Policy Research Working Paper 9355 Mama Knows (and Does) Best Maternal Schooling Opportunities and Child Development in Indonesia Amer Hasan Nozomi Nakajima Marcos A. Rangel Education Global Practice August 2020 Policy Research Working Paper 9355 Abstract This paper leverages quasi-experimental variation in effects are particularly salient at the bottom of the distribu- increased access to basic formal education, introduced by tions of outcomes. Drawing on insights from economics, a large-scale school construction program in Indonesia in psychology, and sociology, the paper examines pathways for the 1970s, to quantify the benefits to the children of women these impacts. Evidence suggests that mothers who were targeted by the program. Novel and rich data allow the exposed to more schooling opportunities during childhood analysis of a range of health, cognitive and socio-emotional demonstrate less hostility toward their children when par- development outcomes for children ages 6 to 8 in 2013. enting and also invest more in their children’s preschool The paper finds that increased maternal access to schooling education. has positive and multidimensional effects on children. The This paper is a product of the Education Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at ahasan1@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Mama Knows (and Does) Best: Maternal Schooling Opportunities and Child Development in Indonesia∗ A MER H ASAN N OZOMI NAKAJIMA M ARCOS A. R ANGEL Education Global Practice Education Policy Sanford School World Bank and Program Evaluation of Public Policy Harvard University Duke University JEL Codes: I28; O15; I21; J13; I26; D10 Keywords: maternal education, school construction, human capital, child outcomes, par- enting ∗ Authors are listed alphabetically. Corresponding author: Rangel (marcos.rangel@duke.edu). We would like to thank the Strategic Impact Evaluation Fund (TF0A0234) for their generous support. Dedy Junaedi, Upik Sabain- ingrum, Anas Sutisna, Lulus Kusbudiharjo and Mulyana were instrumental in managing the fieldwork. Esther Duflo graciously shared the data on the INPRES school construction programs. The findings, interpretations, and conclu- sions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. 1 Introduction Barriers to schooling have been one of the main obstacles to development across the globe. While the motivation for expanding access to schooling often relies on labor productivity gains, there are reasons to believe that a large portion of returns to such policies materialize within families and across generations. Estimating these spillover effects of education is particularly relevant not only because it affects the accounting of social returns to public investments in education, but because it may influence the design of contemporaneous policies targeting early childhood development, like those focusing on parenting and school readiness. The literature on intergenerational impacts of education is large and very active (see reviews in Holmlund, Lindahl, and Plug (2011) and Haveman and Wolfe (1995)),1 but also relatively scarce for developing-country contexts where educational attainment levels are far lower.2 In this paper we take steps towards filling this gap by examining the intergenerational impacts of one of the most prominent schooling-expansion interventions in the developing world, the Indonesia Sekolah Dasar INPRES program of the 1970s. We focus on the children of women originally targeted by the INPRES program and who re- side in rural areas of Indonesia by the early 2010s. We focus on the expansion of maternal educa- tion given the gendered division of labor towards childrearing often found in developing country settings like those of Indonesia. Our analyses capitalize on a novel and comprehensive data set on early childhood development that is representative of rural Indonesia, allowing us to examine the impacts of the INPRES program over multiple developmental outcomes for children born to women targeted by the school construction program. Our outcomes encompass effects over health as well as cognitive and socio-emotional skills. In this way we complement and expand on re- 1 The increased accessibility to Scandinavian registry data has led to an explosion of articles on the topic. Some of the most influential work includes the seminal article by S. E. Black, Devereux, and Salvanes (2005) and more recent iterations which exploit policy-induced variation in educational attainment among older generations such as Fischer, Karlsson, Nilsson, and Schwarz (2019) and Suhonen and Karhunen (2019). Chevalier (2004) and Dickson, Gregg, and Robinson (2016) conduct similar analysis using minimum school-age regulations in the U.K., while Maurin and McNally (2008) do so employing variation generated by the French student revolt of 1968. Work focused on the intergenerational impacts of education in the U.S. context using causal inference reasoning can be found in Currie and Moretti (2003), Oreopoulos, Page, and Stevens (2006), Carneiro, Meghir, and Parey (2013). 2 A notable exception is Andrabi, Das, and Khwaja (2012), who focus on Pakistani women with very low levels of education and their children. 2 cent studies by Akresh, Halim, and Kleemans (2018) and Mazumder, Rosales-Rueda, and Triyana (2019), which are also based on the same program. Departing from the literature that precedes us and these concurrent articles, we examine a broader set of developmental indicators as well as provide evidence regarding the role played by parenting practices on the transmission of policy im- pacts across generations.3 We do so both in terms of activities parents carry out with their children but also in terms of interactions parents have with their children. Our analyses parallel those that focus on time use data, except that instead of focusing on the amount of time spent with the child (Andrabi et al., 2012; Bono, Francesconi, Kelly, & Sacker, 2016; Kalil, Ryan, & Corey, 2012) we focus on also characterizing the quality of parent-child interactions. Our identification strategy follows Esther Duflo’s seminal study on the causal impacts of IN- PRES (Duflo, 2001) by leveraging the timing and geographic (district/kabupaten) differences in intensity of this major school construction project launched in 1973 by the Indonesian government. INPRES led to the construction of 61,807 new primary schools within five years of its launching (Breierova & Duflo, 2004), corresponding to approximately one additional school per 500 Indone- sian children on aggregate. With this historic context in mind, we use the caregiver survey module collected in 2013 as part of the Indonesia Early Childhood Education and Development (ECED) Project. Importantly, the survey contains detailed information about mothers of young children and captures where mothers were born and whether they had moved from their place of birth be- fore completing primary schooling. As such, we are able to locate where the mothers in our data set grew up and characterize whether they potentially benefited from the expansion of primary education as a result of the INPRES program. Our empirical estimates confirm that the INPRES program produced significant gains in edu- cational attainment among the mothers in our sample. Mothers who grew up as part of cohorts in districts that were exposed to more intense doses of schooling expansion received an additional 1.3 years of formal education. These effects are particularly strong at lower levels of schooling, with 3 The literature beyond economics establishes parenting practices as a critical mechanism by which parents influ- ence their childrens development (Dornbusch, Ritter, Leiderman, Roberts, & Fraleigh, 1987; Jenkins & Handa, 2019; Kohn, 1969). 3 sizeable increases in rates of primary education completion (14 percentage points (p.p.)) and of literacy (15 p.p.). When looking at the children of these women, reduced-form estimations indicate a significant relationship between the intensity of a mother’s exposure to the program during child- hood and her offspring’s shorter- and longer-term health outcomes. Extreme stunting and wasting rates decrease by 2.7 p.p. and 3.9 p.p. respectively, but morbidity (flu/fever/stomach-ache) rates in the last two weeks are not affected. We also find a positive relationship between the intensity of a mother’s exposure to the program during childhood and her own child’s educational achievement in language: 0.14 standard deviation (s.d.). These results are particularly strong at the lower end of the distribution of performances in language tests. However, we do not find this relationship in mathematics nor in cognitive reasoning. When we unpack the test into its component modules - which are arranged by difficulty - we find evidence to support the contention that children of mothers exposed to INPRES do better on the items focusing on more basic math skills (ordering quantities). Importantly, these children also outperform their peers in terms of pro-social behavior (0.18 s.d.) and social competence (0.22 s.d.). We examine several of the usual pathways posited in the literature through which this relation- ship may operate and find evidence of an overall marginally significant improvement in resources in the form of assortative mating (more educated partners) and increased household wealth as cen- tral channels, but no evidence of impacts via reduced fertility (quantity-quality trade-off).4 Above all, and in a novel result, we find that mothers exposed to more education opportunities exhibit different child-caring patterns compared to mothers who were not. In particular, we find evidence that mothers who are exogenously exposed to more intensive school construction investments use less hostile parenting practices and are also more likely to send their children to pre-schools. We argue, therefore, that expansion of maternal education not only leads to increased time and fi- nancial investment in their children but also a more effective use of a mother’s child-rearing time and better child-rearing practices. In doing so, we contribute empirical evidence to the small but growing literature on the economics of parenting (Doepke, Sorrenti, & Zilibotti, 2019; Doepke & 4 Since we cannot rule out that partners were directly affected by INPRES policy, all our results are based on reduced-form (intent-to-treat) reasoning. 4 Zilibotti, 2017).5 We also contribute to a broader literature across the social sciences about the importance of early childhood and the long-term impact of parenting practices (Schady, Galiani, & Souza, 2006). For example, the Lancet series on child development assessed a variety of early interventions to improve equity in child development. These reviews find that improvements in parenting practices (such as more responsiveness in feeding infants and young children; encourag- ing learning, book reading, play activities; using positive discipline; and problem-solving related to children’s development, care, and feeding) had positive effects on development (M. M. Black et al., 2017; Engle et al., 2007; S. P. Walker et al., 2011). Our results on child health outcomes fit well within the literature focused on maternal education in developing countries (Desai & Alva, 1998) as well as with the evidence uncovered in Akresh et al. (2018) and Mazumder et al. (2019) on the same INPRES program. The findings on performance on standardized tests in language and sections of the one in math conform with findings presented in Andrabi et al. (2012) for the Pakistani context, even though our results on child time use (or daily activities) are not exactly in line with the reasoning presented there and also those resulting from a direct early-childhood intervention (Attanasio, Cattan, Fitzsimons, Meghir, & Rubio-Codina, 2020). The findings regarding socio-emotional development and school-readiness are novel in this literature and provide strong causal evidence of the intergenerational impact of educational interventions on children’s developmental landmarks. 