Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

In Utero Seasonal Food Insecurity and Cognitive Development: Evidence on Gender Imbalances From Ethiopia

In Utero Seasonal Food Insecurity and Cognitive Development: Evidence on Gender Imbalances From... Abstract Food insecurity is pervasive and highly seasonal in Ethiopia. In this study, we investigate the effect of seasonal food insecurity on child development. Exploiting the Young Lives Ethiopia dataset, we study the gender-specific impact of in utero exposure to seasonal food insecurity on cognitive development and the probability of being on the expected grade for children of age 8 up to 12. We find that at age 8, in utero exposure to food insecurity negatively affects cognitive development, only for boys. At age 12, such exposure significantly reduces cognitive development for all children, but with a significantly higher magnitude for boys. The impact is almost three times bigger compared to the one estimated for girls. Corroborated with other outcomes, we explain such gender imbalances by the accumulative nature of the scarring effect rather than the culling effect or gender differences in parental investment. 1 Introduction Early cognitive abilities play an important role in determining long-term schooling and wages [Currie & Thomas, 2001]. The development of these skills begins in utero and continues to evolve over the life cycle through a dynamic process of skill formation [Heckman, 2007]. Large-scale shocks such as famine, natural disasters and civil wars experienced during prenatal and early life environment have been found to be strong predictors of future outcomes [Almond & Currie, 2011, Currie & Vogl, 2013]. Nonetheless, food shortages are much more frequent and potentially more detrimental on most children’s life cycle. Each year, more people die from hunger than AIDS, malaria and tuberculosis combined [WFP 2013]. Ethiopia is a case in point. According to FAO [2009], about 44 percent of the total population in Ethiopia were undernourished between 2004 and 2006. This could be attributed to chronic food insecurity, a pervasive phenomenon in the country. A substantial number of people in Ethiopia are facing difficulties in feeding themselves on a regular basis around the rainy and planting seasons. According to the International Food Policy Research Institute and the Ethiopian Development Research Institute, more than 25 percent of households in Tigray region, close to 30 percent of households in Oromia (the most populous region) and 25 percent of households in Southern Nations, Nationalities and Peoples’ (SNNP) region reported food gaps during the rainy season in 2006 [Hoddinott et al., 2011]. For Amhara (the second most populous region) the food gap stands at less than 20 percent.1 In the same year, close to 20 percent and 15 percent of households reported food gaps for 3 months and 4 months, respectively. Such chronic under-nutrition, in particular at an early age, is likely to have long-term consequences in terms of health, schooling and socio-economic outcomes [Alderman et al., 2006, Miller, 2017]. The positive impact of early childhood nutrition on education has also been established [Glewwe et al., 2001, Maluccio et al., 2009]. The impact of prenatal exposure to seasonal food insecurity is largely unknown. In this study, we examine the impact of in utero exposure to seasonal food insecurity on cognitive development and grade-for-age. We exploit a unique dataset from the Young Lives Ethiopia (YLE) study and apply a novel identification strategy. We estimate the effect of variation in the number of days of exposure to prenatal food insecurity on cognitive development outcomes, controlling for community and birth month fixed effects together with child and household characteristics. We find that a standard deviation increase in relative food insecurity exposure in utero results in lower maths achievements score at age 12 by about 0.175 standard deviations. Exposure also decreases the odds of being on the correct educational track. More importantly, we shed light on the gender-specific impact of seasonal food insecurity in utero. We find that there are significant gender imbalances. Both at ages 8 and 12, an in utero shock decreases boys’ maths score more severely than girls’. At age 12, we find that boys are less likely to be on the right grade for their age. Our paper directly relates to the emerging literature exploring the effect of prenatal shock on human capital development of children [Almond et al., 2015, Neelsen & Stratmann, 2011]. The so called ‘foetal origins’ hypothesis advocated by Barker describes that conditions in utero (for instance, nutritional deficiencies) have long lasting health effects [Almond & Currie, 2011, Barker, 1990]. Prenatal nutrition shocks should also have significant detrimental effects on brain development [Almond et al., 2015, Almond & Mazumder, 2011, Umana-Aponte, 2011]. To establish causal effects, studies exploit famines and other shocks like natural disasters, wars and disease epidemics as exogenous natural experiments. Almond & Currie [2011] and Currie & Vogl [2013] provide extensive review of the literature.2 More directly related to the context of our study, there is a large number of studies investigating the impact of seasonality, price shocks and weather shocks on households’ vulnerability and child development in Ethiopia [Abay & Hirvonen, 2017, Alem & Söderbom, 2012, Dercon, 2004, Dercon & Krishnan, 2000, Dercon & Porter, 2014, Hill & Porter, 2017, Miller, 2017, Porter, 2012]. However, this literature has not considered the individual exposure to shock in utero, except for Dercon & Porter [2014] and Miller [2017]. Dercon & Porter [2014] find detrimental impact of the 1984/85 Ethiopian famine on height of young adults. However, no effect is found from exposure in utero. On the contrary, Miller [2017] finds significant effects of seasonal food scarcity in utero on height at ages 8 and 12, but no significant difference between boys and girls. Our paper extends Miller [2017]’s work by exploring the impact of seasonal food insecurity on cognitive development and by investigating possible gender imbalances in such an impact.3 Boys have been found to be more vulnerable to shocks in utero such as famine [Almond et al., 2010, Hernández-Julián et al., 2014, Roseboom et al., 2011], conflict [Dagnelie et al., 2018, Valente, 2015], alcohol consumption [Nilsson, 2017] or stress from mother’s grief [Black et al., 2016].4 However, the nature of gender imbalances in the effect of in utero and early life shocks on different health and socio-economic outcomes differs across existing studies. While the Great Chinese Famine has been found to be more detrimental for girls in terms of health and education [Luo et al., 2006, Mu & Zhang, 2011], stronger effects on boys have been found from famines during World War II in Greece, Germany and the Netherlands [Berg et al., 2016] and during the Dutch Potato Famine in the mid-19th century [Lindeboom et al., 2010]. Nilsson [2017] also finds stronger effects of in utero exposure to increased alcohol availability on long-term labour market and educational outcomes and cognitive and non-cognitive ability of boys. The differences in the results are puzzling. The use of different outcome variables and contextual differences may be behind the mixed nature of the evidence, but the impact of in utero shocks on outcomes later in life may result from different mechanisms [Dagnelie et al., 2018, Nilsson, 2017, Valente, 2015]. The scarring effects result from a downward shift of the entire foetal health distribution. Since male foetuses disproportionally stand at the low end of that distribution, deficiencies due to the scarring effects may accumulate overtime and explain more detrimental effects for boys later in life. On the contrary, the culling effect directly relates to selective mortality in utero. If selection in utero is significant, surviving male children would be stronger since in utero shocks have more detrimental effects on boys than girls. As a result, we may find small, or no, effects on boys. Selection effects are likely to be particularly severe for large-scale shocks such as famines and civil wars [Gørgens et al., 2012, Neelsen & Stratmann, 2011]. In the case of relatively mild shocks in food insecurity, we expect the culling effect (selection in utero) to be less of a concern. Results presented in Section 4.1 confirm that prior. Finally, interpreting the impact of shocks in utero on later outcomes requires to consider possible compensating (or exacerbating) investments made by parents in children in response to health endowments after birth [Adhvaryu & Nyshadham, 2016, Almond & Mazumder, 2013]. For instance, Ayalew [2005] finds evidence of compensating health investment in Ethiopia. However, the same author shows evidence of reinforcing investment in terms of education. In our study, we confirm Miller [2017] in finding little evidence of subsequent investment responses by parents. Therefore, our results tend to support the existence of scarring effects that accumulate overtime and dominate possible selection effects or compensating mechanisms. 2 Data and Identification Strategy We exploit data from the YLE surveys. YLE is part of the Young Lives Project, an international study of childhood poverty tracking 12,000 children in four countries (Ethiopia, Peru, Vietnam and India) over a 15-year period. The Ethiopian data originate from 20 sites located in four regions of the country and Addis Ababa, in which more than 96 percent of the Ethiopian children live. These regions include Amhara, Oromia, Tigray and the Southern Nations, Nationalities and Peoples’ Region (SNNPR) (see Figure A.1 in online Appendix A). To choose the 20 sites of the study in each country, a sentinel site sampling approach was applied [Barnett et al., 2013]. In Ethiopia, the purposive sampling process follows the following three principles: (1) oversampling of food deficit districts; (2) the profile of the selected districts/sites should reflect the diversity of the country; and (3) the possibility of tracking children in the future at a reasonable cost. The sites in Ethiopia are selected in such a way that first, four regional states (Amhara, Oromia, SNNPR, Tigray) and one city administration (Addis Ababa) were chosen; second, up to five woredas (districts) were selected from each region (this accounts for 20 districts in total); third, from each woreda at least one kebele (local administrative area) was selected. The selected community may be a sentinel site itself or could be combined with neighbouring communities to create a site. Finally, 100 households with children born in 2001–2002 that constitute the younger cohort and 50 households with a child born in 1994–1995 that make up the older cohort were randomly chosen from each site.5 The YLE survey contains information on children’s health, education, schooling, time-use, feelings and attitudes and cognitive tests. Household information includes family background, education, consumption, social networks, livelihoods and wealth indicators. In this study, we exploit information about the so-called young cohort. The young cohort for Ethiopia comprises 1,999 children born between 2001 and 2002 in the 20 sites across the country. In the baseline survey of 2002, these children were aged between 6 and 18 months old.6 These children were then surveyed again in 2006, 2009 and 2013 (see Figure A.2 in online Appendix A).7 We focus on 24 of 26 communities, since two communities lack the food security information needed for our analysis. We seek to identify the causal impact of in utero exposure to food insecurity on cognitive development and educational progression using the following ordinary least-square specification.8 To shed light on the gender imbalances in the effect of the food insecurity shock, we estimate equation (1) separately for boys and girls. $$\begin{equation} Y_{idc} = \alpha_c + \theta_m + \beta \textrm{Exposure}_{dc} + X_{idc} + \varepsilon_{idc}, \end{equation}$$(1) where |$Y_{idc}$| is the outcome variable designated by various cognitive development measures for individual |$i$|⁠, born on date |$d$|⁠, in community |$c$|⁠. |$\textrm{Exposure}_{dc}$| is the number of days of exposure to seasonal food insecurity in utero, based on each child’s date of birth.9 In the analysis, similar to Miller [2017], we standardise the treatment variable to have a mean of zero and a standard deviation of one within each community to reduce the influence of communities with more severe periods of food insecurity.10|$X_{idc}$| denote the household and child characteristics. We also introduce community and month of birth fixed effects, |$\alpha _c$| and |$\theta _m$|⁠, to deal with omitted factors at the community level that would threaten the causal interpretation of our results.11 Our coefficient of interest, |$\beta $| captures the average effect of a standard deviation change (within a community) in exposure to in utero seasonal food insecurity on maths score and on grade-for-age outcomes. Standard errors are clustered at the community level to deal with correlation within location of residence. Given the low number of communities (24) that might underestimate intra-group correlation, we also show the robustness of our results to the use of wild bootstrapping method [Cameron et al., 2008, Cameron & Miller, 2015]. We report both the robust standard errors clustered at the community level and the wild bootstrap p-values for our main results.12 Our specification deals with several identification concerns. Community fixed effects deal with the threat of systematic differences across communities. For instance, food security is known to vary significantly across communities, mainly due to diverse agro-ecological zones and differences in terms of access to infrastructure. Stifel & Minten [2017] indeed find that households in Ethiopia living in remote areas are systematically more likely to be food insecure. Cognitive developments are also likely to differ across communities. We therefore not only control for household and child characteristics, |$X_{idc}$|⁠, but also for community fixed effects, |$\alpha _c$|⁠. Another issue relates to the confounding role of seasonality. The season of birth has indeed been found to be a strong predictor of health during childhood and later life outcomes [Buckles & Hungerman, 2013, Lokshin & Radyakin, 2012, McEniry & Palloni, 2010]. To deal with national seasonality effects that are unrelated to food insecurity (e.g., national policies), we introduce month of birth fixed effects, denoted |$\theta _m$|⁠. In Section 4 and online Appendix D.4, we will discuss further threats to identification, namely those inherent to mortality selection, endogenous parental responses, fertility selection, reporting errors, the existence of other mechanisms, attrition and missing data issues and exposure to seasonal food insecurity after birth. We now discuss the variables in turn. The dependent variables, designated by |$Y_{idc}$|⁠, are maths achievement scores used to measure children’s quantitative skills, and a measure of grade-for-age.13 We define grade-for-age as a binary variable that takes 1 if a child is in the correct grade for his or her age. The YLE survey contains completed grade. We need current grade to indicate whether the child is on one’s educational expected track. We calculate the current grade level using the information on whether the child is currently enrolled and data on completed grade. Specifically, current grade is equal to completed grade plus 1 if the child is enrolled. Panel A in Table 1 shows the descriptive statistics of our outcome variables: maths score and grade-for-age. As indicated in column (10) the mean values for boys and girls are not statistically different from each other. These descriptive statistics only reveal general patterns in our outcomes and nothing about the role of food insecurity exposure in utero. In our regression analysis, we standardise the maths scores to have a mean of 0 and a standard deviation of 1. Maths achievement tests and grade-for-age have been widely used to measure cognitive development and educational progression [Almond et al., 2015, Shah & Steinberg, 2017]. Table 1 Descriptive Statistics . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . . Full sample . Boys . Girls . Mean diff (Boys–Girls) . . Mean . Std. Dev. . N . Mean . Std. Dev. . N . Mean . Std. Dev. . N . P-values . Panel A: Outcome variables Maths score, restricted sample Maths Age 8 7.153 5.421 1461 7.253 5.493 768 7.043 5.342 693 0.461 Maths Age 12 10.615 6.032 1461 10.551 6.002 768 10.685 6.069 693 0.670 Maths score, unrestricted sample Maths Age 8 6.525 5.368 1695 6.670 5.448 878 6.370 5.280 817 0.250 Maths Age 12 10.503 6.053 1508 10.428 6.030 796 10.587 6.080 712 0.611 Grade-for-age Grade-for-age Age 8 0.606 0.489 1768 0.601 0.490 920 0.612 0.488 848 0.638 Grade-for-age Age 12 0.410 0.492 1757 0.393 0.489 916 0.428 0.495 841 0.135 Panel B: Exposure variable Exposure, 9 months 111.050 49.696 1875 111.385 48.895 970 110.691 50.564 905 0.762 . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . . Full sample . Boys . Girls . Mean diff (Boys–Girls) . . Mean . Std. Dev. . N . Mean . Std. Dev. . N . Mean . Std. Dev. . N . P-values . Panel A: Outcome variables Maths score, restricted sample Maths Age 8 7.153 5.421 1461 7.253 5.493 768 7.043 5.342 693 0.461 Maths Age 12 10.615 6.032 1461 10.551 6.002 768 10.685 6.069 693 0.670 Maths score, unrestricted sample Maths Age 8 6.525 5.368 1695 6.670 5.448 878 6.370 5.280 817 0.250 Maths Age 12 10.503 6.053 1508 10.428 6.030 796 10.587 6.080 712 0.611 Grade-for-age Grade-for-age Age 8 0.606 0.489 1768 0.601 0.490 920 0.612 0.488 848 0.638 Grade-for-age Age 12 0.410 0.492 1757 0.393 0.489 916 0.428 0.495 841 0.135 Panel B: Exposure variable Exposure, 9 months 111.050 49.696 1875 111.385 48.895 970 110.691 50.564 905 0.762 Source: Authors’ computation from Young Lives Data. For maths outcome, in the restricted sample, we restrict the sample to children for whom the outcomes of interest are observed all rounds (ages). Open in new tab Table 1 Descriptive Statistics . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . . Full sample . Boys . Girls . Mean diff (Boys–Girls) . . Mean . Std. Dev. . N . Mean . Std. Dev. . N . Mean . Std. Dev. . N . P-values . Panel A: Outcome variables Maths score, restricted sample Maths Age 8 7.153 5.421 1461 7.253 5.493 768 7.043 5.342 693 0.461 Maths Age 12 10.615 6.032 1461 10.551 6.002 768 10.685 6.069 693 0.670 Maths score, unrestricted sample Maths Age 8 6.525 5.368 1695 6.670 5.448 878 6.370 5.280 817 0.250 Maths Age 12 10.503 6.053 1508 10.428 6.030 796 10.587 6.080 712 0.611 Grade-for-age Grade-for-age Age 8 0.606 0.489 1768 0.601 0.490 920 0.612 0.488 848 0.638 Grade-for-age Age 12 0.410 0.492 1757 0.393 0.489 916 0.428 0.495 841 0.135 Panel B: Exposure variable Exposure, 9 months 111.050 49.696 1875 111.385 48.895 970 110.691 50.564 905 0.762 . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . . Full sample . Boys . Girls . Mean diff (Boys–Girls) . . Mean . Std. Dev. . N . Mean . Std. Dev. . N . Mean . Std. Dev. . N . P-values . Panel A: Outcome variables Maths score, restricted sample Maths Age 8 7.153 5.421 1461 7.253 5.493 768 7.043 5.342 693 0.461 Maths Age 12 10.615 6.032 1461 10.551 6.002 768 10.685 6.069 693 0.670 Maths score, unrestricted sample Maths Age 8 6.525 5.368 1695 6.670 5.448 878 6.370 5.280 817 0.250 Maths Age 12 10.503 6.053 1508 10.428 6.030 796 10.587 6.080 712 0.611 Grade-for-age Grade-for-age Age 8 0.606 0.489 1768 0.601 0.490 920 0.612 0.488 848 0.638 Grade-for-age Age 12 0.410 0.492 1757 0.393 0.489 916 0.428 0.495 841 0.135 Panel B: Exposure variable Exposure, 9 months 111.050 49.696 1875 111.385 48.895 970 110.691 50.564 905 0.762 Source: Authors’ computation from Young Lives Data. For maths outcome, in the restricted sample, we restrict the sample to children for whom the outcomes of interest are observed all rounds (ages). Open in new tab Fig. 1 Open in new tabDownload slide Reported Seasonal Food Insecurity by Calendar MonthSource: Authors’ calculations using data from Young Lives Study, Ethiopia Fig. 1 Open in new tabDownload slide Reported Seasonal Food Insecurity by Calendar MonthSource: Authors’ calculations using data from Young Lives Study, Ethiopia Our main variable of interest, |$\textrm{Exposure}_{dc}$| seeks to capture seasonal food insecurity in utero, by exploiting both food security information at the community level and variations at the individual level based on the date of birth. At the community level, food insecure months are identified in the YLE community surveys, where the community leaders are asked in which months of the year food becomes harder or more expensive to obtain. The alternative is to use weather shocks as a proxy for food shortage. However, the motivation of this study is to estimate the effect of seasonal food insecurity exposure on child cognitive development. Food insecurity in Ethiopia exhibit seasonal patterns, even in absence of drought. Food shortage is likely to exacerbate during drought years, but reported data collected at the community level are more likely to capture seasonal food insecurity. It has also the advantage of explicitly defining hunger and therefore is more likely to capture the direct impact on cognitive development. One disadvantage of reported food insecurity data may be the risk of systematic reporting bias. However, the fact that we are using data collected at community level from community representatives (not at household or individual level) makes the reporting bias minimal. The food insecurity information requested from community leaders was not a measure of food insecurity experienced at personal level that can be subject to erroneous and biased reporting. In addition, community leaders were asked about months that food becomes expensive and scarce in their respective community. Since this is a recurrent occurrence, we believe community representatives would be accurate in their reporting. We use information on community-level food insecurity from the community survey that was conducted in the second round (2006). The same information was also collected in the first round (2002). However, the pattern does not correspond to the conventionally observed seasonality in Ethiopia.14 In particular, the 2002 survey on food insecurity reports higher average relative food insecurity from October to January. But, this period coincides with post-harvest in Ethiopia and is thus characterised by relatively higher availability of food and lower prices. Thus, the information must have been reported and documented with errors. On the contrary, the food insecurity information reported in 2006 corresponds with the reality in Ethiopia. This is further corroborated by monthly food price data. Figure 1 depicts that relative food insecurity is reported from May to September. Figures C.1 and C.2 in online Appendix C also show that food prices both in rural and urban parts of the country are higher from May to September. This is also further confirmed by specific grain prices during 2001–2002 (see Figures C.3; C.4; C.5, C.6 in online Appendix C). As indicated in Figure 1, food insecurity is more likely to be reported during the rainy and planting periods of the main harvesting season. The harvesting season varies across agro-ecological zones, but the main harvesting season would usually fall from October to December. In each month from June to August, more than 20 of the surveyed communities report relative food insecurity. More than 15 of them also report relative food insecurity in May or September. The rest of the year is largely food secure. The seasonal pattern of food insecurity should not come as a surprise. In rural Ethiopia where subsistence agriculture is the prominent form of livelihood, households experience severe food shortages during the rainy/ planting season. Post-harvest, farmers have usually enough food with a high level of supply associated with relatively low prices (Figures C.1 and C.2 in online Appendix C). That is why we observe less food insecurity following harvests (from November to April). But when the rainy and planting seasons come, food availability decreases and pushes market prices upward, threatening food security. More than 60 percent communities report food insecurity for 4 to 5 months in a similar range to Hoddinott et al. [2011] (see Figure B.1 in online Appendix B). Table 2 Calculating the Number of Days a Child Exposed to Prenatal Seasonal Food Shortage Panel A, Community X . Date Conceived on 26 May 2001 Born on 16 Feb 2002 Month May Jun Jul Aug Sep Oct Nov Dec Jan Feb Food insecurity Yes Yes Yes Yes Yes No No No No No Panel B, Community Y Date Conceived on 10 Apr 2001 Born on 11 Jan 2002 Month Apr May Jun Jul Aug Sep Oct Nov Dec Jan Food insecurity No No Yes Yes Yes No No No No No Panel A, Community X . Date Conceived on 26 May 2001 Born on 16 Feb 2002 Month May Jun Jul Aug Sep Oct Nov Dec Jan Feb Food insecurity Yes Yes Yes Yes Yes No No No No No Panel B, Community Y Date Conceived on 10 Apr 2001 Born on 11 Jan 2002 Month Apr May Jun Jul Aug Sep Oct Nov Dec Jan Food insecurity No No Yes Yes Yes No No No No No Open in new tab Table 2 Calculating the Number of Days a Child Exposed to Prenatal Seasonal Food Shortage Panel A, Community X . Date Conceived on 26 May 2001 Born on 16 Feb 2002 Month May Jun Jul Aug Sep Oct Nov Dec Jan Feb Food insecurity Yes Yes Yes Yes Yes No No No No No Panel B, Community Y Date Conceived on 10 Apr 2001 Born on 11 Jan 2002 Month Apr May Jun Jul Aug Sep Oct Nov Dec Jan Food insecurity No No Yes Yes Yes No No No No No Panel A, Community X . Date Conceived on 26 May 2001 Born on 16 Feb 2002 Month May Jun Jul Aug Sep Oct Nov Dec Jan Feb Food insecurity Yes Yes Yes Yes Yes No No No No No Panel B, Community Y Date Conceived on 10 Apr 2001 Born on 11 Jan 2002 Month Apr May Jun Jul Aug Sep Oct Nov Dec Jan Food insecurity No No Yes Yes Yes No No No No No Open in new tab The community-level measurement of food insecurity is then used to determine how much a child is exposed to food insecurity in utero.15 Similar to Miller [2017], we compute the number of days a child has faced a food insecure environment while he/she was in utero. One lives in utero for approximately 38 weeks or 266 days starting from conception. Premature births may be an issue here. A total of 8.7 percent of the children in our sample are indeed born before the end of the term. We have data on the number of weeks of prematurity for only 73 percent of pre-term babies. For the remaining 27 percent, we substitute the missing observations by the median weeks of prematurity, 2 weeks. Thus, for premature babies, the number of days of exposure is calculated after adjustment is made for the reported number of weeks of prematurity. Miller [2017] adopts the same correction. As a result, our measure of food insecurity exposure in the full 9 months is calculated as the number of days a child is facing food insecurity in utero from conception to birth in those 266 days of prenatal experience. The calculation of our prenatal food insecurity exposure is described in Table 2. Assume for example, a child is conceived in a particular community on 26 May 2001. In theory the child will be born on 16 February 2002. In this community, food is relatively unavailable in May, June, July, August and September. The child born in that community will be exposed to prenatal food insecurity for 4 months (June, July, August and September) and 6 days (from May), resulting in 126 days of prenatal food insecurity exposure. Panel B shows a child born in another community on 11 January 2002. This child will be exposed to 3 months (June, July, August) of prenatal food insecurity, resulting in 91 days of exposure. Panel B in Table 1 reports the means, standard deviations, minimums and maximums of exposure in full 9 months. On average, a child has experienced 111 days (3.70 months) of food insecurity out of 266 days. Figure B.2 in online Appendix B also provides the histogram of the exposure measure in full 9 months of gestation. Panel B of Table 1 also shows that both boys and girls are equally affected by food insecurity in utero. 3 Results Table 3 presents the estimated effects of in utero exposure to food insecurity on maths score and the probability of being on the correct educational track at ages 8 and 12. For each outcome, the first panel presents the results without household controls, while the second panel introduces such control variables. Columns (1) and (2) provide estimates from regressions pooling boys and girls together, while the following columns contrast the results between boys (columns 3–4) and girls (columns 5–6). Column (1) in Panels A and B indicates a non-significant effect of exposure on maths score at age 8. However, columns (3) and (5) show there is a significant difference between boys and girls. At age 8, while the coefficients remain non-significant for girls, maths scores for boys are between 0.09 and 0.12 standard deviation lower as a result of one standard deviation change in the exposure to food insecurity (column 3). The detrimental effects of in utero exposure seem to accumulate with age to the point where in utero exposure to food insecurity has a significant and detrimental impact on cognitive development at age 12 for both sexes.16 This is consistent with the idea highlighted by Heckman & Masterov [2007]: disadvantages just like advantages accumulate overtime. Gender imbalances are further confirmed. At age 12, the decrease in maths score for boys by almost one third of a standard deviation (0.27–0.29, in column (4)) is significantly different from the decrease in girl’s score (about 0.1 standard deviation, in column (6)). Gender imbalances are also apparent with the other outcome. At age 12, a standardised deviation increase in food insecurity in utero also decreases the odds of being on the correct grade for one’s age, but only for boys. The gender imbalances in the effects of in utero exposure echo recent findings by Nilsson [2017] of higher vulnerability of male foetuses to alcohol consumption in utero.17 Table 3 Estimated Effect of In Utero Food Insecurity Exposure (Full Pregnancy) . (1) . (2) . (3) . (4) . (5) . (6) . . Full sample . Boys . Girls . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Maths without HH controls Exposure-Std -0.016 -0.169*** -0.120** -0.290*** 0.071 -0.090* (0.027) (0.035) (0.056) (0.067) (0.047) (0.052) [0.606] [0.002] [0.082] [0.002] [0.100] [0.082 ] P-value Boys=Girls (Age 8) 0.034 P-value Boys=Girls (Age 12) 0.038 Observations 1,461 1,461 768 768 693 693 Panel B: Maths with HH controls Exposure-Std -0.017 -0.175*** -0.093* -0.268*** 0.055 -0.111* (0.023) (0.039) (0.053) (0.066) (0.045) (0.059) [0.504] [0.002] [0.092] [0.004] [0.18] [0.05] P-value Boys=Girls (Age 8) 0.086 P-value Boys=Girls (Age 12) 0.089 Observations 1,441 1,441 755 755 686 686 Panel C: Grade-for-age(odds ratio) without HH controls Exposure-Std 0.977 0.804** 0.934 0.713** 1.029 0.860 (0.113) (0.086) (0.103) (0.119) (0.152) (0.166) Observations 1,768 1,757 909 916 844 841 Panel D: Grade-for-age(odds ratio) with HH controls Exposure-Std 0.945 0.781** 0.920 0.701* 0.982 0.817 (0.102) (0.086) (0.100) (0.128) (0.147) (0.170) Observations 1,745 1,734 895 901 836 833 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Ind. Controls Yes Yes Yes Yes Yes Yes . (1) . (2) . (3) . (4) . (5) . (6) . . Full sample . Boys . Girls . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Maths without HH controls Exposure-Std -0.016 -0.169*** -0.120** -0.290*** 0.071 -0.090* (0.027) (0.035) (0.056) (0.067) (0.047) (0.052) [0.606] [0.002] [0.082] [0.002] [0.100] [0.082 ] P-value Boys=Girls (Age 8) 0.034 P-value Boys=Girls (Age 12) 0.038 Observations 1,461 1,461 768 768 693 693 Panel B: Maths with HH controls Exposure-Std -0.017 -0.175*** -0.093* -0.268*** 0.055 -0.111* (0.023) (0.039) (0.053) (0.066) (0.045) (0.059) [0.504] [0.002] [0.092] [0.004] [0.18] [0.05] P-value Boys=Girls (Age 8) 0.086 P-value Boys=Girls (Age 12) 0.089 Observations 1,441 1,441 755 755 686 686 Panel C: Grade-for-age(odds ratio) without HH controls Exposure-Std 0.977 0.804** 0.934 0.713** 1.029 0.860 (0.113) (0.086) (0.103) (0.119) (0.152) (0.166) Observations 1,768 1,757 909 916 844 841 Panel D: Grade-for-age(odds ratio) with HH controls Exposure-Std 0.945 0.781** 0.920 0.701* 0.982 0.817 (0.102) (0.086) (0.100) (0.128) (0.147) (0.170) Observations 1,745 1,734 895 901 836 833 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Ind. Controls Yes Yes Yes Yes Yes Yes Robust standard errors (clustered at the community level) in parentheses. Wild bootstrap p-values in brackets. The asterisks next to the coefficients are for p-values associated with our main (non-wild bootstrap) regressions. *** p<0.01, ** p<0.05, * p<0.1. The dependent variables are standardised maths score and grade-for-age at age 8 and 12. The variable of interest captures prenatal exposure to seasonal food insecurity (full 9 months exposure) standardised to have mean 0 and standard deviation 1 within each community. Ind. controls include age of child in months, number of older siblings and dummies for gender, child ethnicity and prematurity. HH controls include household wealth index and dummies for gender of household head and mother’s education. For maths outcome, we restrict the sample to children for which we observe the outcomes of interest at all age (round) stages. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 3 Estimated Effect of In Utero Food Insecurity Exposure (Full Pregnancy) . (1) . (2) . (3) . (4) . (5) . (6) . . Full sample . Boys . Girls . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Maths without HH controls Exposure-Std -0.016 -0.169*** -0.120** -0.290*** 0.071 -0.090* (0.027) (0.035) (0.056) (0.067) (0.047) (0.052) [0.606] [0.002] [0.082] [0.002] [0.100] [0.082 ] P-value Boys=Girls (Age 8) 0.034 P-value Boys=Girls (Age 12) 0.038 Observations 1,461 1,461 768 768 693 693 Panel B: Maths with HH controls Exposure-Std -0.017 -0.175*** -0.093* -0.268*** 0.055 -0.111* (0.023) (0.039) (0.053) (0.066) (0.045) (0.059) [0.504] [0.002] [0.092] [0.004] [0.18] [0.05] P-value Boys=Girls (Age 8) 0.086 P-value Boys=Girls (Age 12) 0.089 Observations 1,441 1,441 755 755 686 686 Panel C: Grade-for-age(odds ratio) without HH controls Exposure-Std 0.977 0.804** 0.934 0.713** 1.029 0.860 (0.113) (0.086) (0.103) (0.119) (0.152) (0.166) Observations 1,768 1,757 909 916 844 841 Panel D: Grade-for-age(odds ratio) with HH controls Exposure-Std 0.945 0.781** 0.920 0.701* 0.982 0.817 (0.102) (0.086) (0.100) (0.128) (0.147) (0.170) Observations 1,745 1,734 895 901 836 833 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Ind. Controls Yes Yes Yes Yes Yes Yes . (1) . (2) . (3) . (4) . (5) . (6) . . Full sample . Boys . Girls . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Maths without HH controls Exposure-Std -0.016 -0.169*** -0.120** -0.290*** 0.071 -0.090* (0.027) (0.035) (0.056) (0.067) (0.047) (0.052) [0.606] [0.002] [0.082] [0.002] [0.100] [0.082 ] P-value Boys=Girls (Age 8) 0.034 P-value Boys=Girls (Age 12) 0.038 Observations 1,461 1,461 768 768 693 693 Panel B: Maths with HH controls Exposure-Std -0.017 -0.175*** -0.093* -0.268*** 0.055 -0.111* (0.023) (0.039) (0.053) (0.066) (0.045) (0.059) [0.504] [0.002] [0.092] [0.004] [0.18] [0.05] P-value Boys=Girls (Age 8) 0.086 P-value Boys=Girls (Age 12) 0.089 Observations 1,441 1,441 755 755 686 686 Panel C: Grade-for-age(odds ratio) without HH controls Exposure-Std 0.977 0.804** 0.934 0.713** 1.029 0.860 (0.113) (0.086) (0.103) (0.119) (0.152) (0.166) Observations 1,768 1,757 909 916 844 841 Panel D: Grade-for-age(odds ratio) with HH controls Exposure-Std 0.945 0.781** 0.920 0.701* 0.982 0.817 (0.102) (0.086) (0.100) (0.128) (0.147) (0.170) Observations 1,745 1,734 895 901 836 833 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Ind. Controls Yes Yes Yes Yes Yes Yes Robust standard errors (clustered at the community level) in parentheses. Wild bootstrap p-values in brackets. The asterisks next to the coefficients are for p-values associated with our main (non-wild bootstrap) regressions. *** p<0.01, ** p<0.05, * p<0.1. The dependent variables are standardised maths score and grade-for-age at age 8 and 12. The variable of interest captures prenatal exposure to seasonal food insecurity (full 9 months exposure) standardised to have mean 0 and standard deviation 1 within each community. Ind. controls include age of child in months, number of older siblings and dummies for gender, child ethnicity and prematurity. HH controls include household wealth index and dummies for gender of household head and mother’s education. For maths outcome, we restrict the sample to children for which we observe the outcomes of interest at all age (round) stages. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab 4 Discussion Three broad classes of factors may drive our results on the gender imbalances in seasonal food insecurity in utero. First, gender imbalances may be explained by the fact male foetuses are more vulnerable than girls in utero. Deficiencies in human capital development, the so-called scarring effects, may accumulate overtime. Second, the higher vulnerability of boys in utero may result in higher selective mortality in utero, the so-called culling effect and the survival of the stronger boys at and after births. Such alternative explanation would bias the coefficient of interest downward. Third, gender discrimination is usually expected against girls in such a context. Compensating mechanisms would therefore have mitigated the gender imbalances found in the previous section.18 4.1 Mortality selection Our sample only includes surviving children. Although our prenatal shock is of relatively mild (and frequent) nature, we cannot exclude that mortality in utero would drive our estimates towards zero. Surviving children may indeed appear to be the strongest, the healthiest and those with better genes. Similarly, the gender-based analysis could be biased due to differentiated mortality risk for boys and girls. The medical research indeed documents that male foetuses are more vulnerable to shocks and at greater mortality risk than female foetuses [Catalano et al., 2006, Eriksson et al., 2010, Kraemer, 2000, Mizuno, 2000, Shettles, 1961]. Empirical studies also document how negative prenatal exposure could alter sex composition at birth [Almond et al., 2010, Almond & Mazumder, 2011, Dagnelie et al., 2018, Nilsson, 2017, Valente, 2015, Van Ewijk, 2011]. We cannot directly test the effect of the shock on prenatal death differential between boys and girls. We do not have information about miscarriages and prenatal deaths. However, following Van Ewijk [2011], we test the role of selection by estimating the exposure effect on the probability of being a male at ages 1, 5, 8 and 12. We do not find strong evidence for mortality selection. Food insecurity shocks in utero do not seem to translate into changes in the sex ratio (Table 4). Only at age 5, we find a positive coefficient significantly different from zero at 90 percent level of confidence. Such coefficient cannot explain the stronger detrimental impact for boys compared to girls at ages 8 and 12. So, the causal interpretation of our main results is not threatened by mortality selection in utero or after birth. Gender imbalances in cognitive development cannot be explained by selective mortality. Table 4 Effect of Exposure on the Probability of the Child Surveyed Is Male . Logit odds ratio . . (1) . (2) . (3) . (4) . . Age 1 . Age 5 . Age 8 . Age 12 . Panel A: Child surveyed is male, without controls Exposure-Std 1.108 1.112 1.108 1.105 (0.074) (0.076) (0.079) (0.078) Observations 1,875 1,793 1,768 1,754 Panel B: Child surveyed is male, with controls Exposure-Std 1.094 1.109* 1.106 1.098 (0.065) (0.070) (0.071) (0.071) Observations 1,846 1,770 1,745 1,731 Community FE Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes . Logit odds ratio . . (1) . (2) . (3) . (4) . . Age 1 . Age 5 . Age 8 . Age 12 . Panel A: Child surveyed is male, without controls Exposure-Std 1.108 1.112 1.108 1.105 (0.074) (0.076) (0.079) (0.078) Observations 1,875 1,793 1,768 1,754 Panel B: Child surveyed is male, with controls Exposure-Std 1.094 1.109* 1.106 1.098 (0.065) (0.070) (0.071) (0.071) Observations 1,846 1,770 1,745 1,731 Community FE Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is dummy indicating child born is boy. The main independent variable is standardised prenatal exposure to seasonal food insecurity (exposure in whole nine months). Panel A reports results estimated without controls, while panel B shows results estimated with the following control variables: number of older siblings, household wealth index, dummies for child ethnicity, prematurity, gender of household head and mother’s education. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 4 Effect of Exposure on the Probability of the Child Surveyed Is Male . Logit odds ratio . . (1) . (2) . (3) . (4) . . Age 1 . Age 5 . Age 8 . Age 12 . Panel A: Child surveyed is male, without controls Exposure-Std 1.108 1.112 1.108 1.105 (0.074) (0.076) (0.079) (0.078) Observations 1,875 1,793 1,768 1,754 Panel B: Child surveyed is male, with controls Exposure-Std 1.094 1.109* 1.106 1.098 (0.065) (0.070) (0.071) (0.071) Observations 1,846 1,770 1,745 1,731 Community FE Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes . Logit odds ratio . . (1) . (2) . (3) . (4) . . Age 1 . Age 5 . Age 8 . Age 12 . Panel A: Child surveyed is male, without controls Exposure-Std 1.108 1.112 1.108 1.105 (0.074) (0.076) (0.079) (0.078) Observations 1,875 1,793 1,768 1,754 Panel B: Child surveyed is male, with controls Exposure-Std 1.094 1.109* 1.106 1.098 (0.065) (0.070) (0.071) (0.071) Observations 1,846 1,770 1,745 1,731 Community FE Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is dummy indicating child born is boy. The main independent variable is standardised prenatal exposure to seasonal food insecurity (exposure in whole nine months). Panel A reports results estimated without controls, while panel B shows results estimated with the following control variables: number of older siblings, household wealth index, dummies for child ethnicity, prematurity, gender of household head and mother’s education. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab 4.2 Parental responses versus biological effects Parents may respond to in utero shocks by adapting their investment towards children either to compensate or reinforce children’s endowments. If investment responses are compensatory, the effect of prenatal food insecurity shock will tend to understate biological effects. However, parents may also decide to reinforce children’s endowment. In that case our baseline results may overestimate the true biological effect. Recent empirical studies reviewed in Almond & Mazumder [2013] indeed find that parental investments reinforce initial endowment differences. In our case, that would mean that parents discriminate against boys more vulnerable in utero. That would be quite surprising given the abundant report on gender discrimination against girls in Ethiopia [Ayalew, 2005]. On the contrary, compensatory investments would attenuate the established gender imbalances in the previous section. Following Adhvaryu & Nyshadham [2016] and using the YLE survey, we assess whether the behavioural response from parents is driven by food insecurity shock in utero. Specifically, we test the effect of the shock on parental investments at ages 8 and 12 to investigate parental response once the cognitive endowment is realised.19 We employ several parental investment outcomes. These include an indicator to school enrolment, the number of study hours at home (including extra tuition) and an indicator to whether a child is enrolled into a private or a public school, an indicator if parents paid for school fees or tuition (last 12 months), an indicator if parents paid any medical expenditure (last 12 months), the number of meals a child had in the last 24 hours and the total number of food variety a child experienced in the last 24 hours.20 Panel E in Table D.1 in online Appendix D reports the descriptive statistics of these variables. In Table 5, we explore the role of parental investments that are directly related to education that happened at age 8 and 12.21 Overall, with other outcomes, we confirm the conclusions by Miller [2017] that there is limited role for parental investment.22 One exception is the fact the shock decreases the odds of being enrolled in school for girls at age 12. Under-investment in girls’ education at age 12 would tend to attenuate the gender imbalances against boys found earlier. Such under-investment is not confirmed using the time available for study at home or the probability to be sent to a private school or expenditures on school fees or tuition (educational expenditures). Table 5 Childhood Parental Educational Investments . Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Enrolled in to school Exposure-Std 0.892* 0.652 0.836 0.726 1.015 0.359** (0.061) (0.188) (0.150) (0.364) (0.093) (0.184) Observations 1,629 1,398 749 666 742 437 Panel B: Study hour at home(including extra tuition) Exposure-Std -0.005 0.043 -0.003 0.034 -0.005 0.052 (0.025) (0.031) (0.043) (0.043) (0.036) (0.052) Observations 1,744 1,732 904 900 840 832 Panel C: In private school Exposure-Std 1.192 0.940 1.144 0.996 1.277 0.945 (0.278) (0.166) (0.356) (0.264) (0.653) (0.260) Observations 757 851 269 353 323 308 Panel D: Education expenditures (school fees or tuition) Exposure-Std 0.868 1.035 1.072 0.836 0.787 1.247 (0.101) (0.126) (0.222) (0.117) (0.129) (0.225) Observations 1,555 1,631 782 825 724 723 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes . Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Enrolled in to school Exposure-Std 0.892* 0.652 0.836 0.726 1.015 0.359** (0.061) (0.188) (0.150) (0.364) (0.093) (0.184) Observations 1,629 1,398 749 666 742 437 Panel B: Study hour at home(including extra tuition) Exposure-Std -0.005 0.043 -0.003 0.034 -0.005 0.052 (0.025) (0.031) (0.043) (0.043) (0.036) (0.052) Observations 1,744 1,732 904 900 840 832 Panel C: In private school Exposure-Std 1.192 0.940 1.144 0.996 1.277 0.945 (0.278) (0.166) (0.356) (0.264) (0.653) (0.260) Observations 757 851 269 353 323 308 Panel D: Education expenditures (school fees or tuition) Exposure-Std 0.868 1.035 1.072 0.836 0.787 1.247 (0.101) (0.126) (0.222) (0.117) (0.129) (0.225) Observations 1,555 1,631 782 825 724 723 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes For binary outcomes (indicators of school enrolment, type of school enrolled in to and educational expenditures), Logit odds ratio are reported. Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is indicator to school enrolment (panel A), study hours at home (including extra tuition) (panel B), and indicator to whether a child is enrolled in to private or public school (panel C), indicator if parents paid for school fees or tuition for the child (panel D). Controls include (X) household wealth index, number of older siblings and dummies for gender, gender of household head, mother’s education, child ethnicity and prematurity. In the school enrolment regressions many observations are dropped because in several communities all children reported being in school. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 5 Childhood Parental Educational Investments . Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Enrolled in to school Exposure-Std 0.892* 0.652 0.836 0.726 1.015 0.359** (0.061) (0.188) (0.150) (0.364) (0.093) (0.184) Observations 1,629 1,398 749 666 742 437 Panel B: Study hour at home(including extra tuition) Exposure-Std -0.005 0.043 -0.003 0.034 -0.005 0.052 (0.025) (0.031) (0.043) (0.043) (0.036) (0.052) Observations 1,744 1,732 904 900 840 832 Panel C: In private school Exposure-Std 1.192 0.940 1.144 0.996 1.277 0.945 (0.278) (0.166) (0.356) (0.264) (0.653) (0.260) Observations 757 851 269 353 323 308 Panel D: Education expenditures (school fees or tuition) Exposure-Std 0.868 1.035 1.072 0.836 0.787 1.247 (0.101) (0.126) (0.222) (0.117) (0.129) (0.225) Observations 1,555 1,631 782 825 724 723 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes . Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Enrolled in to school Exposure-Std 0.892* 0.652 0.836 0.726 1.015 0.359** (0.061) (0.188) (0.150) (0.364) (0.093) (0.184) Observations 1,629 1,398 749 666 742 437 Panel B: Study hour at home(including extra tuition) Exposure-Std -0.005 0.043 -0.003 0.034 -0.005 0.052 (0.025) (0.031) (0.043) (0.043) (0.036) (0.052) Observations 1,744 1,732 904 900 840 832 Panel C: In private school Exposure-Std 1.192 0.940 1.144 0.996 1.277 0.945 (0.278) (0.166) (0.356) (0.264) (0.653) (0.260) Observations 757 851 269 353 323 308 Panel D: Education expenditures (school fees or tuition) Exposure-Std 0.868 1.035 1.072 0.836 0.787 1.247 (0.101) (0.126) (0.222) (0.117) (0.129) (0.225) Observations 1,555 1,631 782 825 724 723 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes For binary outcomes (indicators of school enrolment, type of school enrolled in to and educational expenditures), Logit odds ratio are reported. Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is indicator to school enrolment (panel A), study hours at home (including extra tuition) (panel B), and indicator to whether a child is enrolled in to private or public school (panel C), indicator if parents paid for school fees or tuition for the child (panel D). Controls include (X) household wealth index, number of older siblings and dummies for gender, gender of household head, mother’s education, child ethnicity and prematurity. In the school enrolment regressions many observations are dropped because in several communities all children reported being in school. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 6 reports results from parental health and nutritional investments such as medical expenditures; meal frequency or the food variety. In this case too, we find little evidence that parents respond to the shock through health and nutritional investments. At age 12, in utero exposure to food insecurity decreases the number of meal frequency, but with no apparent significant difference between boys and girls.23 Table 6 Childhood Parental Health and Nutritional Investments Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Medical expenditures Exposure-Std 1.145 1.140 1.097 1.214 1.203 1.094 (0.133) (0.163) (0.192) (0.328) (0.235) (0.283) Observations 1,746 1,734 905 901 841 829 Panel B: Meal frequency in the last 24 hours Exposure-Std 0.010 -0.052* -0.004 -0.055 0.019 -0.064 (0.021) (0.027) (0.040) (0.048) (0.035) (0.041) Observations 1,746 1,728 905 897 841 831 Panel C: Food variety in the last 24 hours Exposure-Std -0.054 -0.030 -0.092 0.041 -0.040 -0.077 (0.054) (0.073) (0.137) (0.124) (0.144) (0.078) Observations 1,745 1,727 905 897 840 830 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Medical expenditures Exposure-Std 1.145 1.140 1.097 1.214 1.203 1.094 (0.133) (0.163) (0.192) (0.328) (0.235) (0.283) Observations 1,746 1,734 905 901 841 829 Panel B: Meal frequency in the last 24 hours Exposure-Std 0.010 -0.052* -0.004 -0.055 0.019 -0.064 (0.021) (0.027) (0.040) (0.048) (0.035) (0.041) Observations 1,746 1,728 905 897 841 831 Panel C: Food variety in the last 24 hours Exposure-Std -0.054 -0.030 -0.092 0.041 -0.040 -0.077 (0.054) (0.073) (0.137) (0.124) (0.144) (0.078) Observations 1,745 1,727 905 897 840 830 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes For binary outcomes (indicator of if parents paid any medical expenditure to the child), Logit odds ratio are reported. Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is indicator if parents paid any medical expenditure to the child (panel A), the number of meals a child ate in the last 24 hours (panel B) and total number of food variety a child ate in the last 24 hours (panel C). Controls include (X) household wealth index, number of older siblings and dummies for gender, gender of household head, mother’s education, child ethnicity and prematurity. In the school enrolment regressions many observations are dropped because in several communities all children reported being in school. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 6 Childhood Parental Health and Nutritional Investments Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Medical expenditures Exposure-Std 1.145 1.140 1.097 1.214 1.203 1.094 (0.133) (0.163) (0.192) (0.328) (0.235) (0.283) Observations 1,746 1,734 905 901 841 829 Panel B: Meal frequency in the last 24 hours Exposure-Std 0.010 -0.052* -0.004 -0.055 0.019 -0.064 (0.021) (0.027) (0.040) (0.048) (0.035) (0.041) Observations 1,746 1,728 905 897 841 831 Panel C: Food variety in the last 24 hours Exposure-Std -0.054 -0.030 -0.092 0.041 -0.040 -0.077 (0.054) (0.073) (0.137) (0.124) (0.144) (0.078) Observations 1,745 1,727 905 897 840 830 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Medical expenditures Exposure-Std 1.145 1.140 1.097 1.214 1.203 1.094 (0.133) (0.163) (0.192) (0.328) (0.235) (0.283) Observations 1,746 1,734 905 901 841 829 Panel B: Meal frequency in the last 24 hours Exposure-Std 0.010 -0.052* -0.004 -0.055 0.019 -0.064 (0.021) (0.027) (0.040) (0.048) (0.035) (0.041) Observations 1,746 1,728 905 897 841 831 Panel C: Food variety in the last 24 hours Exposure-Std -0.054 -0.030 -0.092 0.041 -0.040 -0.077 (0.054) (0.073) (0.137) (0.124) (0.144) (0.078) Observations 1,745 1,727 905 897 840 830 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes For binary outcomes (indicator of if parents paid any medical expenditure to the child), Logit odds ratio are reported. Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is indicator if parents paid any medical expenditure to the child (panel A), the number of meals a child ate in the last 24 hours (panel B) and total number of food variety a child ate in the last 24 hours (panel C). Controls include (X) household wealth index, number of older siblings and dummies for gender, gender of household head, mother’s education, child ethnicity and prematurity. In the school enrolment regressions many observations are dropped because in several communities all children reported being in school. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab 5 Conclusions We examine the effect of in utero seasonal food insecurity on childhood cognitive development and grade-for-age. We exploit a unique dataset from the YLE. We estimate the effect of variation in the number of days of exposure to prenatal food insecurity on these outcomes, controlling for community and birth month fixed effects together with child and household characteristics. The inclusion of community and month of birth fixed effects means our estimations are unlikely to be affected by seasonality effects, or unobserved heterogeneity at the community level. We find that at age 8, maths are adversely affected by in utero exposure to seasonal food insecurity, but only for boys. At age 12, gender imbalances exacerbate. At age 12, a standard deviation increase in food insecurity in utero decreases maths scores by about one third of a standard deviation for boys, almost three times the decrease observed for girls. Moreover, at age 12, we find that food insecurity in utero decreases the odds of being on the correct grade, but only for boys. Based on the lack of selective mortality in utero and after birth and weak evidence for differentiated parental investment, we conjecture that scarring effects, particularly fierce for male foetuses, accumulate overtime. Such detrimental impacts are likely to have long-term consequences on socio-economic outcomes. Policy interventions that address seasonal food insecurity and programs that target pregnant women to enhance their resilience to seasonal food shortages could protect the development of children and minimise the long-term economic cost. Social safety net or cash transfer programs together with nutrition and micro-nutrient supplementation programs are obvious policy options. In Ethiopia, starting from 2005, the Productive Safety Net Programme (PSNP) aims at addressing seasonal food insecurity. Unfortunately, our data do not allow us to investigate the mitigating effect of the PSNP since the sampled children were in utero between 2000 and 2002, before the implementation of the PSNP. Understanding how specific programs build resilience to seasonal food insecurity is a path for future research. Finally, our paper faces other data limitations. While the focus on seasonal food insecurity seems is arguably policy relevant, the lack of weather data does not allow us to benchmark our results against large-scale weather shocks. Similarly, another limitation is the indirect nature of our evidence behind the gender imbalances. Data on birth weights, miscarriages, stillbirth and neonatal mortality would help us better understand the mechanisms potentially explaining gender imbalances to seasonal food insecurity. Supplementary material Supplementary material is available at Journal of African Economies online. Acknowledgments We thank Ian Walker, Kalle Hirvonen, Maria Navarro Paniagua and Colin Green for their comments. All errors and opinions expressed remain our own. The data used in this publication come from Young Lives, a 15-year study of the changing nature of childhood poverty in Ethiopia, India, Peru and Vietnam (www.younglives.org.uk). Young Lives is funded by UK aid from the Department for International Development (DFID). The views expressed here are those of the authors. They are not necessarily those of Young Lives, the University of Oxford, DFID or other funders. Footnotes 1 " The data may not be representative of the country since the information is obtained from chronically food-insecure woredas (districts). 2 " The literature on the long-term effect of in utero shocks has relied on rare and extreme events such as famine, war and terrorist attacks. In addition to the likely fiercer selection in utero, it has been difficult to distinguish the nutritional impact of shocks from the psychological stress associated with the shock [Currie & Vogl, 2013]. We are not able to distinguish between these insults but in our case, similar to Miller [2017] and Nilsson [2017], we are more likely to directly capture the nutritional impact of shocks. 3 " Since childhood height is correlated with later cognitive development outcomes [Case & Paxson, 2008], our choice of outcomes (maths score and grade-for-age) also complements the analysis in Miller [2017]. 4 " The vulnerability of boys in utero is consistent with the medical literature [Catalano et al., 2006, Eriksson et al., 2010, Kraemer, 2000, Mizuno, 2000, Shettles, 1961]. 5 " See http://www.younglives.org.uk/content/sampling-and-attrition for details. 6 " The survey also collects similar information for the older cohort, born around 1994–1995. These children were 7–8 years old during the first round survey in 2002. We do not have birth information such as prematurity for this cohort that are essential for computing our exposure variables. Thus, this cohort cannot be exploited for our main analysis. We will nonetheless use the information about this cohort to assess the relationship between cognitive development and long-term education outcomes to shed light on the long-run significance of our results. 