 to 0.124 in the LAZ equation, from 0.063 to 0.138 in the WAZ equation, and from 0.011 to 0.058 in the WLZ equation. In the WLZ equation, the coefficient on maternal height is not significant in the OLS regression but is significant at the 0.01 level once nonverbal skills and height are endogenized. Table 3b examines the impact of reading and math test scores on anthropometric outcomes at 36 months. While coefficients on reading scores and math scores are positive and significant when they are used as the only indicator of human capital, the coefficients typically decrease in size once height is included in the LAZ and WAZ regressions. Coefficients on the reading score also are not consistently significant in IV specifications when both IARC scores and height are included. The OLS coefficients on the reading scores are positive and highly significant in the LAZ and WAZ regressions (at 1% and 5% respectively), but are no longer significant once reading scores and

16

height are endogenized. In contrast, the insignficant OLS coefficient on the IARC score in the WLZ regression increases from 0.003 to 0.014 in the IV counterpart, and is significant in the latter at 0.01. Math test scores in the LAZ and WAZ regressions are significant and positive in the OLS specification, but lose significance once math scores and height are endogenized. However, the coefficient on the math test score in the WLZ regression increases from 0.010 in the OLS specification to 0.035 in the IV specification, and improves in significance from being significant at 0.10 to being significant at the 0.01 level. Similar to the results for schooling and Raven’s tests, coefficients on maternal height increase in magnitude once intellectual human capital and height are both treated as endogenous. These reinforce our findings that a failure to account for maternal biological capital may underestimate the returns to investing in mothers’ nutrition. We also note that for all the anthropometric outcomes analyzed in the preceding tables, the failure to include maternal height in the regressions results in poorer performance of the overall regressions, as evidenced in the F statistics, and overestimates of the impact of maternal intellectual capital on anthropometric outcomes. Finally, we examine the impact of various measures of mothers’ human capital on schooling attainment in Table 4. Note that because the G3s under consideration can be as young as 7 years old (and range from 7 to 25 years of age), many children will not have completed school at the time of our data collection, and completed grades of schooling is censored. To deal with the problem of age censoring, we examine the deviation of each child’s completed grades of schooling from the age cohort mean. As expected from the earlier literature, in the OLS estimates mothers’ schooling attainment has a positive and significant coefficient estimate for their children’s schooling relative to the cohort mean. Interestingly, in the OLS estimates for completed grades of schooling as a deviation from the age cohort mean, mothers’ height at age 18 also has a significantly positive coefficient estimate. The coefficient estimate for mothers’ schooling in the grades-of-schooling regression increases from 0.119 to 0.172 when we take schooling and height as endogenously determined. The coefficient estimate for mothers’ reading comprehension increases slightly from 0.016 to 0.020 when we endogenize schooling and height, but the significance drops from the 1% to the 10% level. Interestingly, however, the coefficient estimates on Raven’s scores and math scores all are significant in the OLS specification and lose significance once both test scores (Raven’s and math scores) and height are endogenized. This result suggests that grades completed (and perhaps maternal reading skills, albeit to a weaker degree of significance) are a more important determinant of children’s schooling attainment than nonverbal skills and math scores. Our results also confirm that maternal height remains an underappreciated determinant of children’s progression through school if it is not considered as an endogenous determinant of investments in child schooling. The coefficient for height in the regression that includes maternal schooling attainment increases threefold, from 0.04 to 0.135 when both maternal schooling and height are endogenized. Consistent with the results for maternal schooling attainment, the coefficient for height also increases from 0.052 to 0.163 in the regression using Raven’s scores, from 0.035 to 0.088 in the regression using IARC scores (but loses significance), and from 0.040 to 0.123 in the regression using math scores.

17

5. Conclusions In this paper, we explore, for the first time, how estimates of the impact of maternal schooling attainment and maternal long-run nutritional status on child human capital are affected if both of these components of maternal human capital are treated as behaviorially determined, using an unusually rich longitudinal data set collected over 35 years in Guatemala. .Our estimates are provocative. They suggest that the common procedure of treating maternal human capital as exogenous may yield misleading coefficient estimates for the impacts of maternal human capital on child human capital in the Guatemalan context. Similar results are reported in studies in the United States, Norway, and Nicaragua, which have considered this question but used only a single measure of maternal human capital—schooling attainment. For maternal schooling, the comparison between our OLS and IV estimates under the assumption that the IV estimates are “better” estimates suggests that the OLS estimates may understate slightly the impact on grades of schooling relative to the age-cohort mean. Moreover, using maternal schooling as the only measure of human capital tends to overstate the magnitude of the impact of maternal schooling on child schooling; coefficients on mothers’ schooling decrease once (endogenous) maternal height is included as a regressor. Previous studies have also used only one measure of maternal intellectual capital—schooling. Our study suggests that different measures of intellectual capital seem to capture different aspects of cognitive achievement, and do not have the same impact on child outcomes. For example, once endogeneity of both maternal intellectual capital and height are considered, grades completed, the conventional measure of maternal intellectual capital, is significant at standard levels of significance and positive only in the child schooling regression, is significant only at 10% in the WLZ regression, and does not significantly affect any of the birth outcomes. In contrast, Raven’s scores, a measure of adult nonverbal cognitive skills, are significant in the birth weight equation once maternal human capital is treated as endogenous, but are insignificant for the other child outcomes. Reading and math scores are significant determinants of WLZ, but not of the other outcomes (with the exception of the deviation of child’s schooling attainment from the cohort mean, where reading scores are significant, albeit only at a 10% level). Using schooling as the lone measure of maternal intellectual capital may therefore fail to capture the different pathways by which different aspects of cognitive functioning may affect child outcomes. For maternal height, a comparison of OLS and IV estimates suggests that for all but one of the child outcomes considered (the exception is birth length) the OLS estimates understate, in some cases substantially, the causal impact of maternal long-run nutritional status on child anthropometric and schooling outcomes. The prior literature to our knowledge does not consider the possibility that maternal height should be treated as endogenous in estimates of such intergenerational human capital relations. Our estimates imply, thus, not only that in a number of cases are the common estimates likely to be misleading regarding the magnitude and significance of components of maternal human capital due to endogeneity of maternal human capital, but they are likely to understate the importance of long-run maternal nutritional status relative to maternal intellectual capital, particularly schooling attainment, in determining child human capital. Apparently “biological” human capital, which is thought to work substantially through the development of cognitive potential early in the children’s life cycle (e.g., Engle et al. 2006) is substantially more

18

important relative to “intellectual” human capital for the determination of child human capital, particularly with regard to health and anthropometric outcomes, than would be perceived in the standard estimates that dominate in the previous literature.