2 Data and context 2.1 Data sources and sample restrictions The data used in this paper draw on information collected in the course of an impact evaluation of the Indonesia Early Childhood Education and Development (ECED) Project. The project expanded access to early childhood education and the evaluation focused on the effects on children’s school 5 Economics of parenting includes seminal work by Weinberg (2001), which focuses on the use of physical pun- ishment, and provides stylized facts and argues that parenting should be factored into economic models of child development. More recent empirical work includes Jenkins and Handa (2019). 5 readiness. Data were collected from children, their caregivers and a variety of other respondents in 310 villages in 9 districts across Indonesia. Respondents were 11,263 caregivers who resided in these districts. Among those, 7,982 were mothers who fully reported on outcomes and attributes we employ in our models below, including location of her own birth. They represent 171 distinct districts of birth. Our analysis further restricts this sample to women in birth cohorts surrounding the time of the INPRES program implementation (those born between 1966 and 1975), leaving us with 2,118 mothers from 50 different districts of birth. Figure 1 presents the geographic distribu- tion of contemporaneous district of residence (Panel A), as well as the spatial location of districts of birth for both the full sample (Panel B) and the working sample (Panel C). Previous analysis has compared contemporaneous household level characteristics in the In- donesia ECED data to households observed in the rural subsample of Indonesia’s nationally rep- resentative National Socioeconomic Household Survey (SUSENAS). Hasan, Hyson, and Chang (2013) report that the education levels of the heads of households, the rates of asset ownership among households, and the quality of construction materials used in the home among the house- holds in the ECED data are comparable to those observed in the SUSENAS. Similar analysis using a village level census reveals that the villages in the ECED data share many characteristics in common with the rest of rural Indonesia. As such, the underlying data from which our analytical sample is drawn share several characteristics in common with households and villages typically found in rural Indonesia. Our identification strategy described below relies on both the year and location of birth, and this is because the INPRES program followed specific rules for allocation of schooling expansion funds across districts. Therefore, our second source of data is information on construction planned under INPRES reported in Duflo (2001), which identifies the number of schools to be built in each district after 1974. We merge this district-level data to our working sample of 2,118 mothers in the ECED data. Appendix Table A1 reports descriptive statistics for the districts on Duflo (2001) data as well as for districts of birth captured in the working sample. It is clear from these figures that our working sample represents a set of districts that had a 6 less intensive presence of INPRES schooling expansion relative to Duflo (2001). Nonetheless, the working sample still captures 38 districts in which the number of schools planned (per capita) is larger than the equivalent number measured at the national level (1.8 per 1,000). Therefore, we focus on the contrast between mothers born in districts with “intensive” INPRES presence and those born in “non-intensive” INPRES districts. In Table A2 we also show that the variables influencing INPRES allocation of schools across space correspond to the same ones examined in Duflo (2001), and that this relationship is not significantly different in our working sample. INPRES allocation was indeed remedial, targeting districts where more children could be served, where enrollment rates had been lower, and where distances to be traveled (to existing schools) were larger (indirectly measured by population density). For completeness, we also present the contrast between main demographic characteristics of the full ECED sample and our working sample in Appendix Table A3. It is clear that the main difference between the two comes from the restriction we impose in terms of cohort-of-birth. Therefore, mothers in our working sample are older, slightly less educated (and so are the chil- dren’s fathers) than those in the full sample. It is important to note that since the ECED evaluation design was based on sampling children between the ages of 6-9, our exercises are indeed drawing from a sample of children with older mothers, larger sibships, and likely higher parity than the overall sample in the Indonesia ECED Project. It is key to take these points into consideration when discussing limitations to the external validity of our results. Appendix Table A3 reports that in our working sample, mothers are 41 years old on average, with 32 percent being born in years 1966-1970 and 68 percent in years 1971-1975. On average, they have completed 7.4 years of schooling. Approximately 80 percent of them report having completed primary education and being able to correctly read the sentence “rain came late this year” in Bahasa Indonesia. These mothers (who are close to completing their fertility if not so already) report having given birth to 4.3 children on average. Focus children are on average 7.7 years of age and 50 percent of them are boys. In about 9 percent of the cases, the father is reported to not reside in the household with them. This includes those who report that the father is deceased. 7 On average the non-absent father of the child is 4.6 years older than mothers and has completed 7.7 years of schooling (with 79 percent reporting having completed primary school education). 2.2 Child development outcomes We focus on multiple dimensions of a child’s development. In doing so we also illustrate the raw differences in outcomes across maternal education and literacy status. We start by focusing on three main indicators of child health: low birthweight, wasting and stunting. For low birth weight we classify all reports of birthweight at and below 2,500 grams.6 Stunted growth refers to low height for age, when a child is short for his/her age (WHO standardized population) and is an indicator of chronic malnutrition and carries long-term developmental risks. Wasting refers to low weight for height, the process by which acute food shortage and/or disease causes muscle and fat tissue to waste away. This is also known as acute malnutrition because wasting develops in a relatively short period of time in contrast to stunting. These statistics are especially pertinent since stunting in early life is associated with impaired cognitive ability, lower educational attainment, reduced future productivity, earnings potential and greater risk of poverty (Alderman, Hoddinott, & Kinsey, 2006). Indonesians whose growth was stunted in childhood were shorter (by 3.5 cm) and demonstrated lower cognitive function as young adults and had spent fewer years (by 5 months) enrolled in formal education (Giles, Satriawan, & Witoelar, 2020). Lower adult stature and cognitive ability were in turn associated with lower adult earnings in Indonesia (Sohn, 2015). In Table A4 we report descriptive statistics. We see that 17% of the children in rural Indonesia are born weighing 2,500 grams or less. Approximately 30% of these children also grow up to be wasted (under 2 standard-deviations of the WHO reference population’s mean) and 19% to be stunted. We complement this description by looking at the weight-for-age and height-for-age z- scores for groups of children defined by the literacy status of their mothers/caregivers. Figure 2 6 The child’s birthweight is based on recall by the mother and when possible verified using the birth registration card. The typical child was 3,106 grams at birth. Digital scales were used for weight measurement and tape measures were used for measuring respondent height. These were then compared to World Health Organization Child Growth Standards and WHO Reference 2007 composite data to assess whether respondents are stunted or wasted. 8 contrasts kernel-density estimates for these outcomes and reveals a clear shift to the left for these distributions when focusing on children of illiterate mothers. These are meaningful differences which indicate that the children of illiterate mothers are worse off. The ECED survey also collected child outcomes on a battery of cognitive tests and examina- tions of school-curriculum materials. Children were administered tests to assess their ability in terms of Bahasa Indonesia, mathematics and abstract reasoning. Two versions of the overall test were administered: a shorter test for 6 and 7-year-olds and a longer test for 8 and 9-year-olds. In this paper, we use the common set of items that were included in both versions of the test. We stan- dardized the percentage of correct answers for each test to have a mean of 0 and standard deviation of 1, using the age-specific mean and standard deviation of the entire ECED sample.7 The language test consisted of two sections. The first section (match pictures) evaluated chil- dren’s phonological awareness (i.e., whether they can match pictures that start with a given sound) and letter recognition (i.e., whether they can match pictures that start with a given letter). The sec- ond section (mention objects) assessed children’s vocabulary skills (i.e., whether they can name the word associated with a given image). The mathematics test included two sections. The first section (summation) evaluated children’s ability to add and subtract (i.e., whether they can add to or subtract away from a set of objects). The second section (order numbers) assessed children’s ability to recognize patterns (i.e., whether they can order one- to two-digit numbers in ascending and descending order). Finally, the abstract reasoning section was modeled on the Ravens Progres- sive Matrices. Students were presented with an image that was missing a small section and asked to select the missing pieces from six options, based on color, pattern, and orientation. Like in the case of health outcomes we see a strong positive gradient between child perfor- mance on each of these tests and maternal education. Figure 3 reveals a clear positive gradient between scores and maternal education. When looking at raw scores we see that children of liter- ate mothers get rates of correct responses that are approximately 13.6 p.p. (Language), 15.8 p.p. (Math), and 10.7 p.p. (Reasoning) higher than their peers whose mothers cannot read, for example. 7 Panel A in Table A5 presents descriptive statistics for the raw scores in the working sample (in terms of percentage of correct answers). 