7 " Our main outcome variables are measured in round 3 and round 4 of the YLE surveys. It is important to note that round 3 surveys were implemented between November and February of 2009/10 and round 4 surveys were implemented between November and February of 2013/14. As discussed below, these months are relatively food secure months as they correspond to the post-harvest period. 8 " For binary outcomes, logistic regressions are used instead. 9 " Similar to Miller [2017], date of birth for each child is calculated using age of child in days and the date of interview from the first survey round. 10 " In online Appendix D.2, we discuss the importance of standardisation within communities, together with other functional assumptions (e.g., linearity). Moreover, we show the robustness of our results to using non-standardised treatment variable. 11 " Sibling fixed effects cannot be used in our paper since the YLE surveys did not collect sibling information for our outcomes of interest. 12 " We do not report wild bootstrap p-values for the grade-for-age outcome as the results on this outcome are estimated using logistic regression. 13 " We also report results from other cognitive development measures collected by the YLE study: the Early Grade Reading Assessment (EGRA) and the Peabody Picture Vocabulary Test (PPVT). The EGRA is orally assessed only at age 8. It is implemented to measure the most basic skills for literacy acquisition in the early grades. It involves recognising letters of the alphabet, reading simple words, understanding sentences and paragraphs and listening with comprehension. The PPVT is a widely used test of receptive vocabulary. The descriptive statistics of these variables are presented in Panel A of Table D.1 in online Appendix D.5. These tests are adapted to different languages spoken in the country. Difficulty levels may have changed during translation, and as a result, it is recommended that comparison must be within languages [Cueto & Leon, 2012, Singh, 2015]. We cannot limit our data to a certain language in the country since the geographical concentration of languages in Ethiopia would cancel out the variation in the exposure variable. As a result, we are cautious about interpreting results from these two tests. 14 " Using food insecurity information from 2002 community survey confirms our results but with much lower magnitude. Results are discussed in online Appendix D.2. 15 " We describe the reliability of the community-level food insecurity information in the construction of our in utero exposure in online Appendix C. 16 " We also investigate the impact by trimester, finding stronger effects for boys in the first and second trimesters. For the sake of presentation, we only discuss these results in online Appendix D based on Table D.10. 17 " Detailed results of Table 3 including control variables are provided in Tables D.2 and D.3 of online Appendix D. Online Appendix D provides and discusses additional robustness checks. Gender imbalances in the effect of exposure on other tests are also apparent in Table D.4 of online Appendix D. Exposure decreases reading at age 8, significantly more for boys. Though we find no significant effect on PPVT, exposure has unexpected and positive effect on girl’s PPVT score at age 12. We do not, however, interpret further results from these two outcomes given the lack of accuracy of the cognitive tests in Ethiopia (see footnote 11). Although efficiency might be affected in some cases, our results are largely robust to not standardising the measure of exposure to food insecurity in utero (more subject to high-leverage communities, see Table D.11), to using round 1 food insecurity information (similar to Miller [2017]) to capture seasonal food insecurity at the community level (Table D.13), to non-linear effects in exposure (Table D.14), to relaxing the restriction of including only children for which we observe the outcomes of interest at all age (round) stages in the samples of regressions using the maths score as a dependent variable (Table D.15). These results are further discussed in online Appendix D. 18 " Those mechanisms can equally be seen as threats to the general identification but help us to understand the gender imbalances. Other identification threats, with no obvious gender bias, may affect the magnitude of our coefficients. In online Appendix D, we therefore also examine how our results may be threatened by (1) fertility selection; (2) reporting errors; (3) the existence of other mechanisms; (4) attrition and missing data; and (5) exposure to seasonal food insecurity after birth. Some of these identification threats are also tested on gender-stratified samples to assess their possible consequences on the consistency of our results on the gender imbalances of seasonal food security in utero. 19 " We focus on investment carried out at ages 8 and 12. On the one hand, parents at this stage can observe the realised cognition of their children to decide to reinforce or compensate it. On the other hand, it helps us understand whether differential investment at ages 8 and 12 could explain the difference in the observed effect of the shock on cognition between ages 8 and 12. 20 " Parents/children were asked 7 yes/no questions related to meal frequency for a child. Specifically, they were asked if the child ate any food before breakfast, breakfast, food between breakfast and midday meal, midday meal, food between midday meal and evening meal, evening meal and food after the evening meal. We computed meal frequency for each round (age) as the sum of these frequencies. We top coded at six meals. Moreover, parents/children were asked whether the child ate different types of foods in the last 24 hours. They were asked 17 (at age 8) and 15 (at age 12) yes/no questions. We computed food variety variables for each round (age) as the sum of these frequencies. We top coded at 10 food types. 21 " Nonetheless, in Table D.5 of online Appendix D, we also report results on investment on preschool, an educational investment that happened on or before age 5. We find no significant effect of exposure on preschool investment. 22 " In an unpublished paper, Fan & Porter [2018] investigate the relationship between cognitive ability and educational fees in Ethiopia. They find evidence of compensating effects when cognitive ability is instrumented by weather shocks at early age. However, comparability is limited given the differences in shocks, the timing of the shocks, their IV approach and the lack of heterogeneous analysis across gender. 23 " Gender-specific pre-natal investment is not expected since sex detection before birth is very uncommon in Ethiopia. We nonetheless test the impact of in utero exposure on pre-natal and neo-natal (BCG) investments. We do not find any significant impact of in utero exposure to food insecurity (see Table D.6 in online Appendix D). Moreover, similar to Yi et al. [2015], but at the cost of introducing endogeneity, we have also assessed how much our coefficients of interest may be altered by directly including proxies for possible parental responses into the main regressions. The results in Table D.7, Table D.8 and Table D.9 from online Appendix D confirm that our main results remain unchanged. Parental responses cannot explain the gender imbalances found in our main results. Furthermore, other sources of heterogeneity may explain why the effects accumulate overtime, indirectly shedding light on differentiated ability of households to deal with food insecurity in utero. Results from heterogeneity analysis based on wealth (Table D.16); access to market (Table D.17) and access to road (Table D.18) are commented and discussed in online Appendix D. References Abay , K. and Hirvonen , K. ( 2017 ). ‘ Does market access mitigate the impact of seasonality on child growth? Panel data evidence from Northern Ethiopia ’, The Journal of Development Studies , 53 ( 9 ): 1414 – 29 . Google Scholar Crossref Search ADS WorldCat Adhvaryu , A. and Nyshadham , A. ( 2016 ). ‘ Endowments at birth and parents’ investments in children ’, The Economic Journal , 126 ( 593 ): 781 – 820 . Google Scholar Crossref Search ADS PubMed WorldCat Alderman , H. , Hoddinott , J., and Kinsey , B. ( 2006 ). ‘ Long term consequences of early childhood malnutrition ’, Oxford Economic Papers , 58 ( 3 ): 450 – 74 . Google Scholar Crossref Search ADS WorldCat Alem , Y. and Söderbom , M. ( 2012 ). ‘ Household-level consumption in urban Ethiopia: the effects of a large food price shock ’, World Development , 40 ( 1 ): 146 – 62 . Google Scholar Crossref Search ADS WorldCat Almond , D. and Currie , J. ( 2011 ). ‘ Killing me softly: the fetal origins hypothesis ’, The Journal of Economic Perspectives , 25 ( 3 ): 153 – 72 . Google Scholar Crossref Search ADS PubMed WorldCat Almond D. , Edlund L., Li H., Zhang J. ( 2010 ) ‘ Long-term effects of early-life development: evidence from the 1959 to 1961 China famine ’, in The Economic Consequences of Demographic Change in East Asia, NBER-EASE Volume 19 . Chicago : University of Chicago Press , pp. 321 – 45 . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Almond , D. and Mazumder , B. ( 2013 ). ‘ Fetal origins and parental responses ’, Annual Review of Economics , 5 ( 1 ): 37 – 56 . Google Scholar Crossref Search ADS WorldCat Almond , D. , Mazumder , B., and Ewijk , R. ( 2015 ). ‘ In utero Ramadan exposure and children’s academic performance ’, The Economic Journal , 125 ( 589 ): 1501 – 33 . Google Scholar Crossref Search ADS WorldCat Almond , D. and Mazumder , B. A. ( 2011 ). ‘ Health capital and the prenatal environment: the effect of Ramadan observance during pregnancy ’, American Economic Journal: Applied Economics , 3 ( 4 ): 56 – 85 . Google Scholar Crossref Search ADS WorldCat Ayalew , T. ( 2005 ). ‘ Parental preference, heterogeneity, and human capital inequality ’, Economic Development and Cultural Change , 53 ( 2 ): 381 – 407 . Google Scholar Crossref Search ADS WorldCat Barker , D. J. ( 1990 ). ‘ The fetal and infant origins of adult disease ’, BMJ: British Medical Journal , 301 ( 6761 ): 1111 . Google Scholar Crossref Search ADS WorldCat Barnett , I. , Ariana , P., Petrou , S., Penny , M. E., Galab , S., Woldehanna , T., Escobal , J. A., Plugge , E., Boyden , J., et al. ( 2013 ). ‘ Cohort profile: the Young Lives study ’, International Journal of Epidemiology , 42 ( 3 ): 701 – 8 . Google Scholar Crossref Search ADS PubMed WorldCat Berg , G. J. , Pinger , P. R., and Schoch , J. ( 2016 ). ‘ Instrumental variable estimation of the causal effect of hunger early in life on health later in life ’, The Economic Journal , 126 ( 591 ): 465 – 506 . Google Scholar Crossref Search ADS WorldCat Black , S. E. , Devereux , P. J., and Salvanes , K. G. ( 2016 ). ‘ Does grief transfer across generations? Bereavements during pregnancy and child outcomes ’, American Economic Journal: Applied Economics , 8 ( 1 ): 193 – 223 . Google Scholar Crossref Search ADS WorldCat Buckles , K. S. and Hungerman , D. M. ( 2013 ). ‘ Season of birth and later outcomes: old questions, new answers ’, Review of Economics and Statistics , 95 ( 3 ): 711 – 24 . Google Scholar Crossref Search ADS PubMed WorldCat Cameron , A. C. , Gelbach , J. B., and Miller , D. L. ( 2008 ). ‘ Bootstrap-based improvements for inference with clustered errors ’, The Review of Economics and Statistics , 90 ( 3 ): 414 – 27 . Google Scholar Crossref Search ADS WorldCat Cameron , A. C. and Miller , D. L. ( 2015 ). ‘ A practitioner’s guide to cluster-robust inference ’, Journal of Human Resources , 50 ( 2 ): 317 – 72 . Google Scholar Crossref Search ADS WorldCat Case , A. and Paxson , C. ( 2008 ). Stature and status: height, ability, and labor market outcomes . Journal of Political Economy , 116 ( 3 ): 499 – 532 . Google Scholar Crossref Search ADS PubMed WorldCat Catalano , R. , Bruckner , T., Marks , A. R., and Eskenazi , B. ( 2006 ). ‘ Exogenous shocks to the human sex ratio: the case of September 11, 2001 in New York city ’, Human Reproduction , 21 ( 12 ): 3127 – 31 . Google Scholar Crossref Search ADS PubMed WorldCat Cueto , S. and Leon , J. ( 2012 ). Psychometric characteristics of cognitive development and achievement instruments in round 3 of young lives . Young Lives Technical Note 25, Young Lives, University of Oxford . Currie J. , Thomas D. ( 2001 ) ‘ Early test scores, school quality and SES: long-run effects on wage and employment outcomes ’, in Research in Labor Economics: Worker Wellbeing in a Changing Labor Market , vol. 20 . Bingley, UK : Emerald Group Publishing Limited , pp. 103 – 32 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Currie , J. and Vogl , T. ( 2013 ). ‘ Early-life health and adult circumstance in developing countries ’, Annual Review of Economics , 5 ( 1 ): 1 – 36 . Google Scholar Crossref Search ADS WorldCat Dagnelie O. , De Luca G. D., Maystadt J.