19

6. References Behrman, Jere R., John Hoddinott, John A. Maluccio and Reynaldo Martorell, 2005, “Does it Pay to Become Taller? Or Is What You Know All That Really Matters?,” Philadelphia, PA: University of Pennsylvania, mimeo, 2005. Behrman, Jere R., John Hoddinott, John A. Maluccio, Erica Soler-Hampejsek, Emily L. Behrman, Reynaldo Martorell, Manuel Ramirez and Aryeh D. Stein, 2006a, “What Determines Adult Skills? Impacts of Pre-School, School-Years and Post-School Experiences in Guatemala”, Philadelphia, PA: University of Pennsylvania, mimeo. Behrman, Jere R., Alexis Murphy, Agnes Quisumbing, Usha Ramakrishnan, and Kathryn Yount. 2006b. "What is the Real Impact of Schooling on Age of First Union and Age of First Parenting? New Evidence from Guatemala" Policy Research Working Paper No. 4023. Washington DC: World Bank. Behrman, Jere R. and Mark R. Rosenzweig, 2002,“Does Increasing Women’s Schooling Raise the Schooling of the Next Generation?”, American Economic Review 92:1(March 2002), 323-334 Behrman, Jere R. and Mark R. Rosenzweig, 2004, “Returns to Birthweight,” Review of Economics and Statistics 86:2 (May), 586-601. Behrman, Jere R. and, Mark R. Rosenzweig, 2005,“Does Increasing Women’s Schooling Raise the Schooling of the Next Generation? – Reply” American Economic Review 95:5 (December), 17451751. Behrman, Jere R., Mark R. Rosenzweig, and Paul Taubman, 1994, "Endowments and the Allocation of Schooling in the Family and in the Marriage Market: The Twins Experiment," Journal of Political Economy 102:6 (December), 1131-1174. Behrman, Jere R., Mark R. Rosenzweig, and Paul Taubman, 1996, "College Choice and Wages: Estimates Using Data on Female Twins," Review of Economics and Statistics 73:4 (November), 672-685. Behrman, Jere R. and Barbara L. Wolfe, 1987a, "How Does Mother's Schooling Affect the Family's Health, Nutrition, Medical Care Usage, and Household Sanitation?" Journal of Econometrics 36, 185-204. Behrman, Jere R. and Barbara L. Wolfe, 1987b, "Investments in Schooling in Two Generations in Pre- Revolutionary Nicaragua: The Roles of Family Background and School Supply," Journal of Development Economics 27:1-2 (October), 395-420 (reprinted in International Trade, Investment, Macro Policies and History: Essays in Memory of Carlos F. Diaz-Alejandro (eds., Pranab Bardhan, Jere R. Behrman, and Albert Fishlow), Amsterdam: North-Holland, 1987), 395-420.

20

Bergeron, Gilles, 1992. “Social and Economic Development in Four Latino Communities of Eastern Guatemala: A Comparative Description,” Food and Nutrition Bulletin, 14:3, 221-236 . Bianchi, Suzanne. 2000. Maternal employment and time with children: Dramatic change or surprising continuity? Demography 37: 139-154. Bianchi, Suzanne, Wight, Vanessa, and Raley, Sara. 2005. Maternal employment and family caregiving: Rethinking time with children in the ATUS. Paper prepared for the ATUS Early Results Conference, Bethesda, MD, December 9, 2005. Black, Sandra E., Paul J. Devereux, and Kjell G. Salvanes, 2005, “Why the Apple Doesn’t Fall Far: Understanding Intergenerational Transmission of Human Capital,” American Economic Review 95:1 (March), 437-449. Brooks-Gunn, J, PK Klebanov, GJ Duncan. 1996. Ethnic differences in child’s intelligence test scores: Role of economic deprivation, home environment, and maternal characteristics Child Development 67:396–408. Cleland, J and JK Van Ginneken. 1988. Maternal education and child survival in developing countries: the search for pathways of influence. Social Science and Medicine 27: 1357–68. Engle, Patrice L., Maureen M. Black, Jere R. Behrman, Meena Cabral de Mello, Paul J. Gertler, Lydia Kapiriri, Reynaldo Martorell, Mary Eming Young, Lancet Child Development Series Steering Committee, 2006, “Strategies to Avoid the Loss of Potential Among 240 Million Children in the Developing World,” Lancet (forthcoming). Estudio 1360. 2002. Changes in the Socioeconomic and Cultural Conditions that Affect the Formation of Human Capital and Economic Productivity. Final report presented to INCAP, 13 May 2002. Fitzgerald, John, Peter Gottschalk, and Robert Moffitt, 1998, “An Analysis of Sample Attrition in Panel Data,” The Journal of Human Resources 33 (2): 251-99. Grajeda, Rubén, Jere R. Behrman, Rafael Flores, John A. Maluccio, Reynaldo Martorell and Aryeh D. Stein, 2005, “The Human Capital Study 2002-04: Tracking, Data Collection, Coverage, and Attrition,” Food and Nutrition Bulletin 26:2 (Supplement 1), S15-S24. Habicht, Jean-Pierre and Reynaldo Martorell, 1992, “Objectives, Research Design, and Implementation of the INCAP Longitudinal Study” Food and Nutrition Bulletin 14:3, 176-190. Maluccio, John A., John Hoddinott, Jere R. Behrman, Reynaldo Martorell, Agnes R. Quisumbing, and Aryeh D. Stein, The Impact of an Experimental Nutritional Intervention in Childhood on Education among Guatemalan Adults, International Food Policy Research Institute, Food Consumption and Nutrition Discussion Paper No.207. June 2006 21

Maluccio, John A., Paúl Melgar, Humberto Méndez, Alexis Murphy and Kathryn M. Yount, 2005, “Social and Economic Development and Change in Four Guatemalan villages: Demographics, Schooling, Occupation, and Assets.” Food and Nutrition Bulletin 26:2 (Supplement 1), S25-S45. Maluccio, John A., Alexis Murphy and Kathryn M. Yount, 2005, “Research note: a Socioeconomic Index for the INCAP Longitudinal Study 1969-77,” Food and Nutrition Bulletin 26:2 (Supplement 1), S120-S124. Martorell, Reynaldo. 1997. "Undernutrition during pregnancy and early childhood and its consequences for cognitive and behavioral development," in ME Young (ed.) Early Child Development: Investing in Our Children’s Future, Amsterdam, The Netherlands: Elsevier, 39-83. Martorell, Reynaldo, Jere R. Behrman, Rafael Flores, and Aryeh D. Stein, 2005, “Rationale for a Follow-up Focusing on Economic Productivity,” Food and Nutrition Bulletin 26:2 (Supplement 1), S5-S14. Martorell Reynaldo, Jean-Pierre Habicht and Juan A. Rivera, 1995, “History and design of the INCAP Longitudinal Study (1969-77) and its Follow up (1988-89),” Journal of Nutrition, 125:4S, 1027S-1041S. Pivaral, V.M., 1972, Características Económicas y Socioculturales de Cuatro Aldeas Ladinas de Guatemala, Ministerio de Educación Pública, Instituto Indigenista Nacional, Guatemala. Plug, Erik , 2004, “Estimating the Effect of Mother's Schooling on Children's Schooling Using a Sample of Adoptees,” American Economic Review 94:1 (March), 358-368. Plug Erik and Wim P. M. Vijverberg, 2003, “Schooling, Family Background and Adoption: Is it Nature or Nurture?” Journal of Political Economy 111: 611-41. Quisumbing, Agnes R., Jere R. Behrman, John A. Maluccio, Alexis Murphy, and Kathryn M. Yount, 2005, “Levels, Correlates and Differences in Human, Physical, and Financial Assets Brought Into Marriages by Young Guatemalan Adults,” Food and Nutrition Bulletin 26:2 (Supplement 1), S55-S67. Ramakrishnan U, Barnhart H, Schroeder DG, Stein AD, Martorell R. Early childhood nutrition, education and fertility milestones in Guatemala. J Nutrition 1999;129:2196-2202. Ramakrishnan, Usha, Kathryn Yount, Jere R. Behrman, Misa Graff, Ruben Grajeda Toledo, and Aryeh D. Stein, 2005, “Fertility Patterns and Reproductive Outcomes among Young Guatemalan Adults,”Food and Nutrition Bulletin 26:2 (Supplement 1), S68-S77..