9 We then turn to the third and final dimension of child development covered in this study: socio- emotional and behavioral development. The survey included instruments based on the Strengths and Difficulties Questionnaire (SDQ), in which mothers assess children’s behavioral and emotional problems along five dimensions: emotional symptoms, conduct problems, hyperactivity/inattention, peer relationship problems and prosocial behavior problems.8 While all five sub-scales are typi- cally used when screening for disorders, broader subscales can be used for analytical purposes. This involves combining the emotional and peer subscales into an internalizing subscale and the conduct and hyperactivity subscales into an externalizing subscale (Goodman, Lamping, & Ploubidis, 2010). Internalizing behaviors are negative behaviors focused inward, and can include fearfulness, social withdrawal, and excessive thoughts, feelings and behaviors focusing on physi- cal symptoms such as pain, weakness or shortness of breath (somatic complaints). In contrast, an externalizing behavior is directed outward toward others. Bullying, vandalism, and arson are ex- amples of externalizing behaviors. Studies have found the Strengths and Difficulties Questionnaire (SDQ) to be a good predictor of children’s scores on the same instrument one year later (Hasan et al., 2013). Figure 4 relates the count of difficulties with maternal education (completed years). In this case, there are some interesting differences in the strength of the gradient. Prosocial difficulties (considerate of others people feelings; sharing with other children; helpful if someone is hurt, upset or ill; kind to younger kids; volunteering to help others) are the most strongly associated with maternal education, followed by internalizing and externalizing difficulties, respectively. Illiterate mothers have kids with prosocial difficulties scores 0.23 standard-deviations (s.d.) higher than kids whose moms are literate. The corresponding numbers for internalizing and externalizing difficulties are 0.08 and -0.02 s.d., respectively. These scores were standardized using the age specific group within the full ECED sample. Appendix Table A5, Panel B, reports the descriptive statistics. The second instrument we employ is the Early Development Instrument, a comprehensive mea- 8 As noted in Pradhan et al. (2013), the SDQ is an informant-based assessment of a child used internationally. It has been translated to Indonesian by Wiguna and Hestyanti (2012). 10 sure of school readiness typically reported on by teachers and parents. In our analysis we compute an average of reports by teachers and caregivers. As described in Hasan et al. (2013), the EDI assesses children’s readiness for school across five developmental domains: i. Physical health and well-being. Children who perform well on the physical health and well- being domain are usually dressed appropriately for school activities and do not come to school hungry or tired. They have established a hand preference, are well-coordinated and have well- developed gross and fine motor skills. ii. Social competence. Children who perform well on the social competence domain get along well with other children, are cooperative and self-confident. They show respect for others, follow rules, exercise self-control, and take responsibility for their actions. These children work neatly and independently the majority of the time. Children who score high on the social competence domain are able to solve problems, follow instructions, and easily adjust to changes. iii. Emotional maturity. Children who score high on the emotional maturity domain demon- strate helping behaviors, often spontaneously. They rarely show signs of anxiousness or aggressive or hyperactive behaviors. They concentrate well, do not have temper tantrums, and are not mean to others. iv. Language and cognitive skills. Children who perform well on the language and cognitive skills domain possess basic literacy skills (know how to handle a book, can identify letters, know some letter sounds, are aware of rhyming words, and are able to write their own name). These children are interested in books, mathematics, and numbers, and have good memories. They are able to read and write simple words or sentences. Additionally, these children have basic numeracy skills (can count to 20, recognize numbers and shapes, compare numbers, sort, and classify). v. Communication skills and general knowledge. Children who score high on the communica- tion skills and general knowledge domain can communicate effectively with ease, can tell stories and engage in imaginative play, articulate clearly, and show reasonable general knowledge. The EDI has well established concurrent and predictive validity in a number of developed economies. In Indonesia the EDI was validated using both the short and long version of the in- 11 strument using teacher and parent reports (Brinkman et al., 2017). Data collected on a cohort of 4 year-olds also shows that as children get older their performance on each of the domains goes up and their vulnerability declines (Hasan et al., 2013). Figure 5 provides a visual depiction of the association between maternal education and child development as reported by teachers and caregivers. Across the board we see a positive relation- ship between the two variables. Language and cognitive development, communication skills and general knowledge and social competence are dimensions with strongest relationship. Literate mothers have kids with raw social development scores 0.30 s.d. above those of kids whose moms do not read. Language and cognitive development scores are 0.43 s.d. higher, and communication skills and general knowledge scores are 0.26 s.d. higher, respectively.9 Once again, Appendix Table A5 reports the descriptive statistics regarding all dimensions (Panel C). 2.3 Parenting, child activities and human capital investments 2.3.1 Parenting Given the association between parenting styles and child outcomes posed by the broader devel- opmental literature, we make full use of our survey instruments to take a unique approach and investigate elements of parenting that may influence child development. In particular, we presume some of these can be influenced by the amount of formal education caregivers acquire.10 The fo- cus is on how families influence their children’s development through their relationships, such as the level of warmth or anger that mothers feel for their children and the kind of discipline that mothers tend to use when children misbehave. These practices are important because children liv- ing in an environment with higher-quality parenting are believed to be more likely to have higher pre-academic skills, better language skills, more social skills, and fewer behavior problems than children who received lower quality parenting (NICHD, 2002). In our survey, mothers were asked to answer a series of questions about their parenting prac- 9 The corresponding numbers for physical health and well-being and emotional maturity are 0.09 and 0.12. 10 Psychology and sociology have a long tradition of associating parenting with socio-economic status (Baumrind, 1991; Kohn, 1969). 12 tices. These practices were measured using items describing parent-child relationships adapted from the Longitudinal Study of Australian Children (Zubrick, Smith, Nicholson, Sanson, & Jack- iewicz, 2008). The questions covered a range of possible practices. Parenting practices are specific behaviors that parents use in their interactions with their child. These include, for example, using reprimands, giving praise, showing physical affection and setting and keeping with rules for be- havior (Bornstein & Zlotnik, 2008; L. O. Walker & Kirby, 2010).11 Following a recent narrative review of parenting, three hierarchical summary constructs are proposed: warmth, consistency, and hostility (Jansen, Daniels, & Nicholson, 2012). We compute these constructs by calculating within each of them the proportion of items the caregiver responds as having undertaken frequently. Many dimensions of parenting have been shown to influence child development, although different terms are often used to describe overlapping or similar constructs. Generally, children show better developmental outcomes when exposed to parenting that is high on the dimensions of warmth, consistency, inductive reasoning and self-efficacy and low on the dimensions of irritabil- ity, hostility and over-protectiveness (Bayer et al., 2011; Bradley et al., 1989; Chang, Schwartz, Dodge, & McBride-Chang, 2003; Chao & Willms, 2002; Paterson & Sanson, 1999; Pettit & Bates, 1989). Parenting styles are multidimensional categories of behaviors and attitudes which classify parents according to where they lie on the distributions of some specific parenting dimensions (Darling & Steinberg, 1993). One of the most well-known classifications of parenting style is that applied by Baumrind and others (Baumrind, 1991; Darling & Steinberg, 1993; Maccoby & Mar- tin, 1983) defining four parenting styles based around levels of over-controlling and responsive parenting: authoritative (high control, high responsiveness), authoritarian (high control, low re- sponsiveness), indulgent/permissive (low control, high responsiveness), and uninvolved/neglectful parenting (low control, low responsiveness). In the Anglo population in Western societies, author- itative parenting has been most consistently associated with positive socioemotional competence, cognitive and health outcomes in children (Baumrind, 1991; Bornstein & Zlotnik, 2008; Jackson, 11 The psychology literature has paid close attention to these practices, particularly in the case of corporal punish- ment and spanking. See Gershoff and Grogan-Kaylor (2016) for a review and Putnick et al. (2012) for an analysis of data from a cross-section of countries. 13 Henriksen, & Foshee, 1998; Smith, 2011). Some dimensions of parenting are not applicable at all ages (e.g. inductive reasoning is not applicable in infancy), and specific parenting behaviors may be appropriate at some ages but not others (e.g. leaving the child alone in their room may be an appropriate discipline strategy for a preschooler, but not for an infant). As a result, both the broader parenting constructs and the specific items used to assess them needed to be mapped against the ages of intended use. We therefore standardized our measures to the age of the child when estimating econometric models. We present descriptive statistics on these measures in Panel A of Table A6. These figures indicate that there is variation on these practices within our sample, and also that the caregivers we study are over-represented in the bottom of the distribution of these scores. When we examine differences by literacy, we see that women who can read are substantially more likely to show warmth and consistency in their parenting, and less likely to show hostility. 2.3.2 Activities Caregivers were also asked to complete an inventory asking whether or not the child took part in a series of typical play and stimulation activities in the past week. This is akin to the inventory of activities employed by time-use surveys in Attanasio et al. (2020) and Andrabi et al. (2012). The descriptive statistics are presented in Panel B of Table A6. Across the board we see high level of engagement in stimulating activities, but also substantial engagement in household chores. We also looked at differences between children of literate and of illiterate mothers, and found that, as expected, the former are more likely to be engaged in reading and music-playing (13 p.p.), but also in every other activity (albeit with smaller differences across groups). 