-F. ( 2018 ) ‘ Violence, selection and infant mortality in Congo ’, Journal of Health Economics , 59 : 153 – 77 . Google Scholar Crossref Search ADS PubMed WorldCat Dercon , S. ( 2004 ). ‘ Growth and shocks: evidence from rural Ethiopia ’, Journal of Development Economics , 74 ( 2 ): 309 – 29 . Google Scholar Crossref Search ADS WorldCat Dercon , S. and Krishnan , P. ( 2000 ). ‘ Vulnerability, seasonality and poverty in Ethiopia ’, The Journal of Development Studies , 36 ( 6 ): 25 – 53 . Google Scholar Crossref Search ADS WorldCat Dercon , S. and Porter , C. ( 2014 ). ‘ Live Aid revisited: long-term impacts of the 1984 Ethiopian famine on children ’, Journal of the European Economic Association , 12 ( 4 ): 927 – 48 . Google Scholar Crossref Search ADS WorldCat Eriksson , J. G. , Kajantie , E., Osmond , C., Thornburg , K., and Barker , D. J. ( 2010 ). ‘ Boys live dangerously in the womb ’, American Journal of Human Biology , 22 ( 3 ): 330 – 5 . Google Scholar Crossref Search ADS PubMed WorldCat Fan , W. and Porter , C. ( 2018 ). ‘ Reinforcement or Compensation? Parental Responses to Children’s Revealed Human Capital Levels ’, Technical report, Young Lives . FAO ( 2009 ) ‘ The State of Food Insecurity in the World: Economic Crises—Impacts and Lessons Learned ’, Technical report . Food and Agriculture Organization of the: United Nations (FAO) . Glewwe , P. , Jacoby , H. G., and King , E. M. ( 2001 ). ‘ Early childhood nutrition and academic achievement: a longitudinal analysis ’, Journal of Public Economics , 81 ( 3 ): 345 – 68 . Google Scholar Crossref Search ADS WorldCat Gørgens , T. , Meng , X., and Vaithianathan , R. ( 2012 ). ‘ Stunting and selection effects of famine: a case study of the Great Chinese Famine ’, Journal of Development Economics , 97 ( 1 ): 99 – 111 . Google Scholar Crossref Search ADS WorldCat Heckman , J. J. ( 2007 ). ‘ The economics, technology, and neuroscience of human capability formation ’, Proceedings of the National Academy of Sciences , 104 ( 33 ): 13250 – 5 . Google Scholar Crossref Search ADS WorldCat Heckman , J. J. and Masterov , D. V. ( 2007 ). ‘ The productivity argument for investing in young children ’. Applied Economic Perspectives and Policy , 29 ( 3 ): 446 – 93 . OpenURL Placeholder Text WorldCat Hernández-Julián , R. , Mansour , H., and Peters , C. ( 2014 ). ‘ The effects of intrauterine malnutrition on birth and fertility outcomes: evidence from the 1974 Bangladesh famine ’, Demography , 51 ( 5 ): 1775 – 96 . Google Scholar Crossref Search ADS PubMed WorldCat Hill R. V. , Porter C. ( 2017 ) ‘ Vulnerability to drought and food price shocks: evidence from Ethiopia ’, World Development , 96 : 65 – 77 . Google Scholar Crossref Search ADS WorldCat Hoddinott J. , Kumar N., Taffesse A. S., Tefera M. ( 2011 ) ‘ Evaluation of Ethiopia’s Food Security Program: Documenting progress in the implementation of the Productive Safety Nets Programme and the Household Asset Building Programme ’, in Unpublished report, International Food Policy Research Institute and Ethiopian Development Research Institute . Kraemer , S. ( 2000 ). ‘ The fragile male ’, BMJ: British Medical Journal , 321 ( 7276 ): 1609 . Google Scholar Crossref Search ADS WorldCat Lindeboom , M. , Portrait , F., and Van den Berg , G. J. ( 2010 ). ‘ Long-run effects on longevity of a nutritional shock early in life: the Dutch Potato famine of 1846–1847 ’, Journal of Health Economics , 29 ( 5 ): 617 – 29 . Google Scholar Crossref Search ADS PubMed WorldCat Lokshin , M. and Radyakin , S. ( 2012 ). ‘ Month of birth and children’s health in India ’, Journal of Human Resources , 47 ( 1 ): 174 – 203 . Google Scholar Crossref Search ADS WorldCat Luo , Z. , Mu , R., and Zhang , X. ( 2006 ). ‘ Famine and overweight in China ’, Applied Economic Perspectives and Policy , 28 ( 3 ): 296 – 304 . OpenURL Placeholder Text WorldCat Maluccio , J. A. , Hoddinott , J., Behrman , J. R., Martorell , R., Quisumbing , A. R., and Stein , A. D. ( 2009 ). ‘ The impact of improving nutrition during early childhood on education among Guatemalan adults ’, The Economic Journal , 119 ( 537 ): 734 – 63 . Google Scholar Crossref Search ADS WorldCat McEniry , M. and Palloni , A. ( 2010 ). ‘ Early life exposures and the occurrence and timing of heart disease among the older adult Puerto Rican population ’, Demography , 47 ( 1 ): 23 – 43 . Google Scholar Crossref Search ADS PubMed WorldCat Miller R. ( 2017 ) ‘ Childhood health and prenatal exposure to seasonal food scarcity in Ethiopia ’, World Development , 99 : 350 – 76 . Google Scholar Crossref Search ADS WorldCat Mizuno , R. ( 2000 ). ‘ The male/female ratio of fetal deaths and births in Japan ’, The Lancet , 356 ( 9231 ): 738 – 9 . Google Scholar Crossref Search ADS WorldCat Mu , R. and Zhang , X. ( 2011 ). ‘ Why does the Great Chinese Famine affect the male and female survivors differently? Mortality selection versus son preference ’, Economics & Human Biology , 9 ( 1 ): 92 – 105 . Google Scholar Crossref Search ADS WorldCat Neelsen , S. and Stratmann , T. ( 2011 ). ‘ Effects of prenatal and early life malnutrition: evidence from the Greek famine ’, Journal of Health Economics , 30 ( 3 ): 479 – 88 . Google Scholar Crossref Search ADS PubMed WorldCat Nilsson , J. P. ( 2017 ). ‘ Alcohol availability, prenatal conditions, and long-term economic outcomes ’, Journal of Political Economy , 125 ( 4 ): 1149 – 207 . Google Scholar Crossref Search ADS WorldCat Porter , C. ( 2012 ). ‘ Shocks, consumption and income diversification in rural Ethiopia ’, Journal of Development Studies , 48 ( 9 ): 1209 – 22 . Google Scholar Crossref Search ADS WorldCat Roseboom , T. J. , Painter , R. C., van Abeelen , A. F., Veenendaal , M. V., and de Rooij , S. R. ( 2011 ). ‘ Hungry in the womb: what are the consequences? Lessons from the Dutch famine ’, Maturitas , 70 ( 2 ): 141 – 5 . Google Scholar Crossref Search ADS PubMed WorldCat Shah , M. and Steinberg , B. M. ( 2017 ). ‘ Drought of opportunities: contemporaneous and long-term impacts of rainfall shocks on human capital ’, Journal of Political Economy , 125 ( 2 ): 527 – 61 . Google Scholar Crossref Search ADS WorldCat Shettles , L. B. ( 1961 ). ‘ Conception and birth sex ratios ’, Obstetrics & Gynecology , 18 ( 1 ): 122 – 30 . OpenURL Placeholder Text WorldCat Singh A. ( 2015 ) ‘ Private school effects in urban and rural India: panel estimates at primary and secondary school ages ’, Journal of Development Economics , 113 : 16 – 32 . Google Scholar Crossref Search ADS WorldCat Stifel D. , Minten B. ( 2017 ) ‘ Market access, welfare, and nutrition: evidence from Ethiopia ’, World Development , 90 : 229 – 41 . Google Scholar Crossref Search ADS WorldCat Umana-Aponte , M. ( 2011 ). Long-term effects of a nutritional shock: The 1980 famine of Karamoja, Uganda. Centre for Market and Public Organisation, University of Bristol, Working Paper, 11/258 . Valente C. ( 2015 ) ‘ Civil conflict, gender-specific fetal loss, and selection: a new test of the Trivers–Willard hypothesis ’, Journal of Health Economics , 39 : 31 – 50 . Google Scholar Crossref Search ADS PubMed WorldCat Van Ewijk , R. ( 2011 ). ‘ Long-term health effects on the next generation of Ramadan fasting during pregnancy ’, Journal of Health Economics , 30 ( 6 ): 1246 – 60 . Google Scholar Crossref Search ADS PubMed WorldCat WFP ( 2013 ). Your Tax Dollars Help Starve Children . https://www.wfp.org/stories/10-things-you-need-know-about-hunger-2013. [Online; accessed 10-December-2016]. Yi , J. , Heckman , J. J., Zhang , J., and Conti , G. ( 2015 ). ‘ Early health shocks, intra-household resource allocation and child outcomes ’, The Economic Journal , 125 ( 588 ): F347 – 71 . Google Scholar Crossref Search ADS PubMed WorldCat © The authors 2020. Published by Oxford University Press on behalf of the Institute of Mathematics and its Applications. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Journal of African Economies Oxford University Press

In Utero Seasonal Food Insecurity and Cognitive Development: Evidence on Gender Imbalances From Ethiopia

Loading next page...
 
/lp/oxford-university-press/in-utero-seasonal-food-insecurity-and-cognitive-development-evidence-F0vi0RH5hn

References (81)

Publisher
Oxford University Press
Copyright
© The authors 2020. Published by Oxford University Press on behalf of the Institute of Mathematics and its Applications. All rights reserved.
ISSN
0963-8024
eISSN
1464-3723
DOI
10.1093/jafeco/ejz028
Publisher site
See Article on Publisher Site

Abstract

Abstract Food insecurity is pervasive and highly seasonal in Ethiopia. In this study, we investigate the effect of seasonal food insecurity on child development. Exploiting the Young Lives Ethiopia dataset, we study the gender-specific impact of in utero exposure to seasonal food insecurity on cognitive development and the probability of being on the expected grade for children of age 8 up to 12. We find that at age 8, in utero exposure to food insecurity negatively affects cognitive development, only for boys. At age 12, such exposure significantly reduces cognitive development for all children, but with a significantly higher magnitude for boys. The impact is almost three times bigger compared to the one estimated for girls. Corroborated with other outcomes, we explain such gender imbalances by the accumulative nature of the scarring effect rather than the culling effect or gender differences in parental investment. 1 Introduction Early cognitive abilities play an important role in determining long-term schooling and wages [Currie & Thomas, 2001]. The development of these skills begins in utero and continues to evolve over the life cycle through a dynamic process of skill formation [Heckman, 2007]. Large-scale shocks such as famine, natural disasters and civil wars experienced during prenatal and early life environment have been found to be strong predictors of future outcomes [Almond & Currie, 2011, Currie & Vogl, 2013]. Nonetheless, food shortages are much more frequent and potentially more detrimental on most children’s life cycle. Each year, more people die from hunger than AIDS, malaria and tuberculosis combined [WFP 2013]. Ethiopia is a case in point. According to FAO [2009], about 44 percent of the total population in Ethiopia were undernourished between 2004 and 2006. This could be attributed to chronic food insecurity, a pervasive phenomenon in the country. A substantial number of people in Ethiopia are facing difficulties in feeding themselves on a regular basis around the rainy and planting seasons. According to the International Food Policy Research Institute and the Ethiopian Development Research Institute, more than 25 percent of households in Tigray region, close to 30 percent of households in Oromia (the most populous region) and 25 percent of households in Southern Nations, Nationalities and Peoples’ (SNNP) region reported food gaps during the rainy season in 2006 [Hoddinott et al., 2011]. For Amhara (the second most populous region) the food gap stands at less than 20 percent.1 In the same year, close to 20 percent and 15 percent of households reported food gaps for 3 months and 4 months, respectively. Such chronic under-nutrition, in particular at an early age, is likely to have long-term consequences in terms of health, schooling and socio-economic outcomes [Alderman et al., 2006, Miller, 2017]. The positive impact of early childhood nutrition on education has also been established [Glewwe et al., 2001, Maluccio et al., 2009]. The impact of prenatal exposure to seasonal food insecurity is largely unknown. In this study, we examine the impact of in utero exposure to seasonal food insecurity on cognitive development and grade-for-age. We exploit a unique dataset from the Young Lives Ethiopia (YLE) study and apply a novel identification strategy. We estimate the effect of variation in the number of days of exposure to prenatal food insecurity on cognitive development outcomes, controlling for community and birth month fixed effects together with child and household characteristics. We find that a standard deviation increase in relative food insecurity exposure in utero results in lower maths achievements score at age 12 by about 0.175 standard deviations. Exposure also decreases the odds of being on the correct educational track. More importantly, we shed light on the gender-specific impact of seasonal food insecurity in utero. We find that there are significant gender imbalances. Both at ages 8 and 12, an in utero shock decreases boys’ maths score more severely than girls’. At age 12, we find that boys are less likely to be on the right grade for their age. Our paper directly relates to the emerging literature exploring the effect of prenatal shock on human capital development of children [Almond et al., 2015, Neelsen & Stratmann, 2011]. The so called ‘foetal origins’ hypothesis advocated by Barker describes that conditions in utero (for instance, nutritional deficiencies) have long lasting health effects [Almond & Currie, 2011, Barker, 1990]. Prenatal nutrition shocks should also have significant detrimental effects on brain development [Almond et al., 2015, Almond & Mazumder, 2011, Umana-Aponte, 2011]. To establish causal effects, studies exploit famines and other shocks like natural disasters, wars and disease epidemics as exogenous natural experiments. Almond & Currie [2011] and Currie & Vogl [2013] provide extensive review of the literature.2 More directly related to the context of our study, there is a large number of studies investigating the impact of seasonality, price shocks and weather shocks on households’ vulnerability and child development in Ethiopia [Abay & Hirvonen, 2017, Alem & Söderbom, 2012, Dercon, 2004, Dercon & Krishnan, 2000, Dercon & Porter, 2014, Hill & Porter, 2017, Miller, 2017, Porter, 2012]. However, this literature has not considered the individual exposure to shock in utero, except for Dercon & Porter [2014] and Miller [2017]. Dercon & Porter [2014] find detrimental impact of the 1984/85 Ethiopian famine on height of young adults. However, no effect is found from exposure in utero. On the contrary, Miller [2017] finds significant effects of seasonal food scarcity in utero on height at ages 8 and 12, but no significant difference between boys and girls. Our paper extends Miller [2017]’s work by exploring the impact of seasonal food insecurity on cognitive development and by investigating possible gender imbalances in such an impact.3 Boys have been found to be more vulnerable to shocks in utero such as famine [Almond et al., 2010, Hernández-Julián et al., 2014, Roseboom et al., 2011], conflict [Dagnelie et al., 2018, Valente, 2015], alcohol consumption [Nilsson, 2017] or stress from mother’s grief [Black et al., 2016].4 However, the nature of gender imbalances in the effect of in utero and early life shocks on different health and socio-economic outcomes differs across existing studies. While the Great Chinese Famine has been found to be more detrimental for girls in terms of health and education [Luo et al., 2006, Mu & Zhang, 2011], stronger effects on boys have been found from famines during World War II in Greece, Germany and the Netherlands [Berg et al., 2016] and during the Dutch Potato Famine in the mid-19th century [Lindeboom et al., 2010]. Nilsson [2017] also finds stronger effects of in utero exposure to increased alcohol availability on long-term labour market and educational outcomes and cognitive and non-cognitive ability of boys. The differences in the results are puzzling. The use of different outcome variables and contextual differences may be behind the mixed nature of the evidence, but the impact of in utero shocks on outcomes later in life may result from different mechanisms [Dagnelie et al., 2018, Nilsson, 2017, Valente, 2015]. The scarring effects result from a downward shift of the entire foetal health distribution. Since male foetuses disproportionally stand at the low end of that distribution, deficiencies due to the scarring effects may accumulate overtime and explain more detrimental effects for boys later in life. On the contrary, the culling effect directly relates to selective mortality in utero. If selection in utero is significant, surviving male children would be stronger since in utero shocks have more detrimental effects on boys than girls. As a result, we may find small, or no, effects on boys. Selection effects are likely to be particularly severe for large-scale shocks such as famines and civil wars [Gørgens et al., 2012, Neelsen & Stratmann, 2011]. In the case of relatively mild shocks in food insecurity, we expect the culling effect (selection in utero) to be less of a concern. Results presented in Section 4.1 confirm that prior. Finally, interpreting the impact of shocks in utero on later outcomes requires to consider possible compensating (or exacerbating) investments made by parents in children in response to health endowments after birth [Adhvaryu & Nyshadham, 2016, Almond & Mazumder, 2013]. For instance, Ayalew [2005] finds evidence of compensating health investment in Ethiopia. However, the same author shows evidence of reinforcing investment in terms of education. In our study, we confirm Miller [2017] in finding little evidence of subsequent investment responses by parents. Therefore, our results tend to support the existence of scarring effects that accumulate overtime and dominate possible selection effects or compensating mechanisms. 