22

Read, Merrill .S. and Jean-Pierre Habicht, 1992, “History of the INCAP Longitudinal Study on the Effects of Early Nutrition Supplementation in Child Growth and Development,” Food and Nutrition Bulletin 14:3, 169–175. Rivera, Juan, Reynaldo Martorell, and Hilda Castro, 1993, “Data collection of the INCAP follow-up study: Organization, coverage, and sample sizes”, Food and Nutrition Bulletin, 14:3, 258-269. Schroeder, Dirk G., Haley J. Kaplowitz and Reynaldo Martorell, 1992, “Patterns and Predictors of Participation and Consumption of Supplement in an Intervention Study in Rural Guatemala,” Food and Nutrition Bulletin 14:3, 191–200. Stein Aryeh D., Jere Behrman, Ann DiGirolamo, Ruben Grajeda Reynaldo Martorell, Agnes Quisumbing, and Usha Ramakrishnan, 2005, “Schooling, Educational Achievement and Cognitive Functioning Among Young Guatemalan Adults,” Food and Nutrition Bulletin 26:2 (Supplement 1), S46-S54. Stern, Nicolas, 2001, “Investing for Growth and Poverty Reduction: Institutions and People,” Speech delivered in Islamabad, Pakistan 29 March 2001 by Senior Vice President and Chief Economist of the World Bank. Stock, James H. and Motohiro Yogo, 2002, “Testing for Weak Instruments in Linear IV Regression,” Cambridge, MA: NBER Technical Working Paper 284. Summers, Lawrence H., 1992, "Investing in All the People," Pakistan Development Review 31:4 (Winter), 367-406. Summers, Lawrence H., 1994, "Investing in All the People: Educating Women in Developing Countries,” Washington, DC: World Bank, Economic Development Institute Seminar Paper No. 45. World Bank. 2001. Engendering Development: Gender Equality in Rights, Resources, and Voice, New York: Oxford University Press for the World Bank.

23

Figure 1 Sample sizes for residents and migrants – Women only Traceable, living in Guatemala

112 (10%) Died between 1969 and 2002

521 (57%) Original Villages

Migrants 1162 Subjects in 69-77 study

57 (5%) Untraceable

993 Traceable in 2002

919 (79%) Stayed in Guatemala

95 (10%) Nearby Villages 222 (24%) In or nearby Guatemala City

74 (6%) Left the Country

81 (9%) Elsewhere in Guatemala

24

Table 1a. Summary of G3 Outcomes Mean Regressions with G2 Intellectual Capital Stocks only¹ Anthropometry at Birth Birth weight (kg) 2.98 Birth length (cm) 48.22

SD

n

0.45 2.12

643 620

36-month Z-scores LAZ WAZ WLZ

-1.78 -1.24 -0.23

1.01 1.08 0.97

518 518 518

Schooling Years schooling²

-0.01

2.60

1215

Regressions with G2 Intellectual and Biological Capital Stocks¹ Anthropometry at Birth Birth weight (kg) 2.98 0.46 Birth length (cm) 48.24 2.13

573 553

36-month Z-scores LAZ WAZ WLZ

-1.79 -1.26 -0.25

1.02 1.09 0.96

456 456 456

Schooling Years schooling²

0.01

2.63

1162

Notes: 1 Summary chosen from specification with grades schooling as indicator G2 mothers' education/cognition 2 Difference in years schooling from age-cohort mean (positive if grades schooled > cohort mean)

25

Table 1b. Summary of explanatory variables and instruments, by G3 outcome

Explanatory variables G2 Mother's Intellectual Capital Stocks¹ Grades Formal Schooling Adult Nonverbal Cognitive Skills (Ravens score) Adult Reading Comprehension (SIA score) Adult Numeric Ability G2 Mother's Intellectual Capital Stocks² Grades Formal Schooling Adult Nonverbal Cognitive Skills (Ravens score) Adult Reading Comprehension (SIA score) Adult Numeric Ability G2 Mother's Nutritional Status³ Height(cm) at age 18 Other Controls³ G3 Gender (1=male) G3 Multiple Birth

Anthropometry at Birth* Mean SD N

Mean

3.37 9.50 26.80 31.38

2.63 3.48 21.09 8.59

643 647 645 645

3.52 9.36 26.49 31.10

2.73 3.68 21.55 8.91

3.38 9.60 26.54 31.21

2.70 3.44 21.10 8.62

573 574 572 572

3.47 9.46 25.92 30.85

149.78

5.38

573

0.50 0.01

0.50 0.12

0.49 0.41 0.37 0.48 0.40 10.48 0.46

Instruments: G2 Mother's Community Characteristics and Shocks³ Lived in communities at 0-36 months of age 0.39 Exposed to atole at 0-36 months of age 0.22 Born in San Juan 0.16 Born in Conacaste 0.36 Born in Espiritu Santo 0.20 Student-teacher ratio in community when G2 age 7 40.86 Local markets available when G2 age 15 0.69

Mean

Schooling SD

518 522 520 520

3.38 10.25 28.09 31.86

2.69 3.46 21.59 8.15

1215 1220 1215 1215

2.80 3.69 21.63 8.98

456 459 457 457

3.38 10.32 27.97 31.76

2.72 3.49 21.71 8.19

1162 1165 1160 1160

149.76

5.28

456

150.04

5.50

1162

573 573

0.52 0.01

0.50 0.10

456 456

0.51 0.01

0.50 0.11

1162 1162

573 573 573 573 573 573 573

0.39 0.22 0.18 0.31 0.24 40.77 0.66

0.49 0.41 0.38 0.46 0.43 10.82 0.47

456 456 456 456 456 456 456

0.30 0.19 0.25 0.33 0.16 42.93 0.69

0.46 0.39 0.43 0.47 0.37 10.94 0.46

1162 1162 1162 1162 1162 1162 1162 (continued)

