The Journal of Development Studies ISSN: 0022-0388 (Print) 1743-9140 (Online) Journal homepage: http://www.tandfonline.com/loi/fjds20 Educated Mothers, Well-Fed and Healthy Children? Assessing the Impact of the 1980 School Reform on Dietary Diversity and Nutrition Outcomes of Zimbabwean Children Marshall Makate & Clifton Makate To cite this article: Marshall Makate & Clifton Makate (2017): Educated Mothers, Well-Fed and Healthy Children? Assessing the Impact of the 1980 School Reform on Dietary Diversity and Nutrition Outcomes of Zimbabwean Children, The Journal of Development Studies, DOI: 10.1080/00220388.2017.1380796 To link to this article: http://dx.doi.org/10.1080/00220388.2017.1380796 View supplementary material Published online: 25 Oct 2017. Submit your article to this journal Article views: 18 View related articles View Crossmark data Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=fjds20 Download by: [UAE University] Date: 26 October 2017, At: 22:38 The Journal of Development Studies, 2017 https://doi.org/10.1080/00220388.2017.1380796 ARTICLE Downloaded by [UAE University] at 22:38 26 October 2017 Educated Mothers, Well-Fed and Healthy Children? Assessing the Impact of the 1980 School Reform on Dietary Diversity and Nutrition Outcomes of Zimbabwean Children MARSHALL MAKATE * & CLIFTON MAKATE ** *Department of Economics, University at Albany State University of New York, Albany, NY, USA, **UNEP-Tongji Institute of Environment for Sustainable Development, Tongji University, Shanghai, China (Original version submitted January/2017; Final version accepted August/2017) ABSTRACT We scrutinise the causal influence of schooling on child dietary diversity and nutrition in Zimbabwe using the exogenous variability in schooling prompted by the 1980 education policy, a natural trial fitting a fuzzy regression discontinuity design. We established that a one-year of learning promotes dietary diversity and nutrition even after accounting for plausible mediating factors. Also, education is more liable to impact dietary practices and nutrition through improvements in health knowledge, literacy, wealth and prenatal care utilisation. These findings suggest that promoting schooling access to girls in resource-poor nations might have far-reaching implications on feeding practices and consequently child nutrition. 1. Introduction Theoretical and empirical research has strongly linked education to nonmarket outcomes (Grossman, 2006). Nevertheless, much of the already established evidence pointing to a positive association between education and health (Chou, Liu, Grossman, & Joyce, 2010; De Walque, 2007; Lindeboom, Llena-Nozal, & van der Klaauw, 2009; Lleras-Muney, 2005; Lundborg, Nilsson, & Rooth, 2014; Mccrary & Royer, 2011) has largely focused on developed countries with less emphasis on less-industrialised nations. Moreover, there is less agreement in the empirical literature on the causal nature of this fundamental relationship (Chou et al., 2010; Cutler & Lleras-Muney, 2010). A fair portion of the previous literature has concentrated on uncovering the plausible connections between educational accomplishments and child health outcomes (Chou et al., 2010; Currie & Moretti, 2003; Grossman, 2006; Lindeboom et al., 2009; Silles, 2015). Policy efforts in several low-income nations has revolved around ensuring a well-schooled and nourished population with specific focus on the underprivileged segments. Thus, it becomes essential to evaluate whether there exist any plausible and positive externalities of parental schooling on child health outcomes. Correspondence Address: Marshall Makate, Department of Economics, University at Albany State University of New York, 1400 Washington Avenue, Albany, NY 12222, USA. Email: [email protected] © 2017 Informa UK Limited, trading as Taylor & Francis Group Downloaded by [UAE University] at 22:38 26 October 2017 2 M. Makate & C. Makate This study examines the fundamental link between mother’s schooling and child dietary diversity – an important correlate of child nutritional status that has received little attention in the empirical literature. We also assess the possible influence of education on child nutrition outcomes – one of the primary targets of the Sustainable Development Goals (SDG) (United Nations, 2015). A lucid appreciation of the plausible links through which education influences dietary diversification and child nutrition outcomes is of particular importance to Zimbabwe, a country currently struggling to meet its public health obligations (Who, 2015). According to the 2015 Zimbabwe Demographic and Health Survey (ZDHS), nearly 27 per cent of children under five years were considered short for their age (that is, stunted), 3 per cent were wasted (that is, thin relative to their height), 8 per cent were underweight (that is, considered thin for their age), and approximately 6 per cent were considered overweight. Given the already established connections between malnutrition and child mortality (Nangalu, Pooni, Bhargav, & Bains, 2016; Pelletier and Frongillo Jr, 1995) and between poor infant and child feeding practices with chronic malnutrition (Reinbott et al., 2016), it becomes imperative for us to comprehend the correlates of dietary diversity. Besides, dietary diversity reflects on dietary quality, nutrient adequacy and food security (Mango, Zamasiya, Makate, Nyikahadzoi, & Siziba, 2014; Mirmiran, Azadbakht, & Azizi, 2006). Theoretically, it is likely to be inversely correlated with child malnutrition (Frost, Forste, & Haas, 2005). The health economics literature offers several explanations of why education might impact health (Grossman, 2006). Two leading accounts point to the productive and allocative efficiency explanations. The latter contends that highly educated individuals have the ability to choose the right mix of health inputs that maximises their health utility and are therefore more likely to accommodate and process new information faster than their less educated equals (Deaton, 2002; Grossman, 1972). Alternatively, the former proposition postulates that improved schooling enhances the marginal product of health, making educated individuals efficient producers of health (Grossman, 1972). In the context of child dietary diversity and nutrition, suppose the school curriculum incorporates lessons on diet and nutrition, then we expect the highly educated to possess superior health knowledge (Johnston, Lordan, Shields, & Suziedelyte, 2015) which consequently contributes to better feeding practices and overall nutrition. Moreover, educated individuals are more likely to be intellectually curious and literate (Cascio & Lewis, 2006), qualities that empower them to understand health information posted through the media, and even the internet, without much difficulty (Bundorf, Wagner, Singer, & Baker, 2006; Wagner, Hu, & Hibbard, 2001). In this paper, we examine the impact of mother’s schooling on child dietary diversity and nutrition outcomes in Zimbabwe using the 1980 education reform that improved secondary school opportunities for children as an identification strategy. This policy provides an ideal setting for us to evaluate the influence of mother’s education on child dietary diversity and nutrition outcomes within a fuzzy regression discontinuity design (RDD). Following previous studies in developing countries (Aguero & Bharadwaj, 2014; Behrman, 2015b; Grépin & Bharadwaj, 2015; Makate & Makate, 2016; Tsai & Venkataramani, 2015), we adopt an identification technique that exploits the exogenous variability in schooling prompted by the reform which augmented secondary schooling prospects for individuals who were 13 years or below in 1980 compared to their counterparts whose age in 1980 fell outside the secondary school starting age (that is, above 13 years). The differences in schooling opportunities created a disconnect in the likelihood of attending secondary school, which we use as an instrumental variable in the empirical framework discussed later. Our contributions to the literature, especially for developing countries, are twofold. First, we furnish new evidence on the causal influence of mother’s schooling on child dietary diversity using a natural trial fitting a fuzzy RDD in which estimates are local average treatment effects and closely resemble the estimates often provided in the programme evaluation literature (Imbens & Angrist, 1994). We examine these issues in the context of a resource-poor country where health care amenities are deficient including deteriorating child health and nutrition outcomes (WHO, 2015). The empirical approach uses more recent and nationally representative data collected by the ZDHS in 2005/2006, 2010/2011 and 2015. Second, we explore the potential alleyways through which mother’s education might influence child feeding and nutrition outcomes. The rest of the paper is structured as follows: Section two gives the background to the study including the related literature and an overview of the 1980 school reform in Zimbabwe; Section three describes the methods; Section four furnishes the results; Section five provides a discussion of the findings; while Section six concludes. Education, child dietary diversity and nutrition 3 2. Background Downloaded by [UAE University] at 22:38 26 October 2017 2.1 Related literature The nexus between education and nonmarket outcomes has received huge attention in the empirical literature (Grossman, 2015). While the focus of early work has mainly been on the establishment of plausible correlations linking education and nonmarket outcomes, many recent studies have credibly uncovered plausible causal effects of schooling on numerous nonmarket outcomes in the developed world (Grossman, 2006). Even though previous studies agree on the positive association between schooling and overall wellbeing in general, empirical evidence on the causal connections is far from conclusive (de Walque, 