2 Data and Identification Strategy We exploit data from the YLE surveys. YLE is part of the Young Lives Project, an international study of childhood poverty tracking 12,000 children in four countries (Ethiopia, Peru, Vietnam and India) over a 15-year period. The Ethiopian data originate from 20 sites located in four regions of the country and Addis Ababa, in which more than 96 percent of the Ethiopian children live. These regions include Amhara, Oromia, Tigray and the Southern Nations, Nationalities and Peoples’ Region (SNNPR) (see Figure A.1 in online Appendix A). To choose the 20 sites of the study in each country, a sentinel site sampling approach was applied [Barnett et al., 2013]. In Ethiopia, the purposive sampling process follows the following three principles: (1) oversampling of food deficit districts; (2) the profile of the selected districts/sites should reflect the diversity of the country; and (3) the possibility of tracking children in the future at a reasonable cost. The sites in Ethiopia are selected in such a way that first, four regional states (Amhara, Oromia, SNNPR, Tigray) and one city administration (Addis Ababa) were chosen; second, up to five woredas (districts) were selected from each region (this accounts for 20 districts in total); third, from each woreda at least one kebele (local administrative area) was selected. The selected community may be a sentinel site itself or could be combined with neighbouring communities to create a site. Finally, 100 households with children born in 2001–2002 that constitute the younger cohort and 50 households with a child born in 1994–1995 that make up the older cohort were randomly chosen from each site.5 The YLE survey contains information on children’s health, education, schooling, time-use, feelings and attitudes and cognitive tests. Household information includes family background, education, consumption, social networks, livelihoods and wealth indicators. In this study, we exploit information about the so-called young cohort. The young cohort for Ethiopia comprises 1,999 children born between 2001 and 2002 in the 20 sites across the country. In the baseline survey of 2002, these children were aged between 6 and 18 months old.6 These children were then surveyed again in 2006, 2009 and 2013 (see Figure A.2 in online Appendix A).7 We focus on 24 of 26 communities, since two communities lack the food security information needed for our analysis. We seek to identify the causal impact of in utero exposure to food insecurity on cognitive development and educational progression using the following ordinary least-square specification.8 To shed light on the gender imbalances in the effect of the food insecurity shock, we estimate equation (1) separately for boys and girls. $$\begin{equation} Y_{idc} = \alpha_c + \theta_m + \beta \textrm{Exposure}_{dc} + X_{idc} + \varepsilon_{idc}, \end{equation}$$(1) where |$Y_{idc}$| is the outcome variable designated by various cognitive development measures for individual |$i$|⁠, born on date |$d$|⁠, in community |$c$|⁠. |$\textrm{Exposure}_{dc}$| is the number of days of exposure to seasonal food insecurity in utero, based on each child’s date of birth.9 In the analysis, similar to Miller [2017], we standardise the treatment variable to have a mean of zero and a standard deviation of one within each community to reduce the influence of communities with more severe periods of food insecurity.10|$X_{idc}$| denote the household and child characteristics. We also introduce community and month of birth fixed effects, |$\alpha _c$| and |$\theta _m$|⁠, to deal with omitted factors at the community level that would threaten the causal interpretation of our results.11 Our coefficient of interest, |$\beta $| captures the average effect of a standard deviation change (within a community) in exposure to in utero seasonal food insecurity on maths score and on grade-for-age outcomes. Standard errors are clustered at the community level to deal with correlation within location of residence. Given the low number of communities (24) that might underestimate intra-group correlation, we also show the robustness of our results to the use of wild bootstrapping method [Cameron et al., 2008, Cameron & Miller, 2015]. We report both the robust standard errors clustered at the community level and the wild bootstrap p-values for our main results.12 Our specification deals with several identification concerns. Community fixed effects deal with the threat of systematic differences across communities. For instance, food security is known to vary significantly across communities, mainly due to diverse agro-ecological zones and differences in terms of access to infrastructure. Stifel & Minten [2017] indeed find that households in Ethiopia living in remote areas are systematically more likely to be food insecure. Cognitive developments are also likely to differ across communities. We therefore not only control for household and child characteristics, |$X_{idc}$|⁠, but also for community fixed effects, |$\alpha _c$|⁠. Another issue relates to the confounding role of seasonality. The season of birth has indeed been found to be a strong predictor of health during childhood and later life outcomes [Buckles & Hungerman, 2013, Lokshin & Radyakin, 2012, McEniry & Palloni, 2010]. To deal with national seasonality effects that are unrelated to food insecurity (e.g., national policies), we introduce month of birth fixed effects, denoted |$\theta _m$|⁠. In Section 4 and online Appendix D.4, we will discuss further threats to identification, namely those inherent to mortality selection, endogenous parental responses, fertility selection, reporting errors, the existence of other mechanisms, attrition and missing data issues and exposure to seasonal food insecurity after birth. We now discuss the variables in turn. The dependent variables, designated by |$Y_{idc}$|⁠, are maths achievement scores used to measure children’s quantitative skills, and a measure of grade-for-age.13 We define grade-for-age as a binary variable that takes 1 if a child is in the correct grade for his or her age. The YLE survey contains completed grade. We need current grade to indicate whether the child is on one’s educational expected track. We calculate the current grade level using the information on whether the child is currently enrolled and data on completed grade. Specifically, current grade is equal to completed grade plus 1 if the child is enrolled. Panel A in Table 1 shows the descriptive statistics of our outcome variables: maths score and grade-for-age. As indicated in column (10) the mean values for boys and girls are not statistically different from each other. These descriptive statistics only reveal general patterns in our outcomes and nothing about the role of food insecurity exposure in utero. In our regression analysis, we standardise the maths scores to have a mean of 0 and a standard deviation of 1. Maths achievement tests and grade-for-age have been widely used to measure cognitive development and educational progression [Almond et al., 2015, Shah & Steinberg, 2017]. Table 1 Descriptive Statistics . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . . Full sample . Boys . Girls . Mean diff (Boys–Girls) . . Mean . Std. Dev. . N . Mean . Std. Dev. . N . Mean . Std. Dev. . N . P-values . Panel A: Outcome variables Maths score, restricted sample Maths Age 8 7.153 5.421 1461 7.253 5.493 768 7.043 5.342 693 0.461 Maths Age 12 10.615 6.032 1461 10.551 6.002 768 10.685 6.069 693 0.670 Maths score, unrestricted sample Maths Age 8 6.525 5.368 1695 6.670 5.448 878 6.370 5.280 817 0.250 Maths Age 12 10.503 6.053 1508 10.428 6.030 796 10.587 6.080 712 0.611 Grade-for-age Grade-for-age Age 8 0.606 0.489 1768 0.601 0.490 920 0.612 0.488 848 0.638 Grade-for-age Age 12 0.410 0.492 1757 0.393 0.489 916 0.428 0.495 841 0.135 Panel B: Exposure variable Exposure, 9 months 111.050 49.696 1875 111.385 48.895 970 110.691 50.564 905 0.762 . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . . Full sample . Boys . Girls . Mean diff (Boys–Girls) . . Mean . Std. Dev. . N . Mean . Std. Dev. . N . Mean . Std. Dev. . N . P-values . Panel A: Outcome variables Maths score, restricted sample Maths Age 8 7.153 5.421 1461 7.253 5.493 768 7.043 5.342 693 0.461 Maths Age 12 10.615 6.032 1461 10.551 6.002 768 10.685 6.069 693 0.670 Maths score, unrestricted sample Maths Age 8 6.525 5.368 1695 6.670 5.448 878 6.370 5.280 817 0.250 Maths Age 12 10.503 6.053 1508 10.428 6.030 796 10.587 6.080 712 0.611 Grade-for-age Grade-for-age Age 8 0.606 0.489 1768 0.601 0.490 920 0.612 0.488 848 0.638 Grade-for-age Age 12 0.410 0.492 1757 0.393 0.489 916 0.428 0.495 841 0.135 Panel B: Exposure variable Exposure, 9 months 111.050 49.696 1875 111.385 48.895 970 110.691 50.564 905 0.762 Source: Authors’ computation from Young Lives Data. For maths outcome, in the restricted sample, we restrict the sample to children for whom the outcomes of interest are observed all rounds (ages). Open in new tab Table 1 Descriptive Statistics . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . . Full sample . Boys . Girls . Mean diff (Boys–Girls) . . Mean . Std. Dev. . N . Mean . Std. Dev. . N . Mean . Std. Dev. . N . P-values . Panel A: Outcome variables Maths score, restricted sample Maths Age 8 7.153 5.421 1461 7.253 5.493 768 7.043 5.342 693 0.461 Maths Age 12 10.615 6.032 1461 10.551 6.002 768 10.685 6.069 693 0.670 Maths score, unrestricted sample Maths Age 8 6.525 5.368 1695 6.670 5.448 878 6.370 5.280 817 0.250 Maths Age 12 10.503 6.053 1508 10.428 6.030 796 10.587 6.080 712 0.611 Grade-for-age Grade-for-age Age 8 0.606 0.489 1768 0.601 0.490 920 0.612 0.488 848 0.638 Grade-for-age Age 12 0.410 0.492 1757 0.393 0.489 916 0.428 0.495 841 0.135 Panel B: Exposure variable Exposure, 9 months 111.050 49.696 1875 111.385 48.895 970 110.691 50.564 905 0.762 . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . . Full sample . Boys . Girls . Mean diff (Boys–Girls) . . Mean . Std. Dev. . N . Mean . Std. Dev. . N . Mean . Std. Dev. . N . P-values . Panel A: Outcome variables Maths score, restricted sample Maths Age 8 7.153 5.421 1461 7.253 5.493 768 7.043 5.342 693 0.461 Maths Age 12 10.615 6.032 1461 10.551 6.002 768 10.685 6.069 693 0.670 Maths score, unrestricted sample Maths Age 8 6.525 5.368 1695 6.670 5.448 878 6.370 5.280 817 0.250 Maths Age 12 10.503 6.053 1508 10.428 6.030 796 10.587 6.080 712 0.611 Grade-for-age Grade-for-age Age 8 0.606 0.489 1768 0.601 0.490 920 0.612 0.488 848 0.638 Grade-for-age Age 12 0.410 0.492 1757 0.393 0.489 916 0.428 0.495 841 0.135 Panel B: Exposure variable Exposure, 9 months 111.050 49.696 1875 111.385 48.895 970 110.691 50.564 905 0.762 Source: Authors’ computation from Young Lives Data. For maths outcome, in the restricted sample, we restrict the sample to children for whom the outcomes of interest are observed all rounds (ages). Open in new tab Fig. 1 Open in new tabDownload slide Reported Seasonal Food Insecurity by Calendar MonthSource: Authors’ calculations using data from Young Lives Study, Ethiopia Fig. 1 Open in new tabDownload slide Reported Seasonal Food Insecurity by Calendar MonthSource: Authors’ calculations using data from Young Lives Study, Ethiopia Our main variable of interest, |$\textrm{Exposure}_{dc}$| seeks to capture seasonal food insecurity in utero, by exploiting both food security information at the community level and variations at the individual level based on the date of birth. At the community level, food insecure months are identified in the YLE community surveys, where the community leaders are asked in which months of the year food becomes harder or more expensive to obtain. The alternative is to use weather shocks as a proxy for food shortage. However, the motivation of this study is to estimate the effect of seasonal food insecurity exposure on child cognitive development. Food insecurity in Ethiopia exhibit seasonal patterns, even in absence of drought. Food shortage is likely to exacerbate during drought years, but reported data collected at the community level are more likely to capture seasonal food insecurity. It has also the advantage of explicitly defining hunger and therefore is more likely to capture the direct impact on cognitive development. One disadvantage of reported food insecurity data may be the risk of systematic reporting bias. However, the fact that we are using data collected at community level from community representatives (not at household or individual level) makes the reporting bias minimal. The food insecurity information requested from community leaders was not a measure of food insecurity experienced at personal level that can be subject to erroneous and biased reporting. In addition, community leaders were asked about months that food becomes expensive and scarce in their respective community. Since this is a recurrent occurrence, we believe community representatives would be accurate in their reporting. We use information on community-level food insecurity from the community survey that was conducted in the second round (2006). The same information was also collected in the first round (2002). However, the pattern does not correspond to the conventionally observed seasonality in Ethiopia.14 In particular, the 2002 survey on food insecurity reports higher average relative food insecurity from October to January. But, this period coincides with post-harvest in Ethiopia and is thus characterised by relatively higher availability of food and lower prices. Thus, the information must have been reported and documented with errors. On the contrary, the food insecurity information reported in 2006 corresponds with the reality in Ethiopia. This is further corroborated by monthly food price data. Figure 1 depicts that relative food insecurity is reported from May to September. Figures C.1 and C.2 in online Appendix C also show that food prices both in rural and urban parts of the country are higher from May to September. This is also further confirmed by specific grain prices during 2001–2002 (see Figures C.3; C.4; C.5, C.6 