26

36-month Z-scores SD N

N

Anthropometry at Birth* 36-month Z-scores Schooling Mean SD N Mean SD N Mean SD N Instruments: G2 Mother's Family & Individual Characteristics³ G1 Mother's schooling 1.09 1.47 573 1.11 1.48 456 1.09 1.57 1162 G1 Father's schooling 1.45 1.92 573 1.50 1.97 456 1.31 1.76 1162 Household wealth index in 1975 -2.91 0.88 573 -2.94 0.86 456 -2.85 0.89 1162 Missing G1 mother's schooling 0.01 0.08 573 0.01 0.09 456 0.01 0.12 1162 Missing G1 father's schooling 0.03 0.17 573 0.03 0.18 456 0.07 0.26 1162 Missing household wealth index in 1975 0.06 0.25 573 0.07 0.26 456 0.15 0.36 1162 Death of G1 mother or father b/f G2 reaches age 15 0.09 0.28 573 0.08 0.27 456 0.07 0.26 1162 Birth year 1969.61 4.23 573 1970.20 4.20 456 1967.40 3.71 1162 G2 is a twin 0.02 0.13 573 0.03 0.17 456 0.02 0.12 1162 Notes: * Summary statistics for anthropometry at birth are based on birth weight sample. See table 1a for birth length sample. 1 These These summary statistics are chosen from specification without height included. Note that summary statistics are at the child-level and thus can include multiple observations per G2 mother. 2 These summary statistics are chosen from specification with height included. Note that summary statistics are at the child-level and thus can include multiple observations per G2 mother. 3 These summary statistics are chosen from specification with mother's grades schooling and height as right hand side variables. Note that summary statistics are at the child-level and thus can include multiple observations per G2 mother.

27

Table 2a. Determinants of G3 Anthropometry at Birth Indicator of G2 Mother's Intellectual Capital G3 Outcome

Grades Formal Schooling Birth Weight (kg) Birth Length (cm) Coeff t/z Coeff t/z

Adult Nonverbal Cognitive Skills (Raven's Test) Birth Weight (kg) Birth Length (cm) Coeff t/z Coeff t/z

OLS: Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

0.002 0.070 -0.656

0.011 0.066 -0.648

F-test p-value n (# clusters)

8.43 0.000 643 (357)

OLS: Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock -0.007 G2 Mother's Height(cm) at age 18 0.023 G3 Gender (1=male) 0.080 G3 Multiple Birth -0.692 F-test p-value n (# clusters)

16.96 0.000 573 (325)

IV: G2 Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

-0.005 0.074 -0.754 23.58 0.000 13.92 0.057 643 (357)

0.21 1.95* -4.36***

-0.010 0.663 -3.031

-0.26 4.09*** -3.14***

10.91 0.000 620 (350)

-0.84 4.62*** 2.15** -5.02***

-0.044 0.106 0.674 -3.291

8.63 0.000 647 (360)

-1.09 4.73*** 3.99*** -3.47***

17.1 0.000 553 (318)

-0.31 2.14** -8.01***

-0.029 0.715 -4.048 28.41 0.000 13.09 0.187 620 (350)

1.93* 1.83* -4.19***

0.013 0.021 0.078 -0.671

0.017 0.065 -0.719 21.95 0.000 10.89 0.069 647 (360)

0.86 4.02*** -3.07***

10.33 0.000 624 (353)

2.25** 4.75*** 2.12** -4.91***

18.29 0.000 574 (326)

-0.44 4.69*** -7.38***

0.021 0.651 -2.995

0.024 0.099 0.677 -3.190

0.92 4.88*** 3.98*** -3.34***

17.38 0.000 554 (319)

1.64 1.88* -7.07***

0.050 0.683 -4.013 28.33 0.000 10.15 0.229 624 (353)

0.97 4.48*** -7.29***

(continued)

28

Indicator of G2 Mother's Intellectual Capital G3 Outcome

Grades Formal Schooling Birth Weight (kg) Birth Length (cm) Coeff t/z Coeff t/z

IV: G2 Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock -0.019 -1.24 G2 Mother's Height(cm) at age 18 0.029 2.55** G3 Gender (1=male) 0.080 2.28** G3 Multiple Birth -0.787 -6.92***

-0.039 0.067 0.723 -4.103

F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

18.06 0.000 5.08 0.299 553 (318)

14.52 0.000 5.22 0.172 573 (325)

-0.59 1.66* 4.66*** -6.2***

Adult Nonverbal Cognitive Skills (Raven's Test) Birth Weight (kg) Birth Length (cm) Coeff t/z Coeff t/z

0.029 0.026 0.078 -0.767 22.54 0.000 4.93 0.335 574 (326)

Note: *** indicates significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level

29

2.35** 2.6*** 2.17** -8.35***

0.075 0.052 0.709 -4.075 20.06 0.000 4.84 0.363 554 (319)

1.25 1.43 4.5*** -6.55***

Table 2b. Determinants of G3 Anthropometry at Birth Indicator of G2 Mother's Intellectual Capital G3 Outcome

Adult Reading Comprehension (SIA score) Birth Weight (kg) Birth Length (cm) Coeff t/z Coeff t/z

Adult Numeric Ability Birth Weight (kg) Birth Length (cm) Coeff t/z Coeff t/z

OLS: Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

0.001 0.066 -0.646

0.002 0.065 -0.648

F-test p-value n (# clusters)

7.90 0.000 645 (359)

OLS: Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock 0.000 G2 Mother's Height(cm) at age 18 0.022 G3 Gender (1=male) 0.079 G3 Multiple Birth -0.685 F-test p-value n (# clusters)

16.36 0.000 572 (325)

IV: G2 Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

0.003 0.067 -0.717

F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

28.64 0.000 8.71 0.076 645 (359)

0.99 1.84* -4.34***

0.003 0.647 -2.978

0.59 3.99*** -3.08***

9.59 0.000 622 (352)

-0.14 4.40*** 2.12** -4.94***

-0.002 0.102 0.674 -3.253

8.51 0.000 645 (359)

-0.34 4.45*** 3.98*** -3.42***

16.42 0.000 552 (318)

1.48 1.92* -8.41***

0.004 0.697 -3.952 27.97 0.000 8.35 0.237 622 (352)

0.66 1.8* -4.4***

-0.001 0.022 0.079 -0.686

0.004 0.065 -0.729 24.32 0.000 6.93 0.063 645 (359)

0.08 3.99*** -3.19***

10.80 0.000 622 (352)

-0.18 4.45*** 2.13** -5.04***

16.62 0.000 572 (325)

0.46 4.61*** -7.23***

0.001 0.646 -3.014

-0.010 0.103 0.680 -3.298

-0.74 4.68*** 4.03*** -3.59***

17.18 0.000 552 (318)

0.63 1.85* -7.87***

0.000 0.706 -4.034

0.00 4.63*** -7.48***

28.77 0.000 6.56 0.219 622 (352) (continued)

30

Indicator of G2 Mother's Intellectual Capital G3 Outcome

Adult Reading Comprehension (SIA score) Birth Weight (kg) Birth Length (cm) Coeff t/z Coeff t/z

IV: G2 Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock 0.000 0.06 G2 Mother's Height(cm) at age 18 0.023 1.82* G3 Gender (1=male) 0.080 2.28** G3 Multiple Birth -0.759 -6.65***