2.3.3 Investments in human capital In addition we use information from the section on the child’s health history to identify whether the child has received a full set of immunizations by the date of the survey. We also collect information on the types of food consumed in the past week and the number of days in the past week which 14 they consumed each food. Food was categorized into seven groups, following the dietary diversity guidelines from the Food and Agriculture Organization of the United Nations (FAO): (1) grains, (2) vitamin A-rich plant food, (3) other fruits and vegetables, (4) meats, fish, and seafood, (5) eggs, (6) dairy, and (7) oils and fats. This information was then combined into two measures of dietary diversity. The first measure is whether children consumed foods from a minimum number of dietary groups (four or more groups) and the second measure is whether children consumed foods from the maximum number of dietary groups (seven) in the past week. Finally, we also asked parents to provide enrollment histories for each child from the 2015/2016 school year going back to the 2012/2013 school year. These were used to construct whether or not a child enrolled in a particular time of school service as well as a measure of their cumulative months of enrollment at the time of the survey. Descriptive statistics on all these variables are presented in Table A7. We see that immunization rates and minimum dietary diversity rates are quite high in this population, yet the majority is not close to getting maximum dietary diversity. Mothers are well aware of pre-school opportunities around them but much fewer of them act on that knowledge on average. Differences between literate and illiterate mothers are sizeable, around 9 p.p. for immunization rates, 5 to 6 p.p. in dietary diversity and about 8.5 p.p. in attendance of pre-schooling institutions (or about 1.1 less months). 3 Empirical design and causal identification strategy Motivated by the correlation between child outcomes and maternal education, as well as by the opportunity to trace changes in maternal access to education with different parenting styles and parental investments, we focus on quasi-experimental variation generated by the INPRES school- ing expansion to investigate impacts across generations. Following Duflo (2001), we combine geographic variation in planned school construction intensity (per 1,000 children) with a woman’s year of birth in order to formulate a difference-in-differences empirical model. We differ from that 15 seminal article in that we elect to focus on a binary classification of construction intensity per dis- trict (above or below the aggregated number of schools per 1,000 children planned for the whole country). Our analysis also focuses on the contrast between two cohorts for which primary school construction would have provided a different treatment dose. In the ECED data, the vast majority of mothers are born after 1957-1962 (the comparison cohort used in Duflo (2001)) because the sampling was based on being mothers to young children in 2013. Thus, our analysis contrasts women born between 1966 and 1970 with those born between 1971 and 1975. Since the timing of construction was variable within districts and older children are unlikely to benefit from newly constructed primary schools, the latter cohort (1971-1975) is expected to have accrued most of the benefits of increased access to elementary education generated by INPRES. We start by estimating models that confirm the effect of planned school construction on the education of mothers interviewed in 2013. For the effect on the mother’s children, we focus on a reduced-form or intent-to-treat analysis because the impact of school construction over the second generation may result from multiple mechanisms. In Figure A1, we reproduce these findings using data from the 2010 Census. The additional data allows us to examine the parallel-trends hypotheses (at least in the dimension of maternal schooling) which sustain our empirical design but, due to aforementioned child-mother age restrictions in our working sample, could not be directly performed.12 Our full econometric specification based on this reasoning and focused on maternal educational outcomes is as follows: Yicd “ µd ` φc ` τ Intensed ˆ Y oungc ` β 1 Zd ˆ Y oungc ` ηicd (1) where Intensed is an indicator function for districts where INPRES construction was more inten- sive than the national level, Y oungc flags mothers in 1971 to 1975 birth-cohorts. Fixed-effects for district of birth (µd ) and year-of-birth (φc ) guarantee that the parameter τ is interpreted as a difference-in-differences coefficient. We complement the model with interactions between other 12 We also rely on the fact that studies of INPRES utilizing alternative data sources have shown that the parallel trend assumption seems to hold (Akresh et al., 2018; Duflo, 2001; Mazumder et al., 2019). 16 district-of-birth characteristics (Zd ) and the indicator for belonging to the most exposed cohorts in order to alleviate concerns that other district-of-birth characteristics and not INPRES intensity could explain the results we find.13 Standard errors are clustered at the district of birth of mothers. Another potential threat to identification is related to the migration of mothers. In our data, mothers’ district of residence in 2013 is limited to the nine rural districts in the Indonesia ECED data. Thus, we may be concerned that migration from high INPRES birth districts to our nine districts reflects particularly aspirational individuals (i.e., those likely to obtain more education and to experience upward geographic mobility) whereas migration from low INPRES birth districts to our nine districts selects on particularly low-aspirational individuals (i.e., those likely to obtain less education and to experience downward geographic mobility). While we cannot directly rule out this possibility, our analysis of migration patterns among mothers in our data suggests that this is unlikely to be a major threat to validity. Figure 6 presents the mobility pattern across districts by whether the district experienced low intensity or high intensity INPRES. Each dot is a birth district, sorted by levels of Human Development Index (a measure of overall development). Birth districts that are more (less) developed are on higher (lower) values of the y-axis. For each birth district, we calculate the median HDI of the district where mothers are residing in 2013. The difference between this median 2013 HDI and the birth district HDI (x-axis) indicates the level of mobility experienced by mothers coming from each birth district. The figure clearly shows similar and consistent migration patterns across low and high intensity INPRES districts: mothers born in higher HDI districts experience larger downward mobility, while mothers born in lower HDI districts experience larger upward mobility. This suggests that differential migration patterns between low and high intensity INPRES districts are unlikely to bias our results. We expand the empirical model above to investigate the effect of maternal education on chil- dren’s outcomes. Specifically, we consider children’s outcomes to likely be a function of the age and gender of the focus child in our study. We do so with alternative controls (Xicd ) for age and an indicator function for child gender. Therefore, the model taken to the data is: 13 These are the same robustness exercises conducted in Duflo (2001). 17 Yicd “ µ ˜c ` τ ˜d ` φ ˜1 Zd ˆ Y oungc ` α1 Xicd ` η ˜Intensed ˆ Y oungc ` β ˜icd (2) ˜ captures the intent-to-treat (ITT) effect of INPRES on children belonging where the parameter τ to the second generation. Again, standard errors are clustered at the district of birth of mothers. Finally, in order to emphasize impacts at particular points of the distribution of the dependent vari- able, we estimate linear probability models with quantile indicator variables replacing continuous measures of the outcomes on the left-hand side of these models.14 4 Results 4.1 INPRES and maternal education Table 1 summarizes the impact of school construction on maternal education. For each panel, the cells in columns 1, 2 and 3 illustrate the simple two-by-two tables of our difference-in-differences identification strategy. Mothers’ birth districts are classified in terms of school construction inten- sity as either “high INPRES” (column 1) or “low INPRES” (column 2).15 Mothers belong to low exposure cohorts born between 1966-1970 or to high exposure cohorts born between 1971-1975. Recall that by design, school construction planned under INPRES was more intense in districts where school enrollment rates were low. Therefore, for both cohorts, mothers born in districts with low INPRES intensity have higher levels of educational attainment (Panel A), primary school completion rate (Panel B) and literacy rate (Panel C) than mothers born in districts with high pro- gram intensity. While educational outcomes remained similar across birth cohorts in low INPRES districts, there were large improvements in educational outcomes over time in high INPRES dis- tricts. Thus, the difference-in-differences yields positive impacts of school expansion on maternal education. 14 For the case of maternal education we use the same strategy to identify particular ranges in years of education affected by INPRES. 15 As shown in Appendix Table A1, the difference between the number of schools constructed per 1,000 children in high and low intensity districts is 1.10. 18 Our preferred model specification in column 5 shows that mothers exposed to higher doses of the INPRES program completed an additional 1.1 years of education, improved primary comple- tion rate by 13.3 percentage points, and increased literacy rates by 13.2 percentage points. Our estimates are much larger than the estimates in Akresh et al. (2018); Duflo (2001); Mazumder et al. (2019). However, larger estimates are expected given that our study focuses on mothers living in rural Indonesia in adulthood. Given negative selection into rural areas in developing countries (Young, 2013), mothers in our study are those who are particularly disadvantaged. These disad- vantages are clear from the low levels of education among mothers born in high INPRES districts. In our sample, mothers born in the high intensity districts have an average of 5.8 and 6.8 years of education in the low and high exposure birth cohorts (respectively). In contrast, the educational attainment of high INPRES districts in Akresh et al. (2018); Duflo (2001); Mazumder et al. (2019) is over 8 years across birth cohorts. Our identification strategy assumes that there are no omitted time-varying and district-specific variables that are correlated with school construction. A key threat to identification is if improve- ments in educational attainment among mothers in high intense INPRES districts simply reflect mean reversion. This is plausible since program intensity was directly determined by enrollment rates in 1972. Another key threat to our identification assumption is if other governmental interven- tions occurred during the same time and used similar allocation as the school construction program. For example, the second largest government program during this period was for water and sani- tation (Duflo, 2001). Thus, our diff-in-diff estimates may be upwardly biased if it confounds the effect of school construction with mean reversion that would have occurred in the absence of the program or with other government interventions. To assess mean reversion, we estimate a model that includes an interaction between high expo- sure birth cohort and enrollment rates in 1971. Similarly, we check the sensitivity of our treatment effect estimates to other major government programs during the same period by estimating a model that includes an interaction between high exposure birth cohort and the intensity of the water and sanitation program. The inclusion of these interaction increases our treatment effect from 1.064 19 (column 5) to 1.242 (column 6) and 1.133 (column 7), which suggests that mean reversion and omitted programs are not likely to be driving our main results. We also report a wild bootstrap p-value to account for the somewhat small number of clusters (50 birth districts) in our sample. The large impact of school construction on maternal education in our sample underscores the im- portance of expanding educational opportunities for women in rural areas, who are often the most marginalized members of society. 4.2 Children’s outcomes Having established the impact of school construction on maternal education, we next turn to the intent-to-treat impacts on children’s outcomes. Table 2 reports on health at birth in Panel A. Col- umn 1 presents difference in differences estimates which control for caregiver district and year of birth fixed effects as well as child gender and child’s age fixed effects. Column 2 adds an interac- tion term for whether the mother is from the treated cohort and the number of school age children in 1973. Column 3 replaces the child’s age fixed effects with a linear term for the child’s age. Esti- mates are shown with empirical standard errors as well as p-values derived from a wild-bootstrap. Each table also includes two columns which contain reference values to ease in interpreting the magnitudes of the point estimates. The column labeled “1996-1970 maternal cohort high INPRES mean” reports the average value of the outcome among least likely to be exposed (comparison cohort) mothers in districts that received an intensive dose of INPRES. The column labeled “1996- 1970 maternal cohort high-low INPRES gap” reports the difference between high and low INPRES districts among comparison cohort mothers. While the point estimates indicate that the birthweight was higher for children born to mothers exposed to larger doses of the program relative to those born to mothers exposed to smaller doses of the program, the point estimates are not statistically significant. Similarly, the program did not have statistically significant impacts on the incidence of low birthweight. Panel B reports on health in infancy and finds that there are no significant impacts of the program on whether children were sick in the past fortnight. Weight-for-age and whether the 20 child is wasted are also similar. When looking at extreme wasting, there is some evidence that children of mothers exposed to larger doses of the program are less likely to be extremely wasted. Given the average prevalence of extreme wasting (8.7%) the point estimate is substantial at -3.9. However, the p-value is 0.11. In contrast, the evidence is stronger for children’s height-for-age. The program increased height-for-age by 0.15 s.d. Give the sample mean in the high INPRES district, this translates to a 10% increase in height-for-age. The program seems to have been particularly effective at reducing extreme stunting rates (2.72 p.p.). Given that 4.6% of children born to mothers from high INPRES districts are extremely stunted, the effect is substantial. Table 3 follows the same format as the previous table and reports the impact of the program on an array of cognitive assessments in the next generation. On the test of Bahasa Indonesia, children of mothers exposed to high doses of the program scored 0.14 s.d. higher than children of those exposed to low doses of the program. These are substantial effects given that the difference be- tween comparison cohort mothers in high and low INPRES districts is -0.63 s.d. This suggests that INPRES helped close the pre-existing gap in education between high and low INPRES districts. When examining the bottom end of the test score distribution, there is even stronger evidence of the lasting impacts of maternal exposure to INPRES. Children of treated mothers are 11 p.p. less likely to be in the bottom quartile. This is the equivalent to 38% of the gap between comparison cohort mothers in high and low INPRES districts. Tests of mathematics do not reveal similar results. The point estimates are smaller and not significant. However, when we unpack the test into its constituent sections and assess children’s performance on the two segments of the test separately, we find evidence to suggest that children of treated mothers were more likely to do better at the basic skill of ordering numbers. We find, for example, that the effect measured on this part of the exam is equivalent to 0.19 s.d. (p-value 0.072) using the same specification of column 2 in Table 3. The effects are insignificant and close to zero for the addition section of the exam, on the other hand. Together these results are aligned with the fact that the impacts are concentrated on the lower end of the outcome distribution. Finally, there is no evidence that maternal exposure to INPRES improves children’s cognitive 21 reasoning skills, based on a test using the Raven’s progressive matrices. This is not surprising, since these skills are likely not generated by exposure to more investment in educational activities. Applying the Strengths and Difficulties Questionnaire (SDQ) to this sample of children reveals that there is no impact - either practical or statistical - on children’s internalizing behavior or externalizing behavior (Table 4).16 However, the program reduces children’s pro-social behavior problems by -0.18 s.d. For context, the gap between high and low INPRES districts is 0.32 s.d. Thus, the estimate is practically very large, with over 50% improvement in pro-social behavior, and statistically significant. The Early Development Instrument (Table 5) corroborates the findings from the SDQ. The coefficients on four out of the five EDI domains are practically small and statistically insignificant. However, there is a substantial positive effect on children’s social competence. Children of mothers who were exposed to high doses of the program have 0.2 s.d. higher social competence than children of mothers who were exposed to low doses of the program. The children are 8 p.p. less likely to be in the bottom quartile of the EDI domain distribution for their age. This is equivalent to 72% of the gap between high and low INPRES districts. 4.3 Mediators Next, we examine several channels of transmission between maternal education and child develop- ment, by drawing on insights from economics, psychology, and sociology. We begin by examining characteristics of the children’s father. The two columns in Table 6 differ only in that column 2 adds an interaction between the treated cohort and the number of school-aged children in 1973. The program did not affect the presence of fathers in the household and age of fathers. However, there is evidence that mothers who were exposed to higher doses of the program have children with more educated men. The program increased educational attainment of fathers by three-quarters of a year and primary school completion rate of fathers by 8 p.p. This lends some evidence to assorta- tive mating as one channel of transmission for the benefits of exposure to INPRES. Yet, we cannot 16 Recall that the SDQ is scored such that lower scores indicate better child development outcomes. 22 rule out the possibility that fathers were directly affected by INPRES themselves when growing up. In terms of fertility, there is no program effect on the number of live births. Nor is there any impact on the likelihood that mothers are working outside the household. However, there is clear evidence that mothers exposed to larger doses of the program are in wealthier households by about 0.18 s.d. The data also suggest that mothers exposed to larger doses of the program are 10 p.p. less likely to be in the bottom quartile of the wealth distribution. Table 7 allows us to capture a novel feature of the data - detailed information on behaviors related to parenting. We examine three aspects of parenting: warmth, consistency and hostility. The point estimates for warmth and consistency are small and statistically insignificant. However, we find program effects on parental hostility. Mothers who were exposed to larger doses of the program are much less likely to be in the bottom quartile of the parental hostility distribution. These point estimates are fairly large. The mean prevalence of parental hostility in high INPRES districts is 34.4% while the point estimate is -9.3 p.p. This suggests that exposure to INPRES led to a 27% reduction in the likelihood that mothers are in the bottom quartile of the parental hostility distribution. In practice, this means that the program improved parenting practices by reducing the most hostile parenting behaviors. In addition to effects on parenting practices, we investigate whether the school expansion im- proved maternal engagement in different activities with their children (Table 8). With the exception of playing music, the point estimates on all other activities – ranging from reading to storytelling to household chores – are small and statistically insignificant. Children of mothers exposed to larger doses of the program are 8 p.p less likely to report having played music with their mothers in the last week. This correspond to a 12% reduction in music activities in high INPRES districts. We examine investments made in children’s human capital in Table 9. We do not find that the school construction affected children’s immunization rates. This is likely due to the growing global awareness around the importance of immunizations as well as the big push for ensuring immunizations by the government of Indonesia around 2013 (World Health Organization, 2013). 23 This is underscored by the high immunization rates in the high INPRES districts (83%) as well as the relatively small gap between high and low INPRES districts. We also examine children’s dietary diversity. 95% of children born to mothers from high INPRES districts meet the minimum requirements of dietary diversity. Consequently, it is not surprising that we do not detect effects of the school construction on children’s diet. Similarly, there are no differences in terms of maximum dietary diversity. Finally, we examine effects on investments in children’s education. Mothers who were ex- posed to larger doses of the program are 8 p.p. more likely to know the location of the closest pre-school. This effect size translates to a 12% increase in knowledge of preschools among high INPRES districts. Likewise, the program led to a 5 p.p increase in the children ever-attending a pre-school. Given that only 28% of mothers in the high INPRES districts report that their child ever-attended preschool, the point estimate is a substantial increase of 20%. Lastly, the program improved the next generation’s early education investments by inducing longer months of enroll- ment in preschool by 31%. 5 Conclusion The evidence presented in this paper should reaffirm to policy makers the critical role of ensuring access to education for all. The evidence shows that there are lasting benefits - not just to the individuals who acquire greater schooling themselves but to a wide array of developmental markers for their children. We find significant and meaningful evidence of intent-to-treat effects not only in terms of long-run health status (extreme stunting and to a lesser extent extreme wasting) but also in terms of cognitive outcomes (language and mathematics scores in early grades of primary school) as well as in terms of children’s social behavior (measured using both the SDQ and the EDI). We find evidence to support the fact that the effects of maternal education operate through a multitude of related and self-reinforcing pathways. Treated mothers are likely to belong to better resourced households. Importantly, treated mothers engage in better parent-child interactions - 24 specifically they are much less likely to be hostile and/or violent. This conforms well with a long list of contributions to the psychology literature linking parental usage of physical punishment and the social-emotional development of their children. These mothers who had facilitated access to schooling also choose to send their children to pre-school at higher rates and for longer durations. This points to a connection between the knowledge they acquire in school and what they ultimately do as caregivers. All our results on child outcomes fit well within the literature focused on maternal education in developing countries (Andrabi et al., 2012; Desai & Alva, 1998) as well as with the evidence uncovered on the same INPRES program (Akresh et al., 2018; Mazumder et al., 2019). The find- ings regarding socio-emotional development and school-readiness are novel in this literature and reaffirm strong causal evidence of the intergenerational impact of educational interventions on chil- dren’s developmental landmarks. As countries grapple with the global learning crisis (World Bank, 2018, 2019), investments in access to and quality of education will continue to be of paramount importance. These investments will reap returns not only for those who receive them directly, but also for generations to come. 25 References Akresh, R., Halim, D., & Kleemans, M. (2018). Long-term and intergenerational effects of education: Evidence from school construction in Indonesia. NBER Working Paper Series. Alderman, H., Hoddinott, J., & Kinsey, B. (2006). Long term consequences of early childhood malnutrition. Oxford Economic Papers, 58(3), 450–474. Andrabi, T., Das, J., & Khwaja, A. I. (2012). What did you do all day? Maternal education and child outcomes. Journal of Human Resources, 47 (4), 873–912. Attanasio, O., Cattan, S., Fitzsimons, E., Meghir, C., & Rubio-Codina, M. (2020). Estimating the production function for human capital: Results from a randomized controlled trial in Colombia. American Economic Review, 110(1), 48–85. Baumrind, D. (1991). The influence of parenting style on adolescent competence and substance use. The Journal of Early Adolescence, 11(1), 56–95. Bayer, J. K., Ukoumunne, O. C., Lucas, N., Wake, M., Scalzo, K., & Nicholson, J. M. (2011). Risk factors for childhood mental health symptoms: National longitudinal study of Australian children. Pediatrics, 128(4), e865–e879. Black, M. M., Walker, S. P., Fernald, L. C., Andersen, C. T., DiGirolamo, A. M., Lu, C., . . . others (2017). Early childhood development coming of age: Science through the life course. The Lancet, 389(10064), 77–90. Black, S. E., Devereux, P. J., & Salvanes, K. G. (2005). Why the apple doesn’t fall far: Un- derstanding intergenerational transmission of human capital. American Economic Review, 95(1), 437–449. Bono, E. D., Francesconi, M., Kelly, Y., & Sacker, A. (2016). Early maternal time investment and early child outcomes. The Economic Journal, 126(596), F96–F135. Bornstein, M., & Zlotnik, D. (2008). Parenting styles and their effects. In M. M. Haith & J. B. Benson (Eds.), Encyclopedia of infant and early childhood development (pp. 496–509). Elsevier. Bradley, R. H., Caldwell, B. M., Rock, S. L., Ramey, C. T., Barnard, K. E., Gray, C., . . . others (1989). Home environment and cognitive development in the first 3 years of life: A collab- orative study involving six sites and three ethnic groups in north america. Developmental Psychology, 25(2), 217. Breierova, L., & Duflo, E. (2004). The impact of education on fertility and child mortality: Do fathers really matter less than mothers? NBER Working Paper Series. Brinkman, S. A., Kinnell, A., Maika, A., Hasan, A., Jung, H., & Pradhan, M. (2017). Validity and reliability of the Early Development Instrument in Indonesia. Child Indicators Research, 10(2), 331–352. Carneiro, P., Meghir, C., & Parey, M. (2013). Maternal education, home environments, and the development of children and adolescents. Journal of the European Economic Association, 11, 123–160. Chang, L., Schwartz, D., Dodge, K. A., & McBride-Chang, C. (2003). Harsh parenting in relation to child emotion regulation and aggression. Journal of Family Psychology, 17 (4), 598. Chao, R. K., & Willms, J. D. (2002). The effects of parenting practices on childrens outcomes. In J. D. Willms (Ed.), Vulnerable children: Findings from Canada’s national longitudinal survey of children and youth (pp. 149–165). Edmonton, Alberta: University of Alberta Press. 26 Chevalier, A. (2004). Parental education and child’s education: A natural experiment. IZA Dis- cussion Paper. Currie, J., & Moretti, E. (2003). Mother’s education and the intergenerational transmission of hu- man capital: Evidence from college openings. The Quarterly Journal of Economics, 118(4), 1495–1532. Darling, N., & Steinberg, L. (1993). Parenting style as context: An integrative model. Psycholog- ical Bulletin, 113(3), 487. Desai, S., & Alva, S. (1998). Maternal education and child health: Is there a strong causal relationship? Demography, 35(1), 71–81. Dickson, M., Gregg, P., & Robinson, H. (2016). Early, late or never? When does parental education impact child outcomes? The Economic Journal, 126(596), F184–F231. Doepke, M., Sorrenti, G., & Zilibotti, F. (2019). The economics of parenting. Annual Review of Economics, 11, 55–84. Doepke, M., & Zilibotti, F. (2017). Parenting with style: Altruism and paternalism in intergenera- tional preference transmission. Econometrica, 85(5), 1331–1371. Dornbusch, S. M., Ritter, P. L., Leiderman, P. H., Roberts, D. F., & Fraleigh, M. J. (1987). The relation of parenting style to adolescent school performance. Child Development, 1244– 1257. Duflo, E. (2001). Schooling and labor market consequences of school construction in Indonesia: Evidence from an unusual policy experiment. American Economic Review, 91(4), 795–813. Engle, P. L., Black, M. M., Behrman, J. R., De Mello, M. C., Gertler, P. J., Kapiriri, L., . . . others (2007). Strategies to avoid the loss of developmental potential in more than 200 million children in the developing world. The Lancet, 369(9557), 229–242. Fischer, M., Karlsson, M., Nilsson, T., & Schwarz, N. (2019). The long-term effects of long terms – Compulsory schooling reforms in Sweden. Journal of the European Economic Association, jvz071. Gershoff, E. T., & Grogan-Kaylor, A. (2016). Spanking and child outcomes: Old controversies and new meta-analyses. Journal of Family Psychology, 30(4), 453. Giles, J., Satriawan, E., & Witoelar, F. (2020). Early childhood nutrition, availability of health service providers and life outcomes as young adults: Evidence from Indonesia. Working Paper. Goodman, A., Lamping, D. L., & Ploubidis, G. B. (2010). When to use broader internalising and externalising subscales instead of the hypothesised five subscales on the Strengths and Difficulties Questionnaire (SDQ): Data from British parents, teachers and children. Journal of Abnormal Child Psychology, 38(8), 1179–1191. Hasan, A., Hyson, M., & Chang, M. C. (2013). Early childhood education and development in poor villages of Indonesia: Strong foundations, later success. The World Bank. Haveman, R., & Wolfe, B. (1995). The determinants of children’s attainments: A review of methods and findings. Journal of Economic Literature, 33(4), 1829–1878. Holmlund, H., Lindahl, M., & Plug, E. (2011). The causal effect of parents’ schooling on children’s schooling: A comparison of estimation methods. Journal of Economic Literature, 49(3), 615–51. Jackson, C., Henriksen, L., & Foshee, V. A. (1998). The Authoritative Parenting Index: Predicting health risk behaviors among children and adolescents. Health Education & Behavior, 25(3), 319–337. 27 Jansen, E., Daniels, L., & Nicholson, J. (2012). The dynamics of parenting and early feeding constructs and controversies: A viewpoint. Early Child Development and Care, 182(8), 967–981. Jenkins, J. M., & Handa, S. (2019). Parenting skills and early childhood development: Production function estimates from longitudinal data. Review of Economics of the Household, 17 (1), 121–147. Kalil, A., Ryan, R., & Corey, M. (2012). Diverging destinies: Maternal education and the devel- opmental gradient in time with children. Demography, 49(4), 1361–1383. Kohn, M. (1969). Class and conformity: A study in values. University of Chicago Press. Maccoby, E. E., & Martin, J. A. (1983). Socialization in the context of the family: Parent-child interaction. In P. Mussen & E. Herrington (Eds.), Handbook of child psychology (pp. 1–102). Wiley. Maurin, E., & McNally, S. (2008). Vive la revolution! Long-term educational returns of 1968 to the angry students. Journal of Labor Economics, 26(1), 1–33. Mazumder, B., Rosales-Rueda, M., & Triyana, M. (2019). Intergenerational human capital spillovers: Indonesia’s school construction and its effects on the next generation. In AEA Papers and Proceedings (Vol. 109, pp. 243–49). NICHD. (2002). Early child care and children’s development prior to school entry: Results from the NICHD Study of Early Child Care. American Educational Research Journal, 39(1), 133–164. Oreopoulos, P., Page, M. E., & Stevens, A. H. (2006). The intergenerational effects of compulsory schooling. Journal of Labor Economics, 24(4), 729–760. Paterson, G., & Sanson, A. (1999). The association of behavioural adjustment to temperament, parenting and family characteristics among 5-year-old children. Social Development, 8(3), 293–309. Pettit, G. S., & Bates, J. E. (1989). Family interaction patterns and children’s behavior problems from infancy to 4 years. Developmental Psychology, 25(3), 413. Pradhan, M., Brinkman, S. A., Beatty, A., Maika, A., Satriawan, E., de Ree, J., & Hasan, A. (2013). Evaluating a community-based early childhood education and development pro- gram in Indonesia: Study protocol for a pragmatic cluster randomized controlled trial with supplementary matched control group. Trials, 14(1), 259. Putnick, D. L., Bornstein, M. H., Lansford, J. E., Chang, L., Deater-Deckard, K., Di Giunta, L., . . . others (2012). Agreement in mother and father acceptance-rejection, warmth, and hostility/rejection/neglect of children across nine countries. Cross-Cultural Research, 46(3), 191–223. Schady, N., Galiani, S., & Souza, A. P. (2006). Early childhood development in Latin America and the Caribbean. Econom´ ıa, 6(2), 185–225. Smith, M. (2011). Measures for assessing parenting in research and practice. Child and Adolescent Mental Health, 16(3), 158–166. Sohn, K. (2015). The height premium in Indonesia. Economics & Human Biology, 16, 1–15. Suhonen, T., & Karhunen, H. (2019). The intergenerational effects of parental higher education: Evidence from changes in university accessibility. Journal of Public Economics, 176, 195– 217. Walker, L. O., & Kirby, R. S. (2010). Conceptual and measurement issues in early parenting prac- tices research: An epidemiologic perspective. Maternal and Child Health Journal, 14(6), 28 958–970. Walker, S. P., Wachs, T. D., Grantham-McGregor, S., Black, M. M., Nelson, C. A., Huffman, S. L., . . . others (2011). Inequality in early childhood: Risk and protective factors for early child development. The Lancet, 378(9799), 1325–1338. Weinberg, B. A. (2001). An incentive model of the effect of parental income on children. Journal of Political Economy, 109(2), 266–280. Wiguna, T., & Hestyanti, Y. (2012). SDQ: Information for researchers and professionals about the Strengths and Difficulties Questionnaire. Indonesian translation. London: YouthinMind. World Bank. (2018). World Development Report 2018: Learning to realize educations promise. World Bank. World Bank. (2019). Ending learning poverty : What will it take? World Bank. World Health Organization. (2013). Indonesia introduces five-in-one vaccine for children. Re- trieved from https://www.who.int/immunization/newsroom/indonesia five in one 20130822/en/ Young, A. (2013). Inequality, the urban-rural gap, and migration. The Quarterly Journal of Economics, 128(4), 1727–1785. Zubrick, S. R., Smith, G. J., Nicholson, J., Sanson, A., & Jackiewicz, T. A. (2008). Parenting and families in Australia. FaHCSIA Social Policy Research Paper. 29 Tables Table 1: School construction and maternal education: Levels, Differences and Difference-in- differences. High Low Difference in Differences Robustness INPRES INPRES Checks [1] [2] [3] [4] [5] [6] [7] Panel A: Years of Education 1966-1970 Birth Cohort 5.776 9.411 (0.609) (0.447) 1971-1975 Birth Cohort 6.773 9.155 (0.562) (0.510) Difference 0.997 -0.255 1.253 1.049 1.064 1.242 1.133 (0.192) (0.177) (0.258) (0.261) (0.306) (0.481) (0.500) [0.016] [0.008] [0.024] [0.052] Panel B: Completed Primary Education (Yes=1) 1966-1970 Birth Cohort 0.631 0.939 (0.086) (0.022) 1971-1975 Birth Cohort 0.766 0.935 (0.080) (0.020) Difference 0.135 -0.004 0.139 0.121 0.133 0.142 0.143 (0.030) (0.021) (0.036) (0.039) (0.041) (0.074) (0.074) [0.040] [0.012] [0.152] [0.096] Panel C: Literate (Yes=1) 1966-1970 Birth Cohort 0.633 0.934 (0.051) (0.028) 1971-1975 Birth Cohort 0.79 0.937 (0.058) (0.017) Difference 0.157 0.003 0.154 0.13 0.132 0.154 0.142 (0.030) (0.020) (0.035) (0.033) (0.039) (0.063) (0.066) [0.008] [0.016] [0.044] [0.076] Controls (D-in-D’s) District and year of birth FE’s NO YES YES YES YES Treated cohort interaction with school-aged children in 1973 (ln scale) NO NO YES YES YES Treated cohort interaction with enrollment per capita in 1971 (ln scale) NO NO NO YES YES Treated cohort interaction with 1970’s sanitation program NO NO NO NO YES Sample 674 1,444 2,118 2,118 2,118 2,118 2,118 Notes: Standard-errors in parentheses are clustered at the district of birth. Wild-bootstrap significance p-values in brackets. Table 2: School Construction and Child Health: Difference-in-differences estimates of intent- to-treat. 1966-1970 Maternal Cohort Difference in Differences High INPRES High-Low INPRES Mean Gap [1] [2] [3] Panel A: Health at birth Child’s birthweight (grams) 3,076.8 -44.6 44.07 37.01 37.21 (45.93) (44.68) (44.44) [0.456] [0.480] [0.464] spaceLow birthweight incidence (%) 20.5 8.7 0.20 0.47 0.42 (2.84) (2.79) (2.77) [0.916] [0.736] [0.980] Panel B: Health in infancy Child was sick in past fortnight 57.4 1.1 -2.32 -2.07 -2.02 (cough/flu/fever/stomach ache) (%) (4.36) (4.29) (4.33) [0.664] [0.656] [0.668] Child’s weight-for-age (WHO z-score) 1.64 -0.30 0.09 0.10 0.10 (0.09) (0.09) (0.09) [0.456] [0.420] [0.460] spaceChild is wasted (%) 33.5 13.4 -3.95 -4.26 -4.45 (3.85) (3.82) (3.83) [0.408] [0.364] [0.404] spaceChild is extremely-wasted (%) 8.7 4.7 -3.61 -3.90 -3.94 (2.04) (1.97) (1.97) [0.148] [0.108] [0.112] Child’s height-for-age (WHO z-score) 1.41 -0.30 0.14 0.15 0.15 (0.07) (0.07) (0.07) [0.072] [0.056] [0.080] spaceChild is stunted (%) 20.7 8.1 -2.58 -2.76 -2.96 (2.32) (2.32) (2.31) [0.188] [0.244] [0.224] spaceChild is extremely stunted (%) 4.6 1.5 -2.55 -2.68 -2.72 (1.25) (1.23) (1.22) [0.064] [0.048] [0.048] Controls (D-in-D’s) Caregiver district and year of birth FE’s YES YES YES Treated cohort interaction with # school-aged children in 1973 (ln scale) NO YES YES Child gender YES YES YES Child-age FE’s YES YES NO Child’s age (linearly, in months) NO NO YES Notes: Sample is 2,118 children. Standard-errors in parentheses are clustered at the district of maternal birth. Wild- bootstrap significance p-values in brackets. Table 3: School Construction and Child Cognitive and School Curriculum Tests: Difference-in-differences estimates. 1966-1970 Maternal Cohort Difference in Differences High INPRES High-Low INPRES Mean Gap [1] [2] [3] Language (school curriculum) [age-specific z-score] 0.08 -0.63 0.154 0.148 0.137 (0.077) (0.082) (0.078) [0.052] [0.096] [0.096] spaceChild in bottom quartile of age-specific language score dist. (%) 39.4 28.9 -10.700 -10.975 -10.921 (3.061) (3.055) (3.110) [0.016] [0.008] [0.008] Mathematics (school curriculum) [age-specific z-score] 0.04 -0.60 0.098 0.097 0.090 (0.119) (0.112) (0.119) [0.524] [0.512] [0.528] spaceChild in bottom quartile of age-specific math score dist. (%) 33.7 25.0 -4.587 -4.938 -4.961 (5.900) (5.959) (6.017) [0.492] [0.556] [0.528] Cognitive reasoning [age-specific z-score] -0.11 -0.48 0.069 0.056 0.055 (0.081) (0.073) (0.078) [0.428] [0.456] [0.520] spaceChild in bottom quartile of age-specific cognitive reasoning score dist. (%) 35.8 16.1 1.872 2.210 2.169 (3.404) (3.267) (3.353) [0.632] [0.600] [0.676] Controls (D-in-D’s) Caregiver district and year of birth FE’s YES YES YES Treated cohort interaction with # school-aged children in 1973 (ln scale) NO YES YES Child gender YES YES YES Child-age FE’s YES YES NO Child’s age (linearly, in months) NO NO YES Notes: Sample is 2,118 children. Standard-errors in parentheses are clustered at the district of maternal birth. Wild-bootstrap significance p-values in brackets. Table 4: School Construction and Child Socio-Emotional Difficulties (SDQ): Difference-in-differences estimates. 1966-1970 Maternal Cohort Difference in Differences High INPRES High-Low INPRES Mean Gap [1] [2] [3] Prosocial behavior score (reversed) [z-score] 0.06 0.32 -0.17 -0.184 -0.183 (0.071) (0.071) (0.071) [0.092] [0.048] [0.072] spaceChild in top quartile of age-specific prosocial difficulty dist. (%) 39.9 14.1 -4.746 -5.155 -5.132 (3.205) (3.131) (3.121) [0.256] [0.208] [0.228] Internalizing score (emotional and peer problems) [z-score] -0.09 -0.19 0.013 0.015 0.016 (0.106) (0.107) (0.106) [0.904] [0.916] [0.928] spaceChild in top quartile of age-specific internalizing difficulty dist. (%) 28.7 -8.9 -1.138 -0.98 -0.915 (5.125) (5.093) (5.013) [0.820] [0.868] [0.868] Externalizing score (behavioral and hyperactivity) [z-score] -0.10 -0.20 -0.037 -0.055 -0.054 (0.091) (0.085) (0.085) [0.692] [0.532] [0.520] spaceChild in top quartile of age-specific externalizing difficulty dist. (%) 30.5 -7.5 -0.778 -1.51 -1.512 (4.807) (4.371) (4.433) [0.920] [0.732] [0.756] Controls (D-in-D’s) Caregiver district and year of birth FE’s YES YES YES Treated cohort interaction with # school-aged children in 1973 (ln scale) NO YES YES Child gender YES YES YES Child-age FE’s YES YES NO Child’s age (linearly, in months) NO NO YES Notes: Sample is 2,118 children. Standard-errors in parentheses are clustered at the district of maternal birth. Wild-bootstrap significance p-values in brackets. Table 5: School Construction and Child School Readiness (EDI): Difference-in-differences estimates. 1966-1970 Maternal Cohort Difference in Differences High INPRES High-Low INPRES Mean Gap [1] [2] [3] Physical health and well-being 0.05 0.12 -0.047 -0.039 -0.038 (0.087) (0.086) (0.088) [0.616] [0.644] [0.700] spaceChild in bottom quartile of age-specific physical health and well-being (%) 28.7 -4.0 -3.651 -4.393 -4.319 (4.618) (4.446) (4.655) [0.440] [0.372] [0.368] Social competence -0.07 -0.19 0.198 0.219 0.217 (0.069) (0.064) (0.065) [0.012] [0.000] [0.000] spaceChild in bottom quartile of age-specific social competence (%) 30.3 11.1 -7.549 -8.075 -8.081 (2.474) (2.229) (2.223) [0.052] [0.004] [0.024] Emotional maturity 0.09 0.12 0.061 0.086 0.085 (0.062) (0.053) (0.052) [0.332] [0.100] [0.160] spaceChild in bottom quartile of age-specific emotional maturity (%) 25.5 -4.6 -1.077 -2.075 -2.022 (2.666) (2.685) (2.720) [0.688] [0.416] [0.440] Language and cognitive development -0.08 -0.14 -0.029 -0.016 -0.017 (0.084) (0.083) (0.084) [0.700] [0.880] [0.876] spaceChild in bottom quartile of age-specific language/cognitive development (%) 30.5 8.3 -4.322 -4.774 -4.655 (3.301) (3.341) (3.293) [0.264] [0.132] [0.220] Communications and general knowledge -0.03 -0.14 0.033 0.045 0.043 (0.072) (0.072) (0.074) [0.628] [0.528] [0.592] spaceChild in bottom quartile of age-specific commun./knowledge (%) 33.0 1.6 3.054 2.413 2.55 (3.749) (3.824) (3.692) [0.456] [0.600] [0.584] Controls (D-in-D’s) Caregiver district and year of birth FE’s YES YES YES Treated cohort interaction with # school-aged children in 1973 (ln scale) NO YES YES Child gender YES YES YES Child-age FE’s YES YES NO Child’s age (linearly, in months) NO NO YES Notes: Sample is 2,118 children. Standard-errors in parentheses are clustered at the district of maternal birth. Wild-bootstrap significance p-values in brackets. Table 6: School construction, mating pattern, fertility and household wealth: Difference-in- differences estimates. Difference in Differences [1] [2] Father not in household (includes deceased) (Yes=1) 0.019 0.002 (0.035) (0.028) [0.696] [0.952] Father’s age (in years) [if present] -0.759 -0.941 (1.119) (1.038) [0.680] [0.652] Father’s schooling (in years) [if present] 0.661 0.733 (0.334) (0.277) [0.204] [0.068] spaceFather completed primary ed. (Yes=1) [if present] 0.084 0.085 (0.033) (0.030) [0.060] [0.080] Mom’s total number of live births -0.055 0.013 (0.227) (0.200) [0.844] [0.972] Mom currently working outside household (Yes=1) -0.030 -0.031 (0.044) (0.051) [0.520] [0.584] Household wealth index (z-score) 0.154 0.178 (0.084) (0.085) [0.116] [0.080] spaceHousehold in bottom quartile of wealth dist. (Yes=1) -0.089 -0.105 (0.035) (0.039) [0.040] [0.032] Controls (D-in-D’s) Caregiver district and year of birth FE’s YES YES Treated cohort interaction with # school-aged children in 1973 (ln scale) NO YES Notes: Standard-errors in parentheses are clustered at the district of birth. Sample is 2,118 observations, except for estimates conditional on father presence (1,926 observations). Wild-bootstrap significance p-values in brackets. Table 7: School Construction and Parenting: Difference-in-differences estimates. 1966-1970 Maternal Cohort Difference in Differences High INPRES High-Low INPRES Mean Gap [1] [2] [3] Parenting warmth -0.03 -0.20 0.002 0.005 0.008 (0.062) (0.063) (0.063) [0.976] [0.896] [0.896] spaceParenting in bottom quartile of age-specific warmth scale (%) 35.1 9.8 -0.895 -0.963 -1.181 (3.055) (3.126) (3.060) [0.828] [0.728] [0.764] Parenting consistency -0.13 -0.23 -0.042 -0.045 -0.046 (0.104) (0.104) (0.104) [0.704] [0.712] [0.692] spaceParenting in bottom quartile of age-specific consistency scale (%) 33.9 9.0 1.193 1.089 1.056 (2.940) (2.932) (2.930) [0.704] [0.696] [0.704] Parenting hostility (reversed) 0.05 0.05 0.102 0.107 0.107 (0.109) (0.110) (0.110) [0.428] [0.360] [0.344] spaceParenting in bottom quartile of age-specific reversed hostility scale (%) 34.4 1.6 -8.999 -9.548 -9.303 (3.886) (4.338) (3.929) [0.064] [0.080] [0.056] Controls (D-in-D’s) Caregiver district and year of birth FE’s YES YES YES Treated cohort interaction with # school-aged children in 1973 (ln scale) NO YES YES Child gender YES YES YES Child-age FE’s YES YES NO Child’s age (linearly, in months) NO NO YES Notes: Sample is 2,118 children. Standard-errors in parentheses are clustered at the district of maternal birth. Wild-bootstrap significance p-values in brackets. Table 8: School Construction and Child Activities Inventory: Difference-in-differences esti- mates. 1966-1970 Maternal Cohort Difference in Differences High INPRES High-Low INPRES Mean Gap [1] [2] [3] Child read books/magazines last week (Yes=1) 0.84 0.00 -0.018 -0.014 -0.014 (0.035) (0.035) (0.035) [0.616] [0.716] [0.708] Child told tale/story last week (Yes=1) 0.40 0.17 -0.05 -0.045 -0.044 (0.032) (0.031) (0.031) [0.128] [0.164] [0.192] Child drew last week (Yes=1) 0.91 0.05 -0.034 -0.031 -0.029 (0.016) (0.016) (0.017) [0.096] [0.108] [0.124] Child played music last week (Yes=1) 0.68 -0.09 -0.091 -0.082 -0.081 (0.041) (0.039) (0.039) [0.020] [0.036] [0.028] Child played with toys last week (Yes=1) 0.90 0.01 -0.076 -0.070 -0.069 (0.026) (0.026) (0.026) [0.072] [0.084] [0.100] Child did household chores last week (Yes=1) 0.59 -0.07 -0.022 -0.003 -0.004 (0.030) (0.035) (0.035) [0.496] [0.868] [0.864] Child played outdoors last week (Yes=1) 0.86 -0.09 0.018 0.011 0.011 (0.020) (0.021) (0.021) [0.404] [0.656] [0.712] Controls (D-in-D’s) Caregiver district and year of birth FE’s YES YES YES Treated cohort interaction with # school-aged children in 1973 (ln scale) NO YES YES Child gender YES YES YES Child-age FE’s YES YES NO Child’s age (linearly, in months) NO NO YES Notes: Sample is 2,118 children. Standard-errors in parentheses are clustered at the district of maternal birth. Wild- bootstrap significance p-values in brackets. Table 9: School Construction and Human Capital Investments: Difference-in-differences estimates. 1966-1970 Maternal Cohort Difference in Differences High INPRES High-Low INPRES Mean Gap [1] [2] [3] Panel A: Health Child has been immunized 0.83 -0.12 -0.012 -0.011 -0.011 (0.023) (0.023) (0.023) [0.704] [0.612] [0.584] Child receives minimum dietary diversity (Yes=1) 0.95 -0.01 -0.008 -0.008 -0.008 (0.013) (0.013) (0.013) [0.624] [0.656] [0.596] Child receives maximum dietary diversity (Yes=1) 0.32 -0.02 0.08 0.082 0.082 (0.062) (0.063) (0.062) [0.408] [0.464] [0.512] Panel B: Education Mother knows closest day-care/pre-school (Yes=1) 0.66 -0.30 0.08 0.081 0.081 (0.029) (0.029) (0.030) [0.008] [0.024] [0.012] Child ever attended day-care/pre-school (Yes=1) 0.28 -0.17 0.041 0.053 0.055 (0.035) (0.035) (0.035) [0.236] [0.176] [0.108] Months child attended day-care/pre-school 4.04 -2.62 1.009 1.227 1.265 (0.501) (0.496) (0.501) [0.088] [0.072] [0.036] Controls (D-in-D’s) Caregiver district and year of birth FE’s YES YES YES Treated cohort interaction with # school-aged children in 1973 (ln scale) NO YES YES Child gender YES YES YES Child-age FE’s YES YES NO Child’s age (linearly, in months) NO NO YES Notes: Sample is 2,118 children. Standard-errors in parentheses are clustered at the district of maternal birth. Wild- bootstrap significance p-values in brackets. Figures Figure 1: Map of districts in overall data and in analytical sample Notes: Panel A shows the nine districts (red) in the Indonesia ECED Data. Mothers resided in these districts in 2013. Panel B shows the birth districts (blue) of all mothers in the Indonesia ECED Data. 171 districts are represented. Panel C shows the maternal birth districts (blue) in the Indonesia ECED Data with at least one mother born between 1966 and 1970, and at least one mother born between 1971 and 1975. 50 districts are represented. Geographic boundaries in the maps are based on 2010 boundaries in Panel A and 1995 boundaries in Panel B and Panel C. Figure 2: Child Health by Maternal Literacy Notes: Figures based on 7,982 observations in ECED. Vertical lines (in red) represent cut-offs for wasting/stunting (-2) and extreme wasting/stunting. Figure 3: Child Performance in Cognitive and Curriculum Tests versus Maternal Education Notes: Figures based on 7,982 observations in ECED. Lines are local-polynomial fits of degree one. Figure 4: Child’s Strengths and Difficulties versus Maternal Education Notes: Figures based on 7,982 observations in ECED. Lines are local-polynomial fits of degree one. Figure 5: Childs Early Development versus Maternal Education Notes: Figures based on 7,982 observations in ECED. Lines are local-polynomial fits of degree one. EDI is average of mother-reported and teacher-reported scores. Low INPRES district High INPRES district 80 75 Birth district HDI 70 65 Downward Upward Downward Upward mobility mobility mobility mobility -10 -5 0 5 -10 -5 0 5 Median 2013 district HDI - Birth district HDI Figure 6: Migration of mothers across districts Notes: Each dot is a birth district in our analysis. The INPRES intensity of each birth district is classified as either low intensity (left) or high intensity (right). The y-axis position of each birth district is determined by the district’s Human Development Index (HDI). Higher values of HDI indicate better conditions. The x-axis position of each birth district is the median difference between the HDI of the district where the mother is living in 2013 and the HDI of the district of birth. X-axis greater than 0 indicates upward mobility: the HDI of the district in adulthood is better than the HDI of the district of birth. X-axis less than 0 indicates downward mobility: the HDI of the district in adulthood is worst than the HDI of the district of birth. X-axis of 0 indicates no change in mobility. HDI is measured in 2013 using the Indonesia Database For Policy And Economic Research (INDO-DAPOER). .22 .2 .18 .16 Difference−in−Differences .14 .12 .1 .08 .06 .04 .02 0 −.02 1+ 2+ 3+ 4+ 5+ 6+ 7+ 8+ 9+ + + + + + 10 11 12 13 14 Educational attainment (years) Figure 7: Difference-in-Differences: Cumulative distribution of maternal education. Notes: Figures based on 2,118 observations in ECED. These are point estimates of the difference-in-differences coefficient in a sequence linear probability models. All models control for (ln) of school-aged population in maternal district of birth around 1973 (interacted with ”treated cohort“). Appendix Mama Knows (and Does) Best: Maternal education and child development in Indonesia by Hasan, Amer; Nozomi Nakajima and Marcos A. Rangel Table A1: Maternal Birth-District Characteristics. Full INPRES data Working sample [Duflo, 2001] Mean SD N Mean SD N Intensive INPRES intervention (Yes=1) 0.54 287 0.44 50 District school enrollment in 1971 (per population) 17.8 (9.70) 287 16.5 (7.77) 50 Schools planned (count) 222.6 (172.90) 287 316.2 (166.56) 50 Schools planned (per 1,000 children) 2.10 (1.07) 287 2.03 (0.88) 50 space Schools planned (per 1,000 children) in Intensive Districts 2.77 (1.03) 155 2.46 (0.50) 22 space Schools planned (per 1,000 children) in Non-Intensive Districts 1.32 (0.30) 132 1.36 (0.23) 28 Notes: Working sample restricts birth districts to be those of mothers either born between 1996 and 1970 or between 1971 and 1975. “Intensive INPRES” districts are defined as having school planned per 1,000 children above the equivalent indicator but measured at the national level (1.8 per 1,000). Table A2: Determinant of INPRES Construction Placement District-level. Full INPRES data [Duflo,2001] Working sample Ln(# schools planned) Ln(# schools planned) [1] [2] [3] [4] [5] [6] Ln(Children in District in 1973) 0.77 0.75 0.75 0.81 0.79 0.81 (0.026) (0.025) (0.028) (0.072) (0.073) (0.074) Ln(Enrollment rate in 1971) -0.33 -0.32 -0.50 -0.47 (0.074) (0.075) (0.110) (0.109) Ln(Child-population density) -0.01 -0.03 (0.012) (0.014) R-squared 0.76 0.78 0.78 0.74 0.82 0.83 N 287 287 287 50 50 50 Notes: Standard-errors in parentheses are based on infinitesimal jackknife procedure (Eicker-White Robust). Table A3: Sample characteristics: Mothers, children, fathers and households. ECED sample Working sample Mean SD Mean SD Maternal demographics and education Mother’s age 34.6 (6.46) 41.3 (2.66) spaceMother was born 1966-1970 (Yes=1) 0.09 0.32 spaceMother was born 1971-1975 (Yes=1) 0.20 0.68 Mother’s schooling (years) 7.6 (3.53) 7.4 (3.94) spaceMother has complete primary ed. (Yes=1) 0.83 0.79 spaceMother can read (Yes=1) 0.87 0.80 Focus child demographics Child’s age 7.6 (0.92) 7.7 (0.91) Child is male (Yes=1) 0.50 0.50 Family composition and wealth Father not in household (includes deceased) (Yes=1) 0.09 0.09 space Father’s age (if in household) 39.7 (7.60) 45.9 (5.61) space Father was born 1966-1970 (Yes=1) 0.18 0.38 space Father was born 1971-1975 (Yes=1) 0.25 0.24 space Father’s schooling (if in household) 7.9 (3.63) 7.7 (4.04) space Father completed primary ed. (if in household) (Yes=1) 0.85 0.79 Sibship size (includes focus child) 4.1 (1.89) 4.3 (1.96) Household wealth index (z-score) 0.12 (0.91) 0.18 (0.95) spaceBottom quartile of wealth (%) 25.0 24.9 N 7,982 2,118 Table A4: Child health outcomes. Working sample Mean SD Child’s birthweight (grams) 3,104.7 (534.6) spaceChild’s low-birthweight (Yes=1) 0.17 Child was sick in past fortnight (cough/flu/fever/stomach ache) (Yes=1) 0.57 Child’s weight-for-age (WHO z-score) -1.55 (1.10) spaceChild is wasted (Yes=1) 0.30 spaceChild is extremely wasted (Yes=1) 0.08 Child’s height-for-age (WHO z-score) -1.31 (0.89) spaceChild is stunted (Yes=1) 0.19 spaceChild is extremely stunted (Yes=1) 0.03 N 2,118 Table A5: Child cognitive/curricular materials outcomes, socio-emotional/behavioral devel- opment and school readiness Working sample Mean SD Panel A: Reasoning/curricular materials Language score (% correct) 64.7 (28.5) Math score (% correct) 67.5 (29.1) Cognitive reasoning score (% correct) 44.1 (24.4) Panel B: Strengths and Difficulties Questionnaire (SDQ) - higher scores mean more difficulties Prosocial behavior score (reversed) [0-10] 2.61 (1.81) Internalizing score (emotional and peer problems) [0-20] 6.09 (2.98) Externalizing score (behavioral problems and hyperactivity) [0-20] 7.51 (2.57) Panel C: School Readiness (EDI) - higher scores mean more developed child Physical health and well-being [0-10] 9.32 (0.93) Social competence [0-10] 7.59 (1.63) Emotional maturity [0-10] 7.02 (1.43) Language and cognitive development [0-10] 9.21 (1.54) Communication and general knowledge [0-10] 6.27 (2.19) N 2,118 Table A6: Parenting and Activity Inventory Working sample Mean SD Panel A: Parenting Warmth (share frequent) 0.68 (0.33) space Lowest Quartile of Warmth Distr. (%) 31.9 (46.6) Consistency (share frequent) 0.35 (0.21) spaceLowest Quartile of Consistency (%) 30.7 (46.2) Hostility (reversed) (share frequent) 0.83 (0.17) spaceLowest Quartile of Hostility (reversed) (%) 35.4 (47.8) Panel B: Child activities Child read books/magazines (Yes=1) 0.84 (0.37) Child told tale/story (Yes=1) 0.32 (0.47) Child drew (Yes=1) 0.89 (0.31) Child played music/sang (Yes=1) 0.70 (0.46) Child played with toys (Yes=1) 0.89 (0.32) Child did household chores (Yes=1) 0.64 (0.48) Child played outdoors (Yes=1) 0.88 (0.32) N 2,118 Table A7: Human Capital Investments Working sample Mean SD Panel A: Health Child has been immunized (Yes=1) 0.87 (0.34) Child receives minimum dietary diversity (Yes=1) 0.96 (0.20) Child receives maximum dietary diversity (Yes=1) 0.34 (0.47) Panel B: Education Mother knows closest day-care/pre-school (Yes=1) 0.77 (0.42) Child ever attended day-care/pre-school (Yes=1) 0.37 (0.48) Months child attended day-care/pre-school 5.34 (8.10) N 2,118 A. B. C. D. Figure A1: Year-by-year difference-in-differences estimates of primary school completion by women (relative to 1951-1955 cohorts) Notes: 99% confidence intervals displayed. Panels A and B estimate based on full sample of 287 districts of birth represented in Duflo (2001). Panels C and D estimate based on the 50 birth districts represented in our main analysis sample. Due to limitations imposed by maternal age in 2013 we base our estimates on a contrast cohort that was likely partially affected by INPRES, which Duflo (2001) completely removed from her analysis. Additional variation seen in this figures motivates our design.