2007; Lindeboom et al., 2009; Lleras-Muney, 2005). Moreover, caveats still remain in the context of low-income countries where evidence on the causal impact of schooling is somewhat still less developed (Lochner, 2011). Education influences child health and nutrition outcomes through various pathways including improved health knowledge, health investments and social economic status among others (Grossman, 2006). It’s plausible that higher educational attainments are more likely to enhance ageappropriate and quality diets essential for growth, nurturing and feeding practices. Parental education is widely regarded as an important determinant of child nutritional status regardless of the food security status of the population (Shi & Zhang, 2011), as inadequate knowledge on proper feeding practices may result in poor nutritional intakes and consequently to malnutrition (Reinbott et al., 2016). Alternatively, food insecurity can be a major constraint for caregivers to make use of gained knowledge, as food availability, affordability and utilisation within the household are directly linked to diets of younger children (Black et al., 2008; Mcdonald et al., 2015). Studies for developing countries probing the causal influence of schooling on health outcomes are not only scarce, but have generally focused on explaining the plausible links with HIV/AIDS knowledge, fertility preferences, child nutrition and child mortality outcomes (Aguero & Bharadwaj, 2014; Behrman 2015a, b; Desai & Alva, 1998; Duflo, Dupas, & Kremer, 2015; Emina, Kandala, & Inugu, 2009; Grépin & Bharadwaj, 2015; Makate & Makate, 2016; Makoka, 2013; Osili & Long, 2008). Since studies have established that the education of the mother (not the father) is likely to impact child health outcomes (Chou et al., 2010; Lindeboom et al., 2009), this study focuses on the role played by mother’s schooling on altering child feeding and nutrition outcomes. Highly schooled mothers are more likely to have longer birth intervals (Cleland & Van Ginneken, 1988; Mukuria, Cushing, & Sangha., 2005., Makate & Makate, 2016) and have wellnourished children (Makoka, 2013). Also, schooling improves the mother’s knowledge regarding child health (Lindeboom et al., 2009; Makoka, 2013), including the causes, prevention and treatment of diseases (Frost et al., 2005) and promotes health-seeking behaviour during pregnancy (Makate & Makate, 2016, Ruel, Habicht, Pinstrup-Andersen, & Gröhn, 1992). Moreover, children of educated mothers are more likely to be fully immunised and live in hygienic environments (Emina et al., 2009; Makoka, 2013). In the case of developing countries, it is often the case that immunisation programmes are provided through large scale state-funded initiatives which might subdue the overall influence of schooling. Other channels include access to economic resources brought about by a higher education (Adler & Newman, 2002; Glewwe, 1999; Grossman, 2006; Lindeboom et al., 2009). Furthermore, educated mothers especially in less-industrialised nations are more likely to reside in metropolitan areas where they have better access to improved health and sanitation services compared to their rural counterparts (Makoka, 2013). Additionally, Alderman, Hentschel, and Sabates (2003) and Desai and Alva (1998) contend that average years of schooling at the community level might yield positive externalities which helps improve sanitation and medical services and consequently child nutrition within those communities. The biggest challenge faced by previous studies has been on the provision of credible estimates reflecting the causal effects of schooling. Notable studies have adopted instrumental variable methods that use, for example, college openings as an instrumental variable for education (Currie & Moretti, 2003), mandatory schooling laws (Chou et al., 2010; Gathmann, Jürges, & Reinhold, 2015; Grytten, Skau, & Sørensen, 2014; Güneş, 2016; Lundborg et al., 2014; Mccrary & Royer, 2011; Parinduri, 2017) and universal primary 4 M. Makate & C. Makate education policies (Behrman, 2015b; Makate & Makate, 2016; Osili & Long, 2008; Tsai & Venkataramani, 2015). We follow the latter strand of literature in our empirical approximation of the causal impact of schooling on child dietary diversity and nutrition outcomes including the possible mechanisms through which schooling might propagate in Zimbabwe. Downloaded by [UAE University] at 22:38 26 October 2017 2.2 The 1980 education policy in Zimbabwe Zimbabwe conquered British oppression to end several years of misery and bondage, becoming a fully independent state in 1980. Independence meant a step into an unknown yet hopeful future as the native Zimbabwean people embarked on a new chapter which saw the rise of their first and only president to date – Robert Mugabe. The pre-independence period was characterised by huge educational imbalances that Mugabe had vowed to correct if voted into power (Gordon, 1994). Before independence, education was deliberately biased in favour of white children who enjoyed free and compulsory schooling up to the age of 15 years while, for blacks, many bottlenecks were put in place with very few children transitioning to secondary school (Nhundu, 1992). In some cases, children from black families even failed to secure places to enrol in primary school due to a deliberate shortage of enrolment places (Riddell, 1980). Several intellectuals have written about Zimbabwe’s 1980 educational restructuring (Dorsey, 1989; Kanyongo, 2005; Nhundu, 1992). Independence ushered in a new era in the learning system as the newly elected government implemented a wide array of reforms including making primary school free and compulsory for all Zimbabwean children regardless of race or ethnicity. Also, the government implemented policies that ensured automatic grade transitioning from primary to secondary school, removed any barriers prohibiting older children that delayed school enrolment to attend if they coveted to do so (Nhundu, 1992). While it is fair to note that enrolment at both secondary and primary levels increased, the most impressive increase was witnessed in secondary schools (Nhundu, 1992). As enrolment soared, the government embarked on a mega school construction operation to fulfil the rising demand for schools which saw primary (secondary) schools doubling from 2401 to 4291 (177 to 1276) between 1980 and 1986 (Dorsey, 1989). Several of the secondary schools were constructed in the countryside (Ansell, 2002) where approximately 70 per cent of Zimbabwe’s population resides (Zimstat, 2016). The upsurge in schools contributed to a rise in secondary school enrolment from nearly 69,686 to 482,000 representing a staggering 592 per cent increase in student enrolment between 1979 and 1986 (World Bank, 2007). Children in Zimbabwe officially enter primary school at the age of six years and spend seven years of classroom-led instruction. On-time and successful completion of the primary school stage (at about age 13 years) guarantees a smooth transition to secondary learning where they spend an additional four years of intellectual growth and development. Fulfilling the requirements of the secondary school level (form four) guarantees the receipt of the ordinary level (O-level) certificate and ensures eligibility for advanced level studies for about two years. Students who successfully complete the A-level phase and receive appropriate grade or subject points smoothly transition to college or university education. 3. Methods 3.1 Data source We use data sourced from the nationally representative ZDHS conducted in 2005–2006, 2010–2011 and 2015. This data relates to the child nutrition outcomes for women, several years after the 1980 education policy in Zimbabwe was implemented. This data is ideal for our purposes as it is a rich source of reliable and comparable information on the household structure, health, labour market, household wealth and educational attainment for household members and is available for many developing countries including Zimbabwe. This is a cross-sectional survey undertaken by Inner City Fund (ICF) International in partnership with local governmental agencies to primarily collect health information for women of reproductive ages 15–49 and their children. The ZDHS, which incorporates Education, child dietary diversity and nutrition 5 Downloaded by [UAE University] at 22:38 26 October 2017 a nationally representative sample of about 11,000 households or more, uses a two-stage stratified sampling technique grounded in the Zimbabwe population censuses as the sampling frames. The 2005–2006 and 2010–2011 surveys used the 2002 population census while the 2015 used the 2012 population census as the primary sampling frame. We specifically chose the three most recent ZDHS datasets since they contain similar and complete information regarding the specific foods given to children within the 24-hour period before each survey and the anthropometric measurements for the most recent child born in the five years preceding each survey. For an excellent description of the ZDHS, the reader is referred to the most recent DHS final report for Zimbabwe (Zimstat, 2016). To lessen the possibility of bias associated with pooling across multiple surveys, we adopted a similar strategy to adjust the survey probability weights as implemented in Makate and Makate (2016). The analysis here uses the birth recode component file of the ZDHS which contains both mother and childlevel features. We are primarily interested in the data for children who are alive at the time of each survey and whose birth mother was born between 1961 and 1971. 3.2 Child dietary diversity and nutrition outcomes The analysis here considers five outcome variables related to child dietary diversity and nutrition. First, as in Amugsi, Mittelmark, and Oduro (2015), we computed an index to measure child dietary diversity for children six months and older based on a series of questions that each respondent was asked during the ZDHS. These questions solicited information regarding the feeding practices of women for their surviving children as observed at each survey date. Respondents were asked whether they had given their children any of the foods mentioned in the groups (a–n) below within the 24 hours before each survey. These food groups include, as recorded in the ZDHS individual questionnaire (Zimstat, 2016): (a) juice or juice drinks; (b) milk such as tinned, powdered or fresh animal milk; (c) bread, rice, noodles, porridge or other foods made from grains; (d) pumpkin, carrots, squas or sweet potatoes that are yellow or orange inside; (e) white potatoes, white yams, manioc, cassava, or any other foods made from roots; (f) any dark green, leafy vegetables; (g) vitamin A fruits (for example, mangoes, papayas and any other vitamin A fruits); (h) any other fruits or vegetables; (i) liver, kidney, heart or other organ meats; (j) any meat, such as beef, pork, lamb, goat, chicken or duck; (k) eggs; (l) fresh or dried fish or shellfish; (m) any foods made from beans, peas, lentils or nuts; and (n) cheese or other food made from milk. Each question had a response of either ‘yes, gave child’ or ‘no, did not give child. We computed an additive index in which each ‘yes’ (‘no’) response received a score of one (zero) to create a dietary diversity score representing food consumption within the recommended 24-hour reference time period for each child (Kennedy, Ballard, & Dop, 2011; Steyn, Nel, Nantel, Kennedy, & Labadarios, 2006). This score ranged from 0–14, with a score of zero indicating that the child did not receive any of the specific foods mentioned and 14 suggesting that the child received all the mentioned foods (complete diversification). The higher score indicates how varied the foods typically consumed by the child are in a 24hour period, as well as the parent’s economic ability to acquire and consume a variety of such foods (Hoddinott & Yohannes, 2002), holding all other things constant. Since a dietary diversity score of zero is less likely to be plausible, we provided additional analysis presented as Supplementary Materials (see Table A1) to show that our estimates for dietary diversity shown in Table 3 are robust to the exclusion of children with a diversity score of zero. Second, child nutrition is measured using anthropometric indicators: child’s height-for-age (H/A) and weight-for-age (W/A), all expressed in terms of standard deviations as compared to the standard median (z-scores) of the World Health Organization (WHO) reference populations (WHO, 2006). The H/A z-score (henceforth HAZ) is a cumulative linear growth measure which can indicate past chronic inadequacies of nutrition or frequent illness in the case of deficiencies. We computed a binary indicator to reflect ‘stunting’ (that is, HAZ < 2Þ. The W/A z-score (henceforth WAZ), a common yardstick of assessing the magnitude of malnutrition over time, is a measure of the body mass relative to the age (WHO, 2006). We also include a binary indicator to measure underweight (that is, WAZ < 2Þ. 6 M. Makate & C. Makate Downloaded by [UAE University] at 22:38 26 October 2017 3.3 Explanatory variables The explanatory variable of interest, education, is measured using the years of completed schooling and a binary indicator for secondary school completion as observed at each survey date. We particularly focus on the schooling attainments of the primary child care giver (the mother in this case) and not the father since father’s schooling is more likely to be highly correlated with mother’s education as highlighted in Chou et al. (2010). Our models account for several explanatory variables believed to influence child dietary diversity and nutrition outcomes. In later discussions, we consider father’s education as a potential mechanism through which mother’s education might impact child dietary diversity and nutrition. We included controls for the mother’s age including its square and the child’s age as well as its square, all observed at the survey date. Since the extent of dietary diversity is likely to depend on the available financial resources and competition for food within the household, we included controls for household wealth, employment status of the respondent and the proportion of children under the age of five years in each household including a binary indicator for female head of household. Including the indicator for female headed households allows us to account for the potential challenges faced by such families in the quest to balance work and life responsibilities. Also, we included controls for the province of birth of the child as well as an urban residence dummy to account for any possible unobserved geographic heterogeneity in birth environments. Additionally, we included region-specific age fixed effects to control for the intensity of the schooling programme which likely varied across Zimbabwe’s 10 provinces. 3.4 Identification approach Identifying the causal influence of the education policy on the potential outcomes (child dietary diversity and nutrition) ðYi Þis calculated as the difference between the individual’s child health outcome when the person is susceptible to the education programme Yi ð1Þ and the child health outcome of the individual when not susceptible to the education policyYi ð0Þ. In other words, we examine the influence of mother’s education on the dietary diversity and nutrition of her offspring. The main complication in causal inference studies is the fact that we do not have the liberty to observe the two states of the world at the same instance, instead we only observe the outcome variables linked with the receipt of treatment by the individual. Let’s define Pi 2 f0; 1g where Pi ¼ 0 defines non-exposure or non-participation and Pi ¼ 1 susceptibility to the programme. Then, we can define the outcomes of each individual as a function of programme eligibility as follows: Yi ¼ ð1 Pi Þ Yi ð0Þ þ Pi Yi ð1Þ ¼ Yi ð0Þ if Pi ¼ 0 Yi ð1Þ if Pi ¼ 1 (1) The ultimate challenge lies in the construction of a suitable comparison group that captures the most probable outcome in the different treatment states. This study adopts a fuzzy RDD to identify the causal influence of the 1980 education policy on dietary diversity and nutrition outcomes for Zimbabwean children. The underlying idea in a RDD is the realisation that susceptibility to the policy depends on whether an individual lies (by chance) above or below an assignment variable Zi (in this case, the respondent’s age at time of policy enactment). Comparing the individuals on each side of the cut-point allows us to compute the causal influence of schooling free of selection distortion. As noted in Imbens and Lemieux (2008), it’s possible that Zi can be somewhat associated to the outcome variables of the study but this correlation is presumed to be smooth. The identification strategy we adopt in this study is borrowed from several related studies in the case of low-income nations (Aguero & Bharadwaj, 2014; Behrman 2015a, b; Makate & Makate, 2016; Tsai & Venkataramani, 2015). The crux of this strategy lies in the realisation that susceptibility to the policy was determined (at least in large part) by whether the respondent’s age was within the secondary school starting range or otherwise. Thus, we can spell out an individual’s susceptibility to this schooling policy as follows: P i ¼ I ½ Zi c (2) Education, child dietary diversity and nutrition 7 where Zi is the respondent’s age at policy enactment. In the case where we have complete programme compliance, the likelihood of susceptibility to the education policy Pi takes either zero (non-exposure) or one (complete exposure) and is a function of the running variable Zi (that is, the individual’s age at policy enactment). However, since grade repetition is a common phenomenon in many African countries including Zimbabwe (Alderman, Gilligan, & Lehrer, 2012; Nishimura, Yamano, & Sasaoka, 2008), we maintain the imperfect compliance hypothesis throughout the paper. This imperfect compliance allows us to adopt a ‘fuzzy’ and not ‘sharp’ RDD (Angrist, Pischke, & Pischke, 2009). In the case of the fuzzy RDD, we have the following: lim ProbðPi ¼ 1jZi ¼ zÞÞ lim ProbðPi ¼ 1jZi ¼ zÞ z#c z"c (3) where the term lim ProbðPi ¼ 1jZi ¼ zÞ represents the limit in the likelihood of exposure to the policy z#c when the value of z nears c from the right side and likewise lim ProbðPi ¼ 1jZi ¼ zÞ as the value of z Downloaded by [UAE University] at 22:38 26 October 2017 z"c approaches c from the left side. According to Hahn, Todd, and van der Klaauw (2001), the causal impact of schooling can thus be defined as follows: lim E ðY j Z ¼ zÞ lim E ðY j Z ¼ zÞ β1 ¼ z#c z"c lim E ðP j Z ¼ zÞ lim E ðP j Z ¼ zÞ z#c (4) z"c Equation (4) expresses the discontinuity in the outcome variable as a ratio of the discontinuity in the policy exposure variable (Hahn et al., 2001). To interpret β1 as causal, several assumptions outlined here (Jacob, Zhu, Somers, & Bloom, 2012; Van Der Klaauw, 2002) must be satisfied. First, it should be the case that the assignment variable is known and may not be manipulated by the individuals. In our case, this assumption is satisfied since exposure is defined well before the education policy is implemented. Second, it must be the case that there is a random assignment of individuals around the age cut-off to ensure comparability of the younger and older cohorts. Although this assumption is rather difficult to directly test, Mccrary (2008) recommends checking the density of the assignment variable using a histogram or kernel density function. We tested this assumption and provide evidence to suggest that it does hold with our analysis data. Third, other variables relevant for the analysis should not be showing any discontinuities within the analysis interval (Hahn et al., 2001; Van Der Klaauw, 2002). We formally test this assumption through smoothness tests and results show no evidence of any violation. 3.5 Empirical specification Our empirical formulation abstracts from a basic linear regression model in which child dietary diversity or nutrition ðYi Þ is a function of mother’s schoolingðSi Þ, other explanatory variables ðXi Þ, and a random disturbance term ðui Þ and specified as follows: Yi ¼ a0 þ β1 Si þ X 0i β2 þ ui (5) Identifying the causal influence of schooling on dietary diversity and nutrition outcomes is convoluted by the endogeneity nature of schooling (Si ) which if ignored results in misleading inferences (Grossman, 2006). Thus, an ordinary least squares (OLS) estimation of Equation (5) is likely to be biased. To minimise this concern, the empirical approach adopts a two-stage least squares (2SLS) method to compute the local average treatment effect (LATE). As noted in Van Der Klaauw (2002), the 2SLS model uses the dummy indicator for susceptibility to the education policy as an instrumental variable for schooling. Our specification is of the type estimated in previous-related studies (Aguero & Bharadwaj, 2014; Behrman, 2015b; Makate & Makate, 2016; Tsai & Venkataramani, 2015). In the first-stage, we 8 M. Makate & C. Makate regressed years of schooling, Si on the instrumental variable Pi representing exposure to the policy, and other background characteristics. This model is formulated as follows: Si ¼ π 0 þ π 1 Pi þ π 2 Pi ðZ1980 13Þ þ π 3 ð1 Pi Þ ðZ1980 13Þ þ X 0i π 4 þ εi (6) where P ¼ 1j½Z1980 13 is the instrumental variable for education, Z1980 as before, is the age of the individual at the time of implementing the education policy, Z1980 is a vector of controls and includes survey year fixed effects, child’s year of birth fixed effects, province fixed effects, province-specific age fixed effects and an urban residence dummy among other relevant variables and εi is an idiosyncratic disturbance term. Equation (6) is estimated using OLS to compute the predicted years of schooling to be used in the second stage regression taking the following form: Downloaded by [UAE University] at 22:38 26 October 2017 Yi ¼ β0 þ β1 S^i þ β2 Pi ðZ1980 13Þ þ β3 ð1 Pi Þ ðZ1980 13Þ þ X 0i β4 þ i (7) where β2 and β3 are the coefficients for the linear approximations on either the right or left side of the age cut-point. The coefficient of interest in our main model shown in Equation (7) is β1 , a reasonable estimate of the causal impact of schooling if all the assumptions mentioned earlier are satisfied. The crux of the identification strategy lies in the observation that individuals in the younger cohorts (that is, those aged 13 and below in 1980) should be comparable in their observable and unobservable characteristics apart from their education levels. Even though Zimbabwe’s independence in 1980 ushered in other policy initiatives apart from the secondary school expansions (Thomas & Maluccio, 1996), these were less likely to have influenced our outcomes in exactly the same way as the education reform as noted here (Grépin & Bharadwaj, 2015). Given that the further we drift from the age cut-point the more precise but biased our estimates become and the narrower this age bandwidth, the more imprecise the estimates become (Lee & Lemieuxa, 2010), we carefully selected our comparison groups bearing this fact in mind. 3.6 Robustness tests To assess the robustness of our estimates, we conducted several specification checks. First, given that there is always a trade-off between bias and precision in selecting the age bandwidth with shorter age bandwidths linked to imprecise but less biased estimates and wider bandwidths allied to precise but more biased estimates (Lee & Lemieuxa, 2010), we considered wider and narrower age bandwidths of 8–20 and 10–18 years in 1980. Second, given that the 1980 education policy was intended to raise secondary school enrolment, we replicated the analysis using a binary indicator for secondary school completion as the instrumented variable. Third, given the potential differences between boys and girls as well as urban and rural communities where many schools were built (Ansell, 2002), we fitted our models disaggregated by the child’s gender and rural/urban residence. Fourth, to minimise potential confounding due to the inclusion of the dummy variable for urban residence, we estimated our models excluding this variable to test the sensitivity of our estimates. 4. Results 4.1 Descriptive statistics Table 1 furnishes the summary statistics of select variables used in the study. We show the descriptive statistics for the control, treatment and partly treated cohorts. The average time spent in school by women in our analysis sample was about 7.49 years with those in the relatively younger (older) cohorts 1967–1971 (1961–1965) having spent an average of 8.42 (7.60) years. Almost 52.60 per cent of the families were headed by females whose average age revolved around 41.71 years as observed at the survey date. Approximately 39.40 per cent of these women read newspapers at least once a week, 36.90 per cent lived in urban communities and came from families averaging about 5.36 people. Education, child dietary diversity and nutrition 9 Table 1. Summary statistics for selected variables used in the analysis Overall Treatment Control Partly treated cohorts cohorts cohorts cohorts (1961–1971)) (1967–1971)) (1961–1965)) (1966) Downloaded by [UAE University] at 22:38 26 October 2017 Variables Mean SD Mean SD Mean SD Mean SD Characteristics for mothers Years of completed schooling 7.492 3.870 8.420 3.353 5.926 4.101 7.600 4.113 Female head of household 0.526 0.499 0.518 0.500 0.532 0.499 0.550 0.498 Age at survey date 41.710 4.310 39.950 4.228 44.261 2.922 43.130 3.857 Age in 1980 13.212 3.090 10.862 1.366 16.907 1.339 14.000 0.000 Employed 0.511 0.500 0.525 0.499 0.480 0.500 0.544 0.499 Frequently read newspapers/magazines 0.394 0.489 0.440 0.496 0.313 0.464 0.418 0.494 General health knowledge Ever heard of oral rehydration therapy 0.550 0.498 0.588 0.492 0.452 0.498 0.607 0.490 Ever heard of any modern family planning 0.465 0.499 0.512 0.500 0.392 0.488 0.439 0.497 method Ever heard of HIV/AIDS 0.947 0.223 0.928 0.258 0.980 0.140 0.944 0.230 Knows when in the ovulatory cycle woman can 0.816 0.388 0.823 0.382 0.793 0.405 0.852 0.356 be pregnant General health knowledge index (1–4) 2.387 0.877 2.473 0.890 2.226 0.823 2.455 0.913 Proportion of children under age five 0.117 0.137 0.123 0.136 0.107 0.136 0.113 0.151 Household size 5.359 2.527 5.250 2.360 5.521 2.718 5.434 2.758 Household wealth (quintiles 1–5) 3.202 1.435 3.226 1.429 3.124 1.445 3.351 1.419 Urban resident 0.369 0.483 0.380 0.485 0.338 0.473 0.417 0.494 Number of mothers 3886 2215 1318 353 Characteristics for children Child is female 0.498 0.500 0.503 0.500 0.491 0.500 0.496 0.500 Any prenatal care 0.934 0.248 0.936 0.245 0.930 0.256 0.931 0.256 Four or more prenatal care visits 0.716 0.451 0.745 0.436 0.644 0.480 0.655 0.479 Number of children 3873 2492 1087 294 Notes: We adjusted all the estimates to be nationally representative. SD = Standard deviation. Estimates for children are based on those children who are 10 years and younger as observed at the survey date. Source: The Zimbabwe Demographic and Health Survey (ZDHS) 2005–2006; 2010–2011; 2015. Many of the interviewed women appeared to be well informed about oral rehydration therapy, modern family planning methods, HIV/AIDS and about their ovulation cycle. The children in our analysis sample are equally split by gender and nearly 93.40 per cent of them were born to mothers who sought prenatal care during their pregnancies. Overall, the younger and older cohorts seem to be somewhat comparable in terms of their features such as family size, child gender, urban residence, proportion of under-five children, women’s age at interview date and female headed households. 4.2 The effects of the schooling reform on schooling The immediate influence of the 1980 education reform in Zimbabwe was a burst in overall secondary school enrolment rates. Since the influence of this reform largely hinged upon the individual’s age at enactment, women in younger cohorts (that is, aged 13 years and younger in 1980) appear to have acquired more years of schooling than their relatively older counterparts at policy enactment, as depicted in Figure 1. The gap in education at age 13 years is quite evident. The two dotted lines demarcate the treatment cohorts (to the left of the first dotted vertical line) and the control cohorts on the right side of the dotted vertical line. As mentioned earlier, we consider girls aged 14 years in 1980 (that is, born in 1966) to have partially profited from the education policy and thus excluded from the analytical sample. In the fuzzy RDD, we use the dummy indicator for age 13 years and below in 1980 as the instrumental variable. Disaggregating the results by rural/urban residence or high/low household wealth shows that women born 1967–1971 were more liable to have acquired more years of schooling than their Downloaded by [UAE University] at 22:38 26 October 2017 10 M. Makate & C. Makate Figure 1. The effect of the 1980 education reform in Zimbabwe on maternal education for the overall sample and subgroups (urban/rural and high/low wealth). Notes: Each dot represents the average years of completed education for each age cohort with a local polynomial fitted for the treatment (age 13 years and younger in 1980) and control (above 13 years in 1980). The vertical dotted lines represent the discontinuity in years of schooling around the age 13 (in 1980) threshold. Source: The Zimbabwe Demographic and Health Surveys, 2005–2006, 2010–2011 and 2015. counterparts born in 1961–1965. Figure 2 also suggests that the 1980 education policy was indeed a nationwide undertaking which impacted girls in the younger cohorts relative to their older counterparts. To better comprehend the influence of the school reform on maternal education, the top section of Table 2 reports the coefficient estimates from regression models with years of schooling as the dependent variable. The results show that women aged 9–19 years in 1980 but excluding those aged 14 years were more liable to have increased their average education by about 1.06 years. Those residing in rural communities and from low-wealth households were liable to have acquired an average of 1.60 and 1.69 years respectively and the results are statistically significant at the 1 per cent significance level. The bottom section of Table 2 reveals that the 1980 education policy enhanced the prospects of completing secondary school. In all the regressions, we included linear age trends below and above the age cut-off, indicators for the survey year, province of residence, urban residence and household wealth (where applicable). Overall, the results suggest that the education policy significantly impacted females living in the countryside and from mostly poor backgrounds. A quick check of the instrumental variable reveals that our chosen instrument performs reasonably well. The F-test statistics reported in Table 3 varied from 11.75 to 14.80 and were all statistically significant with p-values ranging from 0.014 to 0.017 suggesting a fairly strong instrumental variable as noted in Staiger and Stock (1997). 4.3 The effect of maternal schooling on child dietary diversity and nutrition We illustrate in Figure 3 the average child outcomes for women in the 1967–1971 and 1961–1965 cohorts. Figure 3 shows that women born in 1967–1971 appear to have relatively Downloaded by [UAE University] at 22:38 26 October 2017 Education, child dietary diversity and nutrition 11 Figure 2. Secondary school completion (or higher) by province of residence. Notes: 1 = Manicaland, 2 = Mashonaland Central, 3 = Mashonaland East, 4 = Mashonaland West, 5 = Matabeleland North, 6 = Matabeleland South, 7 = Midlands, 8 = Masvingo, 9 = Harare, 10 = Bulawayo. As before, the x-axis represents the age of the respondent in 1980. higher dietary diversity scores than their counterparts born in 1961–1965. Also, children born to mothers in the younger cohorts appear to have relatively higher height-for-age and weight-forage z-scores compared to their older counterparts with the disconnect in the outcomes coinciding with the age 13 threshold. The impact on stunting and underweight rates seem to be less clearcut. One limitation of the graph is the fact that we cannot ascertain the extent to which the influence of schooling was causal. Thus, we proceed to Table 3 which shows the 2SLS regression estimates. The main results of our analysis are presented in Table 3. The top-most part of Table 3 shows the baseline OLS regression estimates (without addressing endogeneity bias) while the 2SLS estimates are furnished in the middle and bottom sections of the table. The baseline estimates reveal a positive correlation between maternal schooling and child dietary diversity, height-for-age (HAZ) and weightfor-age (WAZ) z-scores and negatively correlates with stunting and underweight. Specifically, we noted that a one-year rise in education implies a 0.05-unit increase in dietary diversity and increases HAZ and WAZ by 0.04 and 0.03 standard deviations, respectively. Also, increasing schooling by one year is allied with decreasing stunting rates by nearly 1.20 percentage points and statistically significant at the 5 per cent level. Comparing the OLS to the 2SLS estimates reveals that the 2SLS estimates are much larger. This observed discrepancy can first be attributed to the fact that the OLS estimates did not account for potential endogeneity of education. Also, the fact that the 2SLS estimator measures the local average treatment effect (LATE) as contrasted with the average treatment effect (ATE), the focus of the OLS estimator might account for this discrepancy (Angrist, Imbens, & Rubin, 1996). The standard tests for the first stage instrument strength suggest that our instrumental variable is less likely to be weak since the F statistics range from 11.75 to 14.80, above the minimum standard of 10 (Staiger & Stock, 1997). Also, the corresponding p-values ranging from 0.014 to 0.015 all support our claim of a strong instrumental variable. Downloaded by [UAE University] at 22:38 26 October 2017 12 M. Makate & C. Makate Figure 3. The 1980 education reform in Zimbabwe and its effects on child dietary diversity and nutrition outcomes. Notes: The vertical dotted lines represent the age 13 (left) and 14 (right) thresholds. To the left (right) of the age 13 threshold lies the treatment (control) cohorts. The 2SLS estimates presented in the bottom section of Table 3 and calculated after accounting for potential endogeneity bias reveal a strong and significant influence of maternal schooling on child dietary diversity and nutrition. Specifically, we found that a one-year increase in school increases the dietary diversity score by about 0.43 units and statistically significant at the 5 per cent level. The 0.43-unit increase 100 rise in dietary diversity scores in the dietary diversity score translates into an 18.09 per cent 0:429 2:371 given that the average score for children in our analysis sample was about 2.371. A one-year increase in schooling improves HAZ by about 0.20 which translates into a 14.90 per cent improvement in HAZ knowing that the average HAZ for the youngsters in our analytical sample was -1.37. Also, mothers who spent an added year in school experience a 0.19-unit improvement in their children WAZ and statistically significant at the 5 per cent level. This 0.19-unit improvement in WAZ roughly represents a 26.28 improvement in WAZ. Our results also show that increasing schooling by one year lowers the plausibility of stunting and being underweight by nearly 5.60 and 3.80 percentage points. Given that nearly 32.80 and 12.40 per cent of the children in our sample were stunted and underweight, the noted changes translate to an approximate 17.07 and 31.67 per cent reduction in stunting and being underweight, respectively. These estimates are all statistically significant at the 1 per cent level. The last panel of Table 3 shows the estimates computed when we consider the binary indicator for secondary school completion as the instrumented variable. The results also indicate that our 2SLS estimates are statistically significant (with the exception for stunting) and weakly robust to a different measure of education. Given that our first-stage F-statistics for these estimates are lower (ranging from 4.89 to 15.09) we concentrate on the estimates in the middle section of Table 3. We also conducted several tests to check the robustness of our main estimates provided in Table 3. The results (only the 2SLS estimates) from these robustness checks are furnished in the Supplementary Materials (see Table A2). First, since the precision of the RDD estimates hinges upon the choice of the age bandwidth (Van Der Klaauw, 2002), with shorter bandwidths reflecting (0.291) (0.106) 1.063*** 7.477 3532 0.421*** 0.502 3528 Overall (age 9–19 in 1980) 0.299* 0.754 1232 (0.151) 0.467*** 0.365 2287 1.598*** 6.364 2293 −0.206 9.535 1239 (0.567) Rural sample Urban sample (0.134) (0.285) 0.472* 0.257 1224 1.689*** 5.342 1262 Low wealth (0.231) (0.315) 0.418*** 0.640 2264 0.519 8.663 2270 High wealth (0.124) (0.293) Notes: ***Significant at 1 per cent level; *significant at 10 per cent level. In parentheses are robust standard errors clustered at two different dimensions, first by the province of residence and the age of the woman in 1980. The analysis sample consists of women aged between 9 and 19 years (inclusive) in 1980 but excludes those aged 14 years. We included controls for the province of residence, survey fixed effects, linear age trends above and below the age 13 threshold and an indicator for urban residence. Binary indicator for age 13 Mean of the dependent variable Observations Completed secondary education Binary indicator for age 13 Mean of the dependent variable Observations 2SLS estimates Table 2. First stage results: the impact of the 1980 education reform on schooling in Zimbabwe Downloaded by [UAE University] at 22:38 26 October 2017 Education, child dietary diversity and nutrition 13 (0.131) 0.429** 11.747 0.014 2.371 8.742*** 2538 15.095 0.002 2.370 (2.294) (0.021) (0.132) 0.053** 0.211 2538 0.204** 13.935 0.015 −1.369 2.151** 978 4.890 0.006 −1.369 0.036* 0.165 978 (2) (0.797) (0.070) (0.016) (0.104) Child height-for-age z-score 0.190** 13.935 0.015 −0.723 2.049* 978 4.890 0.006 −0.718 0.030** 0.038 978 (3) (1.015) (0.062) (0.012) (0.050) Child weight-for-age z-score −0.038* 14.802 0.017 0.124 −0.406* 978 4.890 0.006 0.124 −0.056* 13.935 0.015 0.328 −0.567 978 4.890 0.006 0.327 (0.339) (0.027) −0.004 −0.005 978 (5) (0.197) (0.016) (0.003) (0.019) Child is underweight −0.012** (0.005) −0.074** (0.027) 978 (4) Child is stunted Notes: ***Significant at 1 per cent level; **significant at 5 per cent level; *significant at 10 per cent level. The standard errors for coefficient estimates are shown in parentheses and clustered at two different dimensions (that is, by the mother identifier and her age in 1980). The reported estimates are based on the sample women aged 9–19 years in 1980 but excludes women who were aged 14 years in 1980. In every specification, we included linear age slopes on either side of the age 13 threshold, year of child’s birth fixed effects, child’s gender, survey fixed effects, region fixed effects, region-specific age fixed effects, information access indicator (read newspapers/ magazines), household size, fraction of children under age five in the household, household wealth, mother’s age and its square, indicator for female head of household, employment status and a binary variable for urban residence. OLS estimates Years of completed schooling Completed secondary school Observations 2SLS estimates Years of completed schooling First stage F-statistic P-value Mean of the dependent variable Completed secondary school Observations First stage F-statistic P-value Mean of the dependent variable (1) Child dietary diversity score Table 3. Two stage least squares estimates of the impact of maternal education on child dietary diversity and nutrition outcomes Downloaded by [UAE University] at 22:38 26 October 2017 14 M. Makate & C. Makate Education, child dietary diversity and nutrition 15 Downloaded by [UAE University] at 22:38 26 October 2017 imprecise but less biased estimates, panel A in Table A2 shows the estimates from a specification (Equation ) that uses the cohort of mothers who were aged 10–18 years old in 1980. The results indicate a much stronger influence of maternal schooling on child dietary diversity, HAZ, WAZ, stunting and being underweight. Also, considering a wider age bandwidth of 8–20 years in 1980 reveals a robust impact of education on dietary diversity, HAZ and WAZ. The estimates found when we consider secondary school completion reveals a robust effect on HAZ only. Second, restricting the analytical sample to children aged 6–36 months shows that education only impacts WAZ and stunting with statistical significance with expected but insignificant effects on dietary diversity, HAZ and being underweight. Third, since the influence of schooling is likely to differ by place of residence, we split our sample into rural and urban. The results shown in panels C and D show that education had a much stronger and significant influence on dietary diversity and nutrition outcomes for children living in the countryside. Lastly, disaggregating the analysis by the child’s gender reveals that mother’s education significantly impacts child feeding and nutrition outcomes regardless of the child’s gender. Overall, the results from the sensitivity checks are consistent with our main findings presented in Table 3 and appear to show a mildly robust causal influence of mother’s schooling on child dietary diversity and nutrition outcomes. 4.4 What are the potential mechanisms linking education to child feeding practices and nutrition outcomes? The influence of education on child feeding practices and nutrition outcomes is likely to propagate through several channels. We considered several mechanisms connected to the general health knowledge of the mother, literacy, access to information, father’s education, maternal healthcare utilisation, family size, household wealth and the respondent’s position in the family. The 2SLS estimates for these analyses are furnished in Table 4. The results show that schooling improves general health knowledge regarding the oral rehydration method used for treating diarrhoea in children, modern contraceptive methods, transmission of HIV/AIDS and about when in the ovulation cycle a woman can get pregnant. Also, a year spent in school is more likely to promote newspaper readership and general literacy with statistical significance at the 5 and 1 per cent levels, respectively. Furthermore, we noted that highly educated mothers are more likely to come from relatively affluent families and engage in health-seeking behaviour such as prenatal care utilisation and keeping health cards for their children. We failed to establish any evidence to support pathways connected to female headed families, modern contraceptive usage, paternal schooling and share of children under age five years within each household. The credibility of the RDD estimates lie in the legitimacy of the assumptions we noted earlier (see subsection 3.4) in the study. To enhance the credibility of our estimates, we conducted additional tests to rule out the possibility for spurious associations. We present the results for these analyses in the Supplementary Materials. We first checked whether the respondent’s age in 1980 exhibits any discontinuities and that there are no manipulations. A visual inspection of the histogram in Figure A1 shows no evidence of any manipulation or bunching at the discontinuity point. We also checked for potential discontinuities in the assignment variable (that is, the respondent’s age at policy enactment in 1980) to rule out any discontinuities in other explanatory variables included in our regression models. This important check of the covariates allows us to consistently identify the influence of the school reform free of any plausible breaks in the other variables. If the fuzzy RD estimates are valid, we expect to see smoothness or continuity in these variables as well. The results displayed in Figure A2 show no evidence of any discontinuities in other variables other than the assignment variable. 5. Discussion This study examines the influence of maternal educational attainments on child feeding practices and nutrition outcomes using nationally representative data from Zimbabwe. Using the 1980 education reform as a source of exogenous variability in schooling for identification of the causal effects, we found that additional years of schooling by the mother significantly enhances child dietary diversity 3529 12.057 0.091** (0.029) 3532 12.162 0.071 Female head of household (0.041) 3507 11.638 0.083*** Able to read and write (0.008) 3532 12.162 0.012 Family planning (0.023) 1042 44.471 0.011 0.004 0.011 930 10.436 0.004 3532 12.162 0.004 3532 11.799 0.004 0.004 0.004 0.004 Any prenatal Has health card for Proportion of children under Household wealth care child age five (high) (0.237) 0.026** (0.009) 0.075* (0.034) 0.003 (0.004) 0.064*** (0.013) (0.043) Read newspapers 3182 8.386 0.004 Father’s years of education 0.438 3532 12.162 0.131** Health knowledge Notes: ***Significant at 1 per cent level; **significant at 5 per cent level; *significant at 10 per cent level. Shown in parentheses are standard errors clustered at two different dimensions (that is, the region of residence of the woman and her age in 1980). Reported estimates are based on the sample of women aged 9–19 years in 1980 but excluding those aged 14 years in 1980. Years of completed schooling Observations First stage F-statistic P-value Years of completed schooling Observations First stage F-statistic P-value 2SLS estimates Table 4. Two stage least squares estimates: pathways through which education might impact child dietary diversity and nutrition outcomes Downloaded by [UAE University] at 22:38 26 October 2017 16 M. Makate & C. Makate Downloaded by [UAE University] at 22:38 26 October 2017 Education, child dietary diversity and nutrition 17 and nutrition outcomes. These findings are weakly robust to several sensitivity checks and corroborate the findings in the burgeoning literature for less-industrialised countries that has assessed education’s impact on numerous outcomes including fertility (Behrman, 2015a; Duflo et al., 2015), knowledge related to HIV/AIDS (Aguero & Bharadwaj, 2014; Behrman, 2015b; Tsai & Venkataramani, 2015), child mortality (Grépin & Bharadwaj, 2015; Makate & Makate, 2016) and nutrition (Desai & Alva, 1998; Glewwe, 1999; Thomas, Strauss, & Henriques, 1991; Vollmer, Bommer, Krishna, Harttgen, & Subramanian, 2016). The empirical literature has discussed several ways through which education might impact child health and nutrition outcomes. Highly educated individuals may possess an upper hand in their information processing capability and thus become more efficient in producing good health or rather better at discerning and engaging in healthy behaviours (Grossman, 1972). Additionally, we explored other ways through which education might propagate. The finding that increased educational attainments improves child feeding practices and nutrition is not surprising (Desai & Alva, 1998; Kabahenda, Andress, Nickols, Kabonesa, & Mullis, 2014; Thomas et al., 1991). As we have established, mother’s education is likely to improve general health knowledge (Glewwe, 1999; Johnston et al., 2015), literacy and numeracy skills acquired through classroom learning which ultimately help future mothers to comprehend the benefits of balanced diets to their children. Thus, incorporating school-based health and nutrition education programmes (Tamiru et al., 2016) in Zimbabwean schools might also help enhance general health knowledge, which might consequently improve child feeding practices and nutrition outcomes in the country. Furthermore, we found that the influence of schooling on child dietary diversity and nutrition is more likely to work through prenatal care utilisation during pregnancy. This result is not surprising since pregnant women are more likely to receive extensive and appropriate educational advice on healthy eating, living, rest, exclusive early breastfeeding, advice on the dangers of smoking and alcohol consumption during pregnancy, parenting skills, family planning, birth spacing, feeding practices and where to seek additional care if needed (Lincetto, Mothebesoane-Anoh, Gomez, & Munjanja, 2006). The findings in this paper revealed a discrepancy between the OLS and 2SLS estimates as shown in Table 3. While this observation is consistent with previous empirical studies, it remains a puzzle yet to be explored further. We speculate that the observed discrepancy in the context of Zimbabwe might be attributable to the realisation that the school reforms here were non-compulsory in nature and thus would result in a LATE that is different from that calculated when schooling is mandatory. In this case, it’s possible that other children with high ability could not have access to secondary schooling prior to the reforms. Thus, with the lowered costs of schooling these children might self-select into schooling while those of low abilities might have opted to stay away from school which might be the possible reason why the 2SLS estimates are larger than the OLS estimates measuring the ATE. Our study is not without its limitations. First, we certainly acknowledge the possibility that other post-independence changes that have since taken place in Zimbabwe (Thomas & Maluccio, 1996) might drive our estimates. However, we argue that these changes are less likely to have exerted similar cohort-specific effects as the education policy in Zimbabwe and hence less likely to influence our empirical estimates. Moreover, the last few decades in Zimbabwe have been characterised by poor public policies which saw economic progress in the country decline to intolerable levels. Second, our reliance on cross-sectional data makes it difficult to explore any plausible dynamic or intergenerational aspects of schooling on child wellbeing. Third, we could not assess the extent to which paternal health knowledge might influence maternal knowledge which might impact feeding practices and nutrition outcomes for children. Given that the ZDHS does not capture information on paternal health knowledge, we could not explore this possibility. Nevertheless, we make a vital contribution to the literature on the causal effects of mother’s schooling in low-income regions especially sub-Saharan Africa. Fourth, our analysis is only based on the group of children who are alive as observed at the survey date. Thus, we are unable to consider the nutrition outcomes for those children who are deceased, which might be a source of 18 M. Makate & C. Makate selection bias. Since the ZDHS does not provide this information, we are unable to do address this concern. Lastly, our estimates are only limited to one context – Zimbabwe. Future studies might extend to other contexts and explore other pathways through which mother’s education might influence feeding and nutrition outcomes. Downloaded by [UAE University] at 22:38 26 October 2017 6. Conclusion This study provides empirical evidence that mother’s education improves child dietary diversity and nutrition in Zimbabwe – a nation currently struggling to meet the health and wellbeing of its younger generation. The finding that maternal schooling might propagate through its influence on prenatal care utilisation suggests that policies that promote schooling might have far reaching implications to the overall health and wellbeing of the population in developing countries. The nutritional standing of children in Zimbabwe is projected to worsen (Unicef, 2016), hence the need for extensive research to establish ways through which the trend might be reversed. Our analysis suggests that policies that promote secondary school education in Zimbabwe, especially amongst girls, might have far reaching implications on child feeding practices and nutrition outcomes and consequently improve child health in the country. Acknowledgements This paper is adapted from the PhD dissertation of Marshall Makate, completed at the University at Albany, State University of New York in Albany New York, USA under the guidance of Professor Pinka Chatterji. The authors are extremely grateful to the comments received from Econometrics seminar participants at the University at Albany including from the Editor and anonymous reviewers of the journal who helped improve the quality of this paper.All the data and codes used to generate tables and figures of this paper will be made available upon request. The views expressed in this paper are those of the authors and do not necessarily reflect the views of the institutions of affiliation. All remaining errors are ours. Disclosure statement No potential conflict of interest was reported by the authors. The disclosure statement has been inserted. Please correct if this is inaccurate. Supplementary Materials Supplementary Materials are available for this article which can be accessed via the online version of this journal at https://doi.org/10.1080/00220388.2017.1380796 ORCID Marshall Makate Clifton Makate http://orcid.org/0000-0002-2005-2970 http://orcid.org/0000-0002-6061-6638 References Adler, N. E., & Newman, K. (2002). Socioeconomic Disparities In Health: Pathways And Policies Inequality in education, income, and occupation exacerbates the gaps between the health ‘haves’ and ‘have-nots.’. Aguero, J. M., & Bharadwaj, P. (2014). Do the more educated know more about health? Evidence from schooling and HIV knowledge in Zimbabwe. Economic Development and Cultural Change, 62(3), 489–517. doi:10.1086/675398 Downloaded by [UAE University] at 22:38 26 October 2017 Education, child dietary diversity and nutrition 19 Alderman, H., Gilligan, D. O., & Lehrer, K. (2012). The impact of food for education programs on school participation in Northern Uganda. Economic Development and Cultural Change, 61(1), 187–218. doi:10.1086/666949 Alderman, H., Hentschel, J., & Sabates, R. (2003). With the help of one’s neighbors: Externalities in the production of nutrition in Peru. Social Science & Medicine, 56(10), 2019–2031. doi:10.1016/S0277-9536(02)00183-1 Amugsi, D. A., Mittelmark, M. B., & Oduro, A. (2015). Association between maternal and child dietary diversity: An analysis of the Ghana demographic and health survey. PLoS ONE, 10(8), 1–12. doi:10.1371/journal.pone.0136748 Angrist, J. D., Imbens, G. W., & Rubin, D. B. (1996). Identification of causal effects using instrumental variables. Journal of the American Statistical Association, 91(434), 444–455. doi:10.1080/01621459.1996.10476902 Angrist, J. D., Pischke, J.-S., & Pischke, J.-S. (2009). Mostly harmless econometrics: An empiricist’s companion. Princeton: Princeton university press. Ansell, N. (2002). Secondary education reform in Lesotho and Zimbabwe and the needs of rural girls: Pronouncements, policy and practice. Comparative Education, 38(1), 91–112. Retrieved from http://www.jstor.org/stable/3099673. Behrman, J. A. (2015a). Does schooling affect women’s desired fertility? Evidence from Malawi, Uganda, and Ethiopia. Demography, 52(3), 787–809. doi:10.1007/s13524-015-0392-3 Behrman, J. A. (2015b). The effect of increased primary schooling on adult women’s HIV status in Malawi and Uganda: Universal Primary Education as a natural experiment. Social Science & Medicine, 127, 108–115. doi:10.1016/j. socscimed.2014.06.034 Black, R. E., Allen, L. H., Bhutta, Z. A., Caulfield, L. E., De Onis, M., Ezzati, M., . . . Rivera, J. (2008). Maternal and child undernutrition: Global and regional exposures and health consequences. The Lancet, 371(9608), 243–260. doi:10.1016/ S0140-6736(07)61690-0 Bundorf, M. K., Wagner, T. H., Singer, S. J., & Baker, L. C. (2006). Who searches the internet for health information? Health Services Research, 41(3p1), 819–836. doi:10.1111/hesr.2006.41.issue-3p1 Cascio, E. U., & Lewis, E. G. (2006). Schooling and the armed forces qualifying test evidence from school-entry laws. Journal of Human Resources, 41(2), 294–318. doi:10.3368/jhr.XLI.2.294 Chou, S.-Y., Liu, J.-T., Grossman, M., & Joyce, T. (2010). Parental education and child health: Evidence from a natural experiment in Taiwan. American Economic Journal: Applied Economics, 2(1), 33–61. Cleland, J. G., & Van Ginneken, J. K. (1988). Maternal education and child survival in developing countries: The search for pathways of influence. Social Science & Medicine, 27(12), 1357–1368. doi:10.1016/0277-9536(88)90201-8 Currie, J., & Moretti, E. (2003). Mother’s education and the intergenerational transmission of human capital: Evidence from college openings. The Quarterly Journal of Economics, 118(4), 1495–1532. doi:10.1162/003355303322552856 Cutler, D. M., & Lleras-Muney, A. (2010). Understanding differences in health behaviors by education. Journal of Health Economics, 29(1), 1–28. doi:10.1016/j.jhealeco.2009.10.003 de Walque, D. (2007). Does education affect smoking behaviors? Evidence using the Vietnam draft as an instrument for college education. Journal of Health Economics, 26(5), 877–895. doi:10.1016/j.jhealeco.2006.12.005 Deaton, A. (2002). Policy implications of the gradient of health and wealth. Health Affairs, 21(2), 13–30. doi:10.1377/ hlthaff.21.2.13 Desai, S., & Alva, S. (1998). Maternal education and child health: Is there a strong causal relationship? Demography, 35(1), 71–81. Dorsey, B. J. (1989). Educational development and reform in Zimbabwe. Comparative Education Review, 33(1), 40–58. doi:10.1086/446811 Duflo, E., Dupas, P., & Kremer, M. (2015). Education, HIV, and early fertility: Experimental evidence from Kenya. The American Economic Review, 105(9), 2757–2797. doi:10.1257/aer.20121607 Emina, J. B., Kandala, N. B., & Inugu, J. and Ye, Y. (2009). The effect of maternal education on child nutritional status in the Democratic Republic of Congo. 26th International Population Conference of the International Union for the Scientific Study of Population (IUSSP). (Marrakech, Morocco). Frost, M. B., Forste, R., & Haas, D. W. (2005). Maternal education and child nutritional status in Bolivia: Finding the links. Social Science & Medicine, 60(2), 395–407. doi:10.1016/j.socscimed.2004.05.010 Gathmann, C., Jürges, H., & Reinhold, S. (2015). Compulsory schooling reforms, education and mortality in twentieth century Europe. Social Science & Medicine, 127, 74–82. doi:10.1016/j.socscimed.2014.01.037 Glewwe, P. (1999). Why does mother’s schooling raise child health in developing countries? Evidence from Morocco. The Journal of Human Resources, 34(1), 124–159. doi:10.2307/146305 Gordon, R. (1994). Education policy and gender in Zimbabwe. Gender and Education, 6(2), 131–139. doi:10.1080/ 0954025940060203 Grépin, K. A., & Bharadwaj, P. (2015). Maternal education and child mortality in Zimbabwe. Journal of Health Economics, 44, 97–117. doi:10.1016/j.jhealeco.2015.08.003 Grossman, M. (1972). On the concept of health capital and the demand for health. Journal of Political Economy, 80(2), 223–255. doi:10.1086/259880 Grossman, M. (2006). Education and nonmarket outcomes. Handbook of the Economics of Education, 1, 577–633. Grossman, M., 2015. The relationship between health and schooling: What’s new? (NBER Working Papers: 21609). National Bureau of Economic Research, Inc. Grytten, J., Skau, I., & Sørensen, R. J. (2014). Educated mothers, healthy infants. The impact of a school reform on the birth weight of Norwegian infants 1967–2005. Social Science & Medicine, 105, 84–92. doi:10.1016/j. socscimed.2014.01.008 Downloaded by [UAE University] at 22:38 26 October 2017 20 M. Makate & C. Makate Güneş, P. M. (2016). The impact of female education on teenage fertility: Evidence from Turkey. The BE Journal of Economic Analysis & Policy, 16(1), 259–288. Hahn, J., Todd, P., & van der Klaauw, W. (2001). Identification and estimation of treatment effects with a regression‐ discontinuity design. Econometrica, 69(1), 201–209. doi:10.1111/ecta.2001.69.issue-1 Hoddinott, J., & Yohannes, Y. (2002). Dietary diversity as a food security indicator. Food Consumption and Nutrition Division (FCND). Discussion Paper 136. Imbens, G. W., & Angrist, J. D. (1994). Identification and estimation of local average treatment effects. Econometrica, 62(2), 467–475. doi:10.2307/2951620 Imbens, G. W., & Lemieux, T. (2008). Regression discontinuity designs: A guide to practice. Journal of Econometrics, 142(2), 615–635. doi:10.1016/j.jeconom.2007.05.001 Jacob, R. T., Zhu, P., Somers, M.-A., & Bloom, H. S. (2012). A practical guide to regression discontinuity (Citeseer). Johnston, D. W., Lordan, G., Shields, M. A., & Suziedelyte, A. (2015). Education and health knowledge: Evidence from UK compulsory schooling reform. Social Science & Medicine, 127, 92–100. doi:10.1016/j.socscimed.2014.10.026 Kabahenda, M. K., Andress, E. L., Nickols, S. Y., Kabonesa, C., & Mullis, R. M. (2014). Promoting dietary diversity to improve child growth in less-resourced rural settings in Uganda. Journal of Human Nutrition & Dietetics, 27, 143–151. doi:10.1111/ jhn.2014.27.issue-s2 Kanyongo, G. Y. (2005). Zimbabwe’s public education system reforms: Successes and Challenges. International Education Journal, 6(1), 65–74. Kennedy, G., Ballard, T., & Dop, M. C. (2011). Guidelines for measuring household and individual dietary diversity. Rome: Food and Agriculture Organization of the United Nations. Lee, D. S., & Lemieuxa, T. (2010). Regression discontinuity designs in economics. Journal of Economic Literature, 48(2), 281– 355. doi:10.1257/jel.48.2.281 Lincetto, O., Mothebesoane-Anoh, S., Gomez, P., & Munjanja, S. (2006). Antenatal care. Opportunities for Africa’s Newborns: Practical Data, Policy and Programmatic Support for Newborn Care in Africa, 55–62. Lindeboom, M., Llena-Nozal, A., & van der Klaauw, B. (2009). Parental education and child health: Evidence from a schooling reform. Journal of Health Economics, 28(1), 109–131. doi:10.1016/j.jhealeco.2008.08.003 Lleras-Muney, A. (2005). The relationship between education and adult mortality in the United States. The Review of Economic Studies, 72(1), 189–221. doi:10.1111/roes.2005.72.issue-1 Lochner, L. (2011). Non-production benefits of education: Crime, health, and good citizenship (No. w16722). National Bureau of Economic Research. Lundborg, P., Nilsson, A., & Rooth, D.-O. (2014). Parental education and offspring outcomes: Evidence from the Swedish compulsory School Reform. American Economic Journal: Applied Economics, 6(1), 253–278. Makate, M., & Makate, C. (2016). The causal effect of increased primary schooling on child mortality in Malawi: Universal primary education as a natural experiment. Social Science & Medicine, 168, 72–83. doi:10.1016/j.socscimed.2016.09.003 Makoka, D. (2013). The impact of maternal education on child nutrition: Evidence from Malawi, Tanzania, and Zimbabwe. Calverton, MD: ICF International. Mango, N., Zamasiya, B., Makate, C., Nyikahadzoi, K., & Siziba, S. (2014). Factors influencing household food security among smallholder farmers in the Mudzi district of Zimbabwe. Development Southern Africa, 31(4), 625–640. doi:10.1080/ 0376835X.2014.911694 Mccrary, J. (2008). Manipulation of the running variable in the regression discontinuity design: A density test. Journal of Econometrics, 142(2), 698–714. doi:10.1016/j.jeconom.2007.05.005 Mccrary, J., & Royer, H. (2011). The effect of female education on fertility and infant health: Evidence from school entry policies using exact date of birth. The American Economic Review, 101(1), 158–195. doi:10.1257/aer.101.1.158 Mcdonald, C. M., Mclean, J., Kroeun, H., Talukder, A., Lynd, L. D., & Green, T. J. (2015). Household food insecurity and dietary diversity as correlates of maternal and child undernutrition in rural Cambodia. European Journal of Clinical Nutrition, 69(2), 242–246. doi:10.1038/ejcn.2014.161 Mirmiran, P., Azadbakht, L., & Azizi, F. (2006). Dietary diversity within food groups: An indicator of specific nutrient adequacy in Tehranian women. Journal of the American College of Nutrition, 25(4), 354–361. doi:10.1080/07315724.2006.10719546 Mukuria, A., Cushing, J., & Sangha., J. (2005). Nutritional status of children: Results from the demographic and health surveys 1994-2001. Calverton, MD: ORC Macro, MEASURE DHS+. Nangalu, R., Pooni, P. A., Bhargav, S., & Bains, H. S. (2016). Impact of malnutrition on pediatric risk of mortality score and outcome in Pediatric Intensive Care Unit. Indian Journal of Critical Care Medicine, 20(7), 385–390. doi:10.4103/09725229.186218 Nhundu, T. J. (1992). A decade of educational expansion in Zimbabwe: Causes, consequences, and policy contradictions. The Journal of Negro Education, 61(1), 78–98. doi:10.2307/2295630 Nishimura, M., Yamano, T., & Sasaoka, Y. (2008). Impacts of the universal primary education policy on educational attainment and private costs in rural Uganda. International Journal of Educational Development, 28(2), 161–175. doi:10.1016/j. ijedudev.2006.09.017 Osili, U. O., & Long, B. T. (2008). Does female schooling reduce fertility? Evidence from Nigeria. Journal of Development Economics, 87(1), 57–75. doi:10.1016/j.jdeveco.2007.10.003 Parinduri, R. A. (2017). Does education improve health? Evidence from Indonesia. Journal of Development Studies, 53(9), 1358–1375. Downloaded by [UAE University] at 22:38 26 October 2017 Education, child dietary diversity and nutrition 21 Pelletier, D. L., & Frongillo Jr, E. A. (1995). The effects of malnutrition on child mortality in developing countries. Bulletin of the World Health Organization, 73(4), 443–448. Reinbott, A., Schelling, A., Kuchenbecker, J., Jeremias, T., Russell, I., Kevanna, O., . . . Jordan, I. (2016). Nutrition education linked to agricultural interventions improved child dietary diversity in rural Cambodia. The British Journal of Nutrition, 116 (8), 1457–1468. doi:10.1017/S0007114516003433 Riddell, R. (1980). Education for employment. London: Catholic Institute for International Relations. Ruel, M. T., Habicht, J.-P., Pinstrup-Andersen, P., & Gröhn, Y. (1992). The mediating effect of maternal nutrition knowledge on the association between maternal schooling and child nutritional status in Lesotho. American Journal of Epidemiology, 135 (8), 904–914. doi:10.1093/oxfordjournals.aje.a116386 Shi, L., & Zhang, J. (2011). Recent evidence of the effectiveness of educational interventions for improving complementary feeding practices in developing countries. Journal of Tropical Pediatrics, 57(2), 91–98. Silles, M. A. (2015). The intergenerational effect of parental education on child health: Evidence from the UK. Education Economics, 23(4), 455–469. doi:10.1080/09645292.2013.858660 Staiger, D., & Stock, J. H. (1997). Instrumental variables regression with weak instruments. Econometric Society, 557. doi:10.2307/2171753 Steyn, N., Nel, J., Nantel, G., Kennedy, G., & Labadarios, D. (2006). Food variety and dietary diversity scores in children: Are they good indicators of dietary adequacy? Public Health Nutrition, 9(05), 644–650. doi:10.1079/PHN2005912 Tamiru, D., Argaw, A., Gerbaba, M., Nigussie, A., Ayana, G., & Belachew, T. (2016). Improving dietary diversity of school adolescents through school based nutrition education and home gardening in Jimma Zone: Quasi-experimental design. Eating Behaviors, 23, 180–186. doi:10.1016/j.eatbeh.2016.10.009 Thomas, D., & Maluccio, J. (1996). Fertility, contraceptive choice, and public policy in Zimbabwe. The World Bank Economic Review, 10(1), 189–222. doi:10.1093/wber/10.1.189 Thomas, D., Strauss, J., & Henriques, M.-H. (1991). How does mother’s education affect child height? The Journal of Human Resources, 26(2), 183–211. doi:10.2307/145920 Tsai, A. C., & Venkataramani, A. S. (2015). The causal effect of education on HIV stigma in Uganda: Evidence from a natural experiment. Social Science & Medicine, 142, 37–46. doi:10.1016/j.socscimed.2015.08.009 Unicef. (2016). Zimbabwe Humanitarian Situation Report #6 (Author). United Nations. (2015). Transforming Our World: The 2030 Agenda for Sustainable Development. van der Klaauw, W. (2002). Estimating the effect of financial aid offers on college enrollment: A regression–discontinuity approach. International Economic Review, 43(4), 1249–1287. doi:10.1111/iere.2002.43.issue-4 Vollmer, S., Bommer, C., Krishna, A., Harttgen, K., & Subramanian, S. (2016). The association of parental education with childhood undernutrition in low- and middle-income countries: Comparing the role of paternal and maternal education. International Journal of Epidemiology, dyw133. doi:10.1093/ije/dyw133 Wagner, T. H., Hu, T.-W., & Hibbard, J. H. (2001). The demand for consumer health information. Journal of Health Economics, 20(6), 1059–1075. doi:10.1016/S0167-6296(01)00107-2 WHO. (2006). Multicentre growth reference study group 2006. WHO Child Growth Standards: Length/height-for age, weightfor-age, weight for length, weight for height: Methods and development. Geneva: World Health Organization, pp. 312. WHO. (2015). World health statistics 2015. Geneva: World Health Organization. World Bank. (2007). World Development Indicators 2007. Zimstat. (2016). Zimbabwe Demographic and Health Survey 2015: Final Report. Rockville, MD: Author.