-0.001 0.055 0.714 -4.055

F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

17.03 0.000 4.18 0.337 552 (318)

13.54 0.000 4.20 0.145 572 (325)

-0.06 1.19 4.58*** -6.04***

Adult Numeric Ability Birth Weight (kg) Birth Length (cm) Coeff t/z Coeff t/z

-0.002 0.025 0.081 -0.776 14.15 0.000 4.90 0.149 572 (325)

Note: *** indicates significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level

31

-0.36 2.19** 2.28** -6.85***

-0.011 0.064 0.731 -4.199 20.59 0.000 4.86 0.300 552 (318)

-0.37 1.40 4.68*** -6.84***

Table 3a. Determinants of 36 month Z-scores Indicator of G2 Mother's Intellectual Capital G3 Outcome

LAZ Coeff

t/z

OLS: Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

0.058 -0.116 -0.405

3.01*** -1.36 -0.78

F-test p-value n (# clusters)

4.08 0.007 518 (330)

17.39 0.000 456 (294)

IV: G2 Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

0.115 -0.116 -0.539

F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

4.63 0.004 10.88 0.007 518 (330)

0.69 7.93*** -0.96 -2.13**

1.95* 0.09 0.11

0.010 0.060 0.046 -0.010

3.44*** -1.39 -1.15

0.117 0.063 0.645

0.46 5.62*** 0.46 -0.01

6.90 0.000 3.75 0.174 456 (294)

0.036 0.120 0.082 -0.094 5.73 0.000 3.75 0.162 456 (294)

0.002 0.019 0.450

0.09 0.23 0.55

0.001 0.010 0.036 0.466

0.08 0.98 0.40 0.60

0.42 0.791 456 (294)

3.40*** 0.68 0.70

4.22 0.006 10.88 0.009 518 (330)

-0.31 5.01*** -0.85 -1.74*

WLZ t/z

0.12 0.948 518 (330)

8.89 0.000 456 (294)

IV: G2 Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock -0.011 G2 Mother's Height(cm) at age 18 0.133 G3 Gender (1=male) -0.079 G3 Multiple Birth -0.666 F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

0.042 0.009 0.110 1.29 0.279 518 (330)

OLS: Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock 0.013 G2 Mother's Height(cm) at age 18 0.075 G3 Gender (1=male) -0.084 G3 Multiple Birth -0.599 F-test p-value n (# clusters)

Grades Formal Schooling WAZ Coeff t/z Coeff

0.070 0.008 0.449

2.32** 0.09 0.6

1.84 0.141 10.88 0.042 518 (330)

0.91 3.92*** 0.80 -0.21

0.056 0.036 0.018 0.287

1.66* 1.42 0.21 0.44

2.28 0.060 3.75 0.202 456 (294) (continued)

32

Table 3a (continued) Indicator of G2 Mother's Intellectual Capital G3 Outcome

Adult Nonverbal Cognitive Skills (Raven's Test) LAZ WAZ WLZ Coeff t/z Coeff t/z Coeff t/z

OLS: Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

0.006 -0.114 -0.516

F-test p-value n (# clusters)

0.9 0.441 522 (332)

0.4 -1.32 -0.97

OLS: Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock 0.006 G2 Mother's Height(cm) at age 18 0.079 G3 Gender (1=male) -0.082 G3 Multiple Birth -0.629 18.57 0.000 459 (295)

IV: G2 Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

-0.047 -0.073 -0.468

F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

1.48 0.219 7.23 0.001 522 (332)

-0.007 0.063 0.058 -0.040

-0.027 0.072 0.678

-0.42 6.11*** 0.58 -0.06

0.026 0.138 0.065 -0.164

-0.016 0.011 0.053 0.454

-0.85 0.77 0.76

0.001 0.016 0.279

-1.28 1.23 0.59 0.62

0.05 0.20 0.37

0.06 0.979 7.23 0.006 522 (332)

0.7 4.93*** 0.63 -0.41

0.018 0.058 0.015 0.040

F-test 7.28 6.12 1.77 p-value 0.000 0.000 0.136 Weak ID: Cragg Donald F-test 3.32 3.32 3.32 Overid: Hanson J statistic p-value 0.220 0.147 0.138 n (# clusters) 459 (295) 459 (295) 459 (295) Note: *** indicates significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level

33

-1.15 0.46 0.56

0.91 0.456 459 (295)

0.56 0.643 7.23 0.001 522 (332)

-1.01 4.84*** -0.6 -1.62

-0.015 0.039 0.442 0.63 0.597 522 (332)

9.42 0.000 459 (295)

-1.45 -0.86 -1.06

IV: G2 Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock -0.034 G2 Mother's Height(cm) at age 18 0.124 G3 Gender (1=male) -0.057 G3 Multiple Birth -0.619

-0.41 0.25 0.03

0.08 0.973 522 (332)

0.4 8.15*** -0.93 -2.31**

F-test p-value n (# clusters)

-0.006 0.024 0.025

0.56 2.65*** 0.18 0.07

Table 3b. Determinants of 36 month Z-scores Indicator of G2 Mother's Intellectual Capital G3 Outcome Coeff OLS: Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

0.011 -0.138 -0.340

F-test p-value n (# clusters)

8.24 0.000 520 (331)

Adult Reading Comprehension (SIA score) LAZ WAZ WLZ t/z Coeff t/z Coeff t/z

4.67*** -1.60 -0.69

19.43 0.000 457 (294)

IV: G2 Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

0.017 -0.125 -0.549

F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

4.93 0.002 7.19 0.006 520 (331)

2.64*** 7.60*** -1.13 -1.83*

0.007 0.056 0.034 0.076

3.43*** -1.50 -1.24

0.021 0.036 0.496

2.45** 5.34*** 0.34 0.11

6.80 0.000 3.22 0.246 457 (294)

0.009 0.102 0.066 -0.011 6.23 0.000 3.22 0.209 457 (294)

1.59 0.29 0.63

0.003 0.008 0.034 0.520

1.45 0.78 0.38 0.68

1.07 0.371 457 (294)

4.21*** 0.38 0.56

6.27 0.000 7.19 0.036 520 (331)

-0.71 4.87*** -0.80 -1.84*

0.003 0.024 0.502 0.96 0.411 520 (331)

11.03 0.000 457 (294)

IV: G2 Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock -0.004 G2 Mother's Height(cm) at age 18 0.134 G3 Gender (1=male) -0.077 G3 Multiple Birth -0.700 F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

3.96*** -0.03 0.2

5.33 0.001 520 (331)

OLS: Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock 0.006 G2 Mother's Height(cm) at age 18 0.072 G3 Gender (1=male) -0.099 G3 Multiple Birth -0.529 F-test p-value n (# clusters)

0.010 -0.003 0.192

0.015 -0.005 0.528

3.49*** -0.07 0.74

4.11 0.007 7.19 0.259 520 (331)

1.61 3.13*** 0.64 -0.02

0.014 0.012 -0.003 0.496

2.69*** 0.41 -0.03 0.72

3.41 0.010 3.22 0.393 457 (294) (continued)

34

Table 3b. (continued) Indicator of G2 Mother's Intellectual Capital G3 Outcome Coeff OLS: Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth F-test p-value n (# clusters)

LAZ t/z

0.024 -0.146 -0.303 7.38 0.000 520 (331)

OLS: Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock 0.013 G2 Mother's Height(cm) at age 18 0.074 G3 Gender (1=male) -0.100 G3 Multiple Birth -0.505 F-test p-value n (# clusters)

19.71 0.000 457 (294)

IV: G2 Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock G3 Gender (1=male) G3 Multiple Birth

0.040 -0.158 -0.567

F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

4.41*** -1.70* -0.62

2.52** 7.62*** -1.14 -1.66*

2.96*** -1.87* -1.29

4.34 0.005 5.47 0.003 520 (331)

Adult Numeric Ability WAZ Coeff t/z Coeff

0.024 -0.014 0.246

3.99*** -0.14 0.26

WLZ t/z

0.010 0.018 0.538

5.43 0.001

1.35 0.257

520 (331)

520 (331)

0.017 0.057 0.030 0.123

2.63*** 5.46*** 0.30 0.18

0.010 0.007 0.031 0.558

12.03 0.000

1.42 0.229

457 (294)

457 (294)

0.053 0.000 0.531

4.57*** 0.00 0.64

0.039 -0.004 0.687

7.22 0.000 5.47 0.017

3.59 0.014 5.47 0.288

520 (331)

520 (331)

1.94* 0.22 0.69

1.80* 0.78 0.34 0.75

3.25*** -0.05 1.02

IV: G2 Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock G2 Mother's Height(cm) at age 18

-0.012 0.133

-0.82 4.98***

0.022 0.111

1.49 3.60***

0.035 0.026

2.56*** 0.94

G3 Gender (1=male) G3 Multiple Birth

-0.068 -0.762

-0.7 -1.96**

0.058 -0.039

0.56 -0.10

0.018 0.689

0.20 1.17

F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value

6.87 5.94 3.15 0.000 0.000 0.015 3.62 3.62 3.62 0.252 0.174 0.431 457 n (# clusters) (294) 457 (294) 457 (294) Note: *** indicates significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level

35

Table 4. Determinants of G3 Schooling 1 Indicator of G2 Mother's Intellectual Capital

Grades Formal Schooling Coeff t/z

OLS: Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock 0.138 G3 Gender (1=male) 0.292 G3 Multiple Birth -1.623 F-test p-value n (# clusters)

4.77*** 1.88* -2.38**

9.86 0.000 1215 (507)

Adult Nonverbal Cognitive Skills (Raven's Test) Coeff t/z

Adult Reading Comprehension (SIA score) Coeff t/z

Adult Numeric Ability Coeff t/z

0.069 0.262 -1.511

0.018 0.306 -1.592

0.049 0.272 -1.590

5.16 0.002 1220 (510)

OLS: Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock 0.119 3.87*** G2 Mother's Height(cm) at age 18 0.040 2.45** G3 Gender (1=male) 0.324 2.01** G3 Multiple Birth -1.753 -2.54**

0.058 0.052 0.299 -1.683

F-test p-value n (# clusters)

6.49 0.000 1165 (481)

8.58 0.000 1162 (479)

IV: G2 Intellectual Capital Stocks Only G2 Mother's Intellectual Capital Stock 0.194 G3 Gender (1=male) 0.227 G3 Multiple Birth -1.550 F-test p-value Weak ID: Cragg Donald F-test Overid: Hanson J statistic p-value n (# clusters)

3.12*** 1.68* -1.97**

3.40*** 1.49 -2.40**

5.45 0.001 22.66 0.458 1215 (507)

IV: G2 Intellectual Capital Stocks and Nutritional Status G2 Mother's Intellectual Capital Stock 0.172 2.70*** G2 Mother's Height(cm) at age 18 0.135 2.60*** G3 Gender (1=male) 0.353 2.11** G3 Multiple Birth -2.061 -2.87***

0.049 0.192 -1.365

9.86 0.000 1215 (508)

2.51** 3.20*** 1.85* -2.23**

0.016 0.035 0.331 -1.711

0.89 1.26 -1.84*

0.024 0.312 -1.510

3.93*** 2.09** 2.06** -2.44**

0.020 0.088 0.345 -1.858

2.77*** 2.04** -2.28**

4.30*** 2.38** 1.90* -2.51**

0.058 0.244 -1.473

2.60*** 1.62 -2.28**

4.18 0.006 13.98 0.282 1215 (508)

1.90* 1.47 2.11** -2.58**

F-test 5.38 3.51 4.23 p-value 0.000 0.008 0.002 Weak ID: Cragg Donald F-test 6.91 7.11 4.59 Overid: Hanson J statistic p-value 0.943 0.564 0.689 n (# clusters) 1162 (479) 1165 (481) 1160 (479) Notes: 1 difference in years schooling from age-cohort mean (positive if grades schooled > cohort mean) *** indicates significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level

36

0.043 0.040 0.304 -1.726 9.75 0.000 1160 (479)

4.63 0.003 16.08 0.492 1215 (508)

0.42 3.20*** 1.82* -2.54**

5.43*** 1.77* -2.33**

11.62 0.000 1215 (508)

8.81 0.000 1160 (479)

1.80 0.146 11.54 0.098 1220 (510)

0.028 0.163 0.305 -1.962

4.86** 1.98** -2.29**

0.044 0.123 0.313 -1.955 4.15 0.003 5.68 0.652 1160 (479)

1.60 2.24** 1.90* -2.69***

Appendix Table 1: First stage results for select G3 outcomes: endogenous G2 Grades Schooling and Height at age 18 Weight at Birth (n=573, clusters=325) Schooling Height Coeff t Coeff t

36-month LAZ (n=456, clusters=294) Schooling Height Coeff t Coeff t

Years Schooling¹ (n=1162, clusters=479) Schooling Height Coeff t Coeff t

Included instruments: G3 Gender (1=male) G3 Multiple Birth

0.104 -0.130

0.63 -0.22

0.327 2.677

0.79 1.09

0.454 -0.901

2.05** -1.93*

0.206 2.327

0.43 0.66

-0.108 0.449

-0.74 0.54

-0.195 3.228

-0.64 1.91*

Excluded instruments: Lived in communities at 0-36 months of age Exposed to atole at 0-36 months of age Born in San Juan Born in Conacaste Born in Espiritu Santo Student-teacher ratio in community when G2 age 7 Local markets available when G2 age 15

-0.603 0.882 -0.498 -1.010 -0.516 -0.023 -0.8073

-1.36 1.46 -0.85 -2.29** -0.67 -1.16 -1.50

-1.112 0.356 0.972 2.404 -0.141 -0.008 0.076

-1.15 0.26 0.65 2.42** -0.09 -0.22 0.06

-0.313 0.726 -0.325 -0.298 -0.206 -0.035 -1.083

-0.65 1.11 -0.50 -0.60 -0.24 -1.61 -1.70*

-1.091 0.688 0.371 1.988 -0.082 0.004 -0.077

-1.17 0.50 0.23 2.05** -0.05 0.12 -0.05

-0.468 1.360 -1.111 -1.221 -0.473 -0.022 -0.593

-1.07 2.58*** -2.41** -3.25*** -0.75 -1.67* -1.61

-0.533 -0.198 0.307 1.871 -1.079 0.000 -0.636

-0.54 -0.17 0.29 2.16** -0.85 -0.01 -0.82

G1 Mother's schooling G1 Father's schooling Household wealth index in 1975 Missing G1 mother's schooling Missing G1 father's schooling Missing household wealth index in 1975 Death of G1 mother or father b/f G2 reaches age 15 Birth year G2 is a twin

0.293 0.199 0.949 1.277 -1.044 0.954 -0.926 0.011 0.079

2.80*** 2.30** 4.35*** 2.16** -1.84* 1.48 -1.99** 0.29 0.08

0.556 0.011 0.709 -1.780 1.119 -1.147 -1.069 0.138 -5.244

2.49** 0.08 1.94* -0.84 0.71 -1.14 -0.80 1.92* -1.99**

0.327 0.254 1.033 0.783 -1.008 0.985 -0.916 -0.020 0.452

3.12*** 2.66*** 4.53*** 1.34 -1.85* 1.64 -1.94* -0.48 0.62

0.574 -0.080 0.736 -1.155 1.643 -0.204 -2.020 0.158 -3.892

2.53** -0.56 1.90* -0.73 1.26 -0.23 -1.67* 2.00** -1.94*

0.256 0.197 0.739 1.711 -0.561 1.131 -1.474 0.010 0.269

2.75*** 2.81*** 4.19*** 1.79* -1.06 2.77*** -3.26 0.30 0.29

0.510 -0.077 0.805 1.635 -0.472 -1.127 -0.417 0.060 -3.866

2.42** -0.51 2.16** 1.62 -0.45 -1.34 -0.35 0.82 -2.10**

stat 3.18 0.128

P 0.000

stat 9.27 0.239

p 0.000

stat 2.25 0.091

p 0.004

stat p stat P stat P Tests of excluded instruments: F-test of excluded instruments 9.51 0.000 3.20 0.000 7.71 0.000 Partial R2 of excluded instruments 0.278 0.133 0.261 Notes: 1 difference in years schooling from age-cohort mean (positive if grades schooled > cohort mean) *** indicates significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level

37

Appendix Table 2: First stage results for select G3 outcomes: endogenous G2 Nonverbal Cognitive Skills (Ravens score) and Height at age 18 Weight at Birth (n=574, clusters=326) Ravens Height Coeff t Coeff t

36-month LAZ (n=459, clusters=295) Ravens Height Coeff T Coeff t

Years Schooling¹ (n=1165, clusters=481) Ravens Height Coeff t Coeff t

Included instruments: G3 Gender (1=male) G3 Multiple Birth

0.317 -0.228

1.26 -0.21

0.298 2.617

0.72 1.08

0.531 -0.071

1.67* -0.04

0.232 2.096

0.48 0.62

0.182 -1.307

1.08 -1.49

-0.187 3.233

-0.62 1.91*

Excluded instruments: Lived in communities at 0-36 months of age Exposed to atole at 0-36 months of age Born in San Juan Born in Conacaste Born in Espiritu Santo Student-teacher ratio in community when G2 age 7 Local markets available when G2 age 15

-0.764 1.163 -0.500 -0.703 0.810 -0.028 1.584

-1.30 1.52 -0.66 -1.13 0.79 -1.25 2.28**

-1.120 0.427 0.980 2.361 -0.241 -0.009 -0.040

-1.16 0.32 0.66 2.38** -0.15 -0.23 -0.03

0.256 -0.140 0.137 -0.188 1.682 -0.041 1.327

0.33 -0.15 0.16 -0.28 1.49 -1.68* 1.71*

-1.127 0.928 0.358 1.882 -0.430 0.002 -0.486

-1.20 0.67 0.22 1.93* -0.24 0.05 -0.34

0.492 0.470 -1.056 -1.447 -1.672 0.009 0.484

0.93 0.71 -1.65* -2.70*** -2.17** 0.60 1.03

-0.559 -0.128 0.351 1.839 -1.130 -0.002 -0.711

-0.57 -0.11 0.33 2.12** -0.89 -0.05 -0.91

G1 Mother's schooling G1 Father's schooling Household wealth index in 1975 Missing G1 mother's schooling Missing G1 father's schooling Missing household wealth index in 1975 Death of G1 mother or father b/f G2 reaches age 15 Birth year G2 is a twin

0.283 0.091 0.176 1.520 -0.084 1.012 1.066 -0.292 0.224

1.95* 0.81 0.69 2.14** -0.11 1.38 1.44 -7.09*** 0.19

0.553 0.004 0.718 -1.786 1.124 -1.156 -1.073 0.140 -5.322

2.48** 0.03 1.96* -0.84 0.71 -1.15 -0.81 1.94* -2.02**

0.336 0.143 -0.026 4.778 0.465 0.837 0.073 -0.319 0.120

2.28** 1.04 -0.09 2.27** 0.49 1.37 0.09 -6.56*** 0.10

0.576 -0.101 0.756 -1.170 1.633 -0.226 -2.024 0.161 -4.069

2.54** -0.71 1.94* -0.73 1.24 -0.25 -1.68* 2.02** -2.00**

0.384 0.161 -0.241 1.504 -0.875 1.042 0.713 -0.196 -0.619

3.46*** 1.50 -1.09 1.24 -1.53 1.68* 1.07 -4.66*** -0.74

0.509 -0.082 0.816 1.623 -0.481 -1.149 -0.439 0.063 -3.911

2.41** -0.54 2.19** 1.60 -0.46 -1.36 -0.37 0.86 -2.11**

stat 3.21 0.129

p 0.000

stat 4.56 0.126

p 0.000

stat 2.25 0.091

p 0.004

stat P stat p stat P Tests of excluded instruments: F-test of excluded instruments 8.83 0.000 3.21 0.000 4.11 0.000 Partial R2 of excluded instruments 0.195 0.133 0.180 Notes: 1 difference in years schooling from age-cohort mean (positive if grades schooled > cohort mean) *** indicates significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level

38

Appendix Table 3: First stage results for select G3 outcomes: endogenours G2 Reading Comprehension (SIA score) and Height at age 18 Weight at Birth (n=572, clusters=325) SIA Height Coeff T Coeff t

36-month LAZ (n=457, clusters=294) SIA Height Coeff t Coeff t

Years Schooling¹ (n=1160, clusters=479) SIA Height Coeff T Coeff t

Included instruments: G3 Gender (1=male) G3 Multiple Birth

1.569 -3.448

1.08 -0.47

0.283 2.641

0.68 1.08

4.098 -9.148

2.33** -1.96*

0.218 2.134

0.45 0.62

-0.317 2.547

-0.27 0.43

-0.192 3.244

-0.63 1.92*

Excluded instruments: Lived in communities at 0-36 months of age Exposed to atole at 0-36 months of age Born in San Juan Born in Conacaste Born in Espiritu Santo Student-teacher ratio in community when G2 age 7 Local markets available when G2 age 15

-2.884 6.141 -3.627 -1.117 -7.889 -0.167 -6.074

-0.85 1.29 -0.65 -0.30 -1.28 -1.11 -1.34

-1.132 0.443 1.022 2.353 -0.133 -0.010 0.049

-1.17 0.33 0.68 2.37** -0.08 -0.25 0.04

1.050 1.717 -3.300 3.073 -12.285 -0.146 -10.520

0.27 0.33 -0.54 0.76 -1.79* -0.93 -1.95*

-1.143 0.955 0.407 1.867 -0.311 0.001 -0.387

-1.21 0.69 0.25 1.92* -0.16 0.03 -0.25

-0.692 8.532 -5.210 0.355 -5.826 -0.178 -4.402

-0.20 2.00** -1.28 0.10 -1.18 -1.59 -1.44

-0.573 -0.102 0.374 1.830 -1.045 -0.002 -0.649

-0.58 -0.09 0.35 2.11** -0.81 -0.07 -0.81

G1 Mother's schooling G1 Father's schooling Household wealth index in 1975 Missing G1 mother's schooling Missing G1 father's schooling Missing household wealth index in 1975 Death of G1 mother or father b/f G2 reaches age 15 Birth year G2 is a twin

2.641 1.004 7.127 15.311 -2.130 1.679 2.868 -0.331 -2.503

3.53*** 1.71* 4.52*** 2.47** -0.26 0.35 0.70 -1.10 -0.25

0.556 0.003 0.722 -1.789 0.905 -1.148 -1.066 0.141 -5.336

2.49** 0.02 1.97** -0.84 0.51 -1.14 -0.80 1.97** -2.01**

2.553 1.260 7.590 23.932 -5.053 0.273 -1.387 -0.680 3.273

3.26*** 1.86* 4.78*** 4.37*** -0.76 0.07 -0.31 -2.00** 0.49

0.579 -0.102 0.757 -1.077 1.433 -0.214 -2.017 0.162 -4.077

2.55** -0.72 1.95* -0.65 0.97 -0.24 -1.68* 2.03** -2.00**

2.584 1.193 5.174 7.633 -5.726 7.082 -5.189 -0.456 -0.699

4.27*** 2.25** 4.21** 1.16 -1.11 2.12** -1.21 -1.55 -0.09

0.512 -0.082 0.822 1.636 -0.553 -1.135 -0.420 0.064 -3.911

2.43** -0.54 2.21** 1.62 -0.51 -1.34 -0.36 0.88 -2.10**

stat 3.14 0.128

p 0.000

stat 6.90 0.187

P 0.000

stat 2.23 0.091

P 0.004

stat P stat p stat P Tests of excluded instruments: F-test of excluded instruments 7.16 0.000 3.20 0.000 9.23 0.000 Partial R2 of excluded instruments 0.217 0.132 0.213 Notes: 1 difference in years schooling from age-cohort mean (positive if grades schooled > cohort mean) *** indicates significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level

39

Appendix Table 4: First stage results for select G3 outcomes: endogenous G2 Numeric Ability and Height at age 18 Weight at Birth 36-month LAZ (n=572, clusters=325) (n=457, clusters=294) Math Height Math Height Coeff T Coeff t Coeff T Coeff T

Years Schooling¹ (n=1160, clusters=479) Math Height Coeff t Coeff t

Included instruments: G3 Gender (1=male) G3 Multiple Birth

0.875 -3.407

1.41 -0.95

0.283 2.641

0.68 1.08

1.610 -7.815

2.03** -2.42**

0.218 2.134

0.45 0.62

0.507 -0.152

1.12 -0.07

-0.192 3.244

-0.63 1.92*

Excluded instruments: Lived in communities at 0-36 months of age Exposed to atole at 0-36 months of age Born in San Juan Born in Conacaste Born in Espiritu Santo Student-teacher ratio in community when G2 age 7 Local markets available when G2 age 15

-0.339 1.442 -3.278 -2.610 -5.442 0.014 -4.147

-0.20 0.68 -1.50 -1.79* -2.23** 0.27 -2.58**

-1.132 0.443 1.022 2.353 -0.133 -0.010 0.049

-1.17 0.33 0.68 2.37** -0.08 -0.25 0.04

0.644 0.353 -3.416 -0.987 -7.350 0.027 -6.483

0.34 0.15 -1.39 -0.60 -2.52** 0.47 -2.92***

-1.143 0.955 0.407 1.867 -0.311 0.001 -0.387

-1.21 0.69 0.25 1.92* -0.16 0.03 -0.25

-0.602 3.642 -4.402 -3.050 -3.995 -0.020 -3.242

-0.38 1.98** -2.72*** -2.62*** -2.17** -0.55 -3.28***

-0.573 -0.102 0.374 1.830 -1.045 -0.002 -0.649

-0.58 -0.09 0.35 2.11** -0.81 -0.07 -0.81

G1 Mother's schooling G1 Father's schooling Household wealth index in 1975 Missing G1 mother's schooling Missing G1 father's schooling Missing household wealth index in 1975 Death of G1 mother or father b/f G2 reaches age 15 Birth year G2 is a twin

0.973 0.270 1.878 5.757 -3.500 0.066 1.100 -0.160 -4.227

3.17*** 1.10 2.99*** 3.04*** -0.89 0.03 0.62 -1.18 -1.42

0.556 0.003 0.722 -1.789 0.905 -1.148 -1.066 0.141 -5.336

2.49** 0.02 1.97** -0.84 0.51 -1.14 -0.80 1.97** -2.01**

1.031 0.401 1.919 10.844 0.228 0.230 -0.626 -0.298 -1.589

3.05*** 1.45 2.78*** 3.54*** 0.06 0.11 -0.32 -1.97** -0.69

0.579 -0.102 0.757 -1.077 1.433 -0.214 -2.017 0.162 -4.077

2.55** -0.72 1.95* -0.65 0.97 -0.24 -1.68* 2.03** -2.00**

0.755 0.333 1.489 3.078 -4.039 2.465 -1.120 -0.176 -1.542

3.25*** 1.71* 3.60*** 0.95 -1.72* 1.93* -0.78 -1.52 -0.62

0.512 -0.082 0.822 1.636 -0.553 -1.135 -0.420 0.064 -3.911

2.43** -0.54 2.21** 1.62 -0.51 -1.34 -0.36 0.88 -2.10**

stat 3.14 0.128

P 0.000

stat 5.50 0.156

P 0.000

stat 2.23 0.091

P 0.004

P stat P stat P stat Tests of excluded instruments: F-test of excluded instruments 6.87 0.000 3.20 0.000 4.59 0.000 Partial R2 of excluded instruments 0.166 0.132 0.155 Notes: 1 difference in years schooling from age-cohort mean (positive if grades schooled > cohort mean) *** indicates significance at the 1 percent level, ** at the 5 percent level, and * at the 10 percent level

40