in online Appendix C). As indicated in Figure 1, food insecurity is more likely to be reported during the rainy and planting periods of the main harvesting season. The harvesting season varies across agro-ecological zones, but the main harvesting season would usually fall from October to December. In each month from June to August, more than 20 of the surveyed communities report relative food insecurity. More than 15 of them also report relative food insecurity in May or September. The rest of the year is largely food secure. The seasonal pattern of food insecurity should not come as a surprise. In rural Ethiopia where subsistence agriculture is the prominent form of livelihood, households experience severe food shortages during the rainy/ planting season. Post-harvest, farmers have usually enough food with a high level of supply associated with relatively low prices (Figures C.1 and C.2 in online Appendix C). That is why we observe less food insecurity following harvests (from November to April). But when the rainy and planting seasons come, food availability decreases and pushes market prices upward, threatening food security. More than 60 percent communities report food insecurity for 4 to 5 months in a similar range to Hoddinott et al. [2011] (see Figure B.1 in online Appendix B). Table 2 Calculating the Number of Days a Child Exposed to Prenatal Seasonal Food Shortage Panel A, Community X . Date Conceived on 26 May 2001 Born on 16 Feb 2002 Month May Jun Jul Aug Sep Oct Nov Dec Jan Feb Food insecurity Yes Yes Yes Yes Yes No No No No No Panel B, Community Y Date Conceived on 10 Apr 2001 Born on 11 Jan 2002 Month Apr May Jun Jul Aug Sep Oct Nov Dec Jan Food insecurity No No Yes Yes Yes No No No No No Panel A, Community X . Date Conceived on 26 May 2001 Born on 16 Feb 2002 Month May Jun Jul Aug Sep Oct Nov Dec Jan Feb Food insecurity Yes Yes Yes Yes Yes No No No No No Panel B, Community Y Date Conceived on 10 Apr 2001 Born on 11 Jan 2002 Month Apr May Jun Jul Aug Sep Oct Nov Dec Jan Food insecurity No No Yes Yes Yes No No No No No Open in new tab Table 2 Calculating the Number of Days a Child Exposed to Prenatal Seasonal Food Shortage Panel A, Community X . Date Conceived on 26 May 2001 Born on 16 Feb 2002 Month May Jun Jul Aug Sep Oct Nov Dec Jan Feb Food insecurity Yes Yes Yes Yes Yes No No No No No Panel B, Community Y Date Conceived on 10 Apr 2001 Born on 11 Jan 2002 Month Apr May Jun Jul Aug Sep Oct Nov Dec Jan Food insecurity No No Yes Yes Yes No No No No No Panel A, Community X . Date Conceived on 26 May 2001 Born on 16 Feb 2002 Month May Jun Jul Aug Sep Oct Nov Dec Jan Feb Food insecurity Yes Yes Yes Yes Yes No No No No No Panel B, Community Y Date Conceived on 10 Apr 2001 Born on 11 Jan 2002 Month Apr May Jun Jul Aug Sep Oct Nov Dec Jan Food insecurity No No Yes Yes Yes No No No No No Open in new tab The community-level measurement of food insecurity is then used to determine how much a child is exposed to food insecurity in utero.15 Similar to Miller [2017], we compute the number of days a child has faced a food insecure environment while he/she was in utero. One lives in utero for approximately 38 weeks or 266 days starting from conception. Premature births may be an issue here. A total of 8.7 percent of the children in our sample are indeed born before the end of the term. We have data on the number of weeks of prematurity for only 73 percent of pre-term babies. For the remaining 27 percent, we substitute the missing observations by the median weeks of prematurity, 2 weeks. Thus, for premature babies, the number of days of exposure is calculated after adjustment is made for the reported number of weeks of prematurity. Miller [2017] adopts the same correction. As a result, our measure of food insecurity exposure in the full 9 months is calculated as the number of days a child is facing food insecurity in utero from conception to birth in those 266 days of prenatal experience. The calculation of our prenatal food insecurity exposure is described in Table 2. Assume for example, a child is conceived in a particular community on 26 May 2001. In theory the child will be born on 16 February 2002. In this community, food is relatively unavailable in May, June, July, August and September. The child born in that community will be exposed to prenatal food insecurity for 4 months (June, July, August and September) and 6 days (from May), resulting in 126 days of prenatal food insecurity exposure. Panel B shows a child born in another community on 11 January 2002. This child will be exposed to 3 months (June, July, August) of prenatal food insecurity, resulting in 91 days of exposure. Panel B in Table 1 reports the means, standard deviations, minimums and maximums of exposure in full 9 months. On average, a child has experienced 111 days (3.70 months) of food insecurity out of 266 days. Figure B.2 in online Appendix B also provides the histogram of the exposure measure in full 9 months of gestation. Panel B of Table 1 also shows that both boys and girls are equally affected by food insecurity in utero. 3 Results Table 3 presents the estimated effects of in utero exposure to food insecurity on maths score and the probability of being on the correct educational track at ages 8 and 12. For each outcome, the first panel presents the results without household controls, while the second panel introduces such control variables. Columns (1) and (2) provide estimates from regressions pooling boys and girls together, while the following columns contrast the results between boys (columns 3–4) and girls (columns 5–6). Column (1) in Panels A and B indicates a non-significant effect of exposure on maths score at age 8. However, columns (3) and (5) show there is a significant difference between boys and girls. At age 8, while the coefficients remain non-significant for girls, maths scores for boys are between 0.09 and 0.12 standard deviation lower as a result of one standard deviation change in the exposure to food insecurity (column 3). The detrimental effects of in utero exposure seem to accumulate with age to the point where in utero exposure to food insecurity has a significant and detrimental impact on cognitive development at age 12 for both sexes.16 This is consistent with the idea highlighted by Heckman & Masterov [2007]: disadvantages just like advantages accumulate overtime. Gender imbalances are further confirmed. At age 12, the decrease in maths score for boys by almost one third of a standard deviation (0.27–0.29, in column (4)) is significantly different from the decrease in girl’s score (about 0.1 standard deviation, in column (6)). Gender imbalances are also apparent with the other outcome. At age 12, a standardised deviation increase in food insecurity in utero also decreases the odds of being on the correct grade for one’s age, but only for boys. The gender imbalances in the effects of in utero exposure echo recent findings by Nilsson [2017] of higher vulnerability of male foetuses to alcohol consumption in utero.17 Table 3 Estimated Effect of In Utero Food Insecurity Exposure (Full Pregnancy) . (1) . (2) . (3) . (4) . (5) . (6) . . Full sample . Boys . Girls . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Maths without HH controls Exposure-Std -0.016 -0.169*** -0.120** -0.290*** 0.071 -0.090* (0.027) (0.035) (0.056) (0.067) (0.047) (0.052) [0.606] [0.002] [0.082] [0.002] [0.100] [0.082 ] P-value Boys=Girls (Age 8) 0.034 P-value Boys=Girls (Age 12) 0.038 Observations 1,461 1,461 768 768 693 693 Panel B: Maths with HH controls Exposure-Std -0.017 -0.175*** -0.093* -0.268*** 0.055 -0.111* (0.023) (0.039) (0.053) (0.066) (0.045) (0.059) [0.504] [0.002] [0.092] [0.004] [0.18] [0.05] P-value Boys=Girls (Age 8) 0.086 P-value Boys=Girls (Age 12) 0.089 Observations 1,441 1,441 755 755 686 686 Panel C: Grade-for-age(odds ratio) without HH controls Exposure-Std 0.977 0.804** 0.934 0.713** 1.029 0.860 (0.113) (0.086) (0.103) (0.119) (0.152) (0.166) Observations 1,768 1,757 909 916 844 841 Panel D: Grade-for-age(odds ratio) with HH controls Exposure-Std 0.945 0.781** 0.920 0.701* 0.982 0.817 (0.102) (0.086) (0.100) (0.128) (0.147) (0.170) Observations 1,745 1,734 895 901 836 833 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Ind. Controls Yes Yes Yes Yes Yes Yes . (1) . (2) . (3) . (4) . (5) . (6) . . Full sample . Boys . Girls . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Maths without HH controls Exposure-Std -0.016 -0.169*** -0.120** -0.290*** 0.071 -0.090* (0.027) (0.035) (0.056) (0.067) (0.047) (0.052) [0.606] [0.002] [0.082] [0.002] [0.100] [0.082 ] P-value Boys=Girls (Age 8) 0.034 P-value Boys=Girls (Age 12) 0.038 Observations 1,461 1,461 768 768 693 693 Panel B: Maths with HH controls Exposure-Std -0.017 -0.175*** -0.093* -0.268*** 0.055 -0.111* (0.023) (0.039) (0.053) (0.066) (0.045) (0.059) [0.504] [0.002] [0.092] [0.004] [0.18] [0.05] P-value Boys=Girls (Age 8) 0.086 P-value Boys=Girls (Age 12) 0.089 Observations 1,441 1,441 755 755 686 686 Panel C: Grade-for-age(odds ratio) without HH controls Exposure-Std 0.977 0.804** 0.934 0.713** 1.029 0.860 (0.113) (0.086) (0.103) (0.119) (0.152) (0.166) Observations 1,768 1,757 909 916 844 841 Panel D: Grade-for-age(odds ratio) with HH controls Exposure-Std 0.945 0.781** 0.920 0.701* 0.982 0.817 (0.102) (0.086) (0.100) (0.128) (0.147) (0.170) Observations 1,745 1,734 895 901 836 833 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Ind. Controls Yes Yes Yes Yes Yes Yes Robust standard errors (clustered at the community level) in parentheses. Wild bootstrap p-values in brackets. The asterisks next to the coefficients are for p-values associated with our main (non-wild bootstrap) regressions. *** p<0.01, ** p<0.05, * p<0.1. The dependent variables are standardised maths score and grade-for-age at age 8 and 12. The variable of interest captures prenatal exposure to seasonal food insecurity (full 9 months exposure) standardised to have mean 0 and standard deviation 1 within each community. Ind. controls include age of child in months, number of older siblings and dummies for gender, child ethnicity and prematurity. HH controls include household wealth index and dummies for gender of household head and mother’s education. For maths outcome, we restrict the sample to children for which we observe the outcomes of interest at all age (round) stages. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 3 Estimated Effect of In Utero Food Insecurity Exposure (Full Pregnancy) . (1) . (2) . (3) . (4) . (5) . (6) . . Full sample . Boys . Girls . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Maths without HH controls Exposure-Std -0.016 -0.169*** -0.120** -0.290*** 0.071 -0.090* (0.027) (0.035) (0.056) (0.067) (0.047) (0.052) [0.606] [0.002] [0.082] [0.002] [0.100] [0.082 ] P-value Boys=Girls (Age 8) 0.034 P-value Boys=Girls (Age 12) 0.038 Observations 1,461 1,461 768 768 693 693 Panel B: Maths with HH controls Exposure-Std -0.017 -0.175*** -0.093* -0.268*** 0.055 -0.111* (0.023) (0.039) (0.053) (0.066) (0.045) (0.059) [0.504] [0.002] [0.092] [0.004] [0.18] [0.05] P-value Boys=Girls (Age 8) 0.086 P-value Boys=Girls (Age 12) 0.089 Observations 1,441 1,441 755 755 686 686 Panel C: Grade-for-age(odds ratio) without HH controls Exposure-Std 0.977 0.804** 0.934 0.713** 1.029 0.860 (0.113) (0.086) (0.103) (0.119) (0.152) (0.166) Observations 1,768 1,757 909 916 844 841 Panel D: Grade-for-age(odds ratio) with HH controls Exposure-Std 0.945 0.781** 0.920 0.701* 0.982 0.817 (0.102) (0.086) (0.100) (0.128) (0.147) (0.170) Observations 1,745 1,734 895 901 836 833 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Ind. Controls Yes Yes Yes Yes Yes Yes . (1) . (2) . (3) . (4) . (5) . (6) . . Full sample . Boys . Girls . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Maths without HH controls Exposure-Std -0.016 -0.169*** -0.120** -0.290*** 0.071 -0.090* (0.027) (0.035) (0.056) (0.067) (0.047) (0.052) [0.606] [0.002] [0.082] [0.002] [0.100] [0.082 ] P-value Boys=Girls (Age 8) 0.034 P-value Boys=Girls (Age 12) 0.038 Observations 1,461 1,461 768 768 693 693 Panel B: Maths with HH controls Exposure-Std -0.017 -0.175*** -0.093* -0.268*** 0.055 -0.111* (0.023) (0.039) (0.053) (0.066) (0.045) (0.059) [0.504] [0.002] [0.092] [0.004] [0.18] [0.05] P-value Boys=Girls (Age 8) 0.086 P-value Boys=Girls (Age 12) 0.089 Observations 1,441 1,441 755 755 686 686 Panel C: Grade-for-age(odds ratio) without HH controls Exposure-Std 0.977 0.804** 0.934 0.713** 1.029 0.860 (0.113) (0.086) (0.103) (0.119) (0.152) (0.166) Observations 1,768 1,757 909 916 844 841 Panel D: Grade-for-age(odds ratio) with HH controls Exposure-Std 0.945 0.781** 0.920 0.701* 0.982 0.817 (0.102) (0.086) (0.100) (0.128) (0.147) (0.170) Observations 1,745 1,734 895 901 836 833 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Ind. Controls Yes Yes Yes Yes Yes Yes Robust standard errors (clustered at the community level) in parentheses. Wild bootstrap p-values in brackets. The asterisks next to the coefficients are for p-values associated with our main (non-wild bootstrap) regressions. *** p<0.01, ** p<0.05, * p<0.1. The dependent variables are standardised maths score and grade-for-age at age 8 and 12. The variable of interest captures prenatal exposure to seasonal food insecurity (full 9 months exposure) standardised to have mean 0 and standard deviation 1 within each community. Ind. controls include age of child in months, number of older siblings and dummies for gender, child ethnicity and prematurity. HH controls include household wealth index and dummies for gender of household head and mother’s education. For maths outcome, we restrict the sample to children for which we observe the outcomes of interest at all age (round) stages. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab 4 Discussion Three broad classes of factors may drive our results on the gender imbalances in seasonal food insecurity in utero. First, gender imbalances may be explained by the fact male foetuses are more vulnerable than girls in utero. Deficiencies in human capital development, the so-called scarring effects, may accumulate overtime. Second, the higher vulnerability of boys in utero may result in higher selective mortality in utero, the so-called culling effect and the survival of the stronger boys at and after births. Such alternative explanation would bias the coefficient of interest downward. Third, gender discrimination is usually expected against girls in such a context. Compensating mechanisms would therefore have mitigated the gender imbalances found in the previous section.18 4.1 Mortality selection Our sample only includes surviving children. Although our prenatal shock is of relatively mild (and frequent) nature, we cannot exclude that mortality in utero would drive our estimates towards zero. Surviving children may indeed appear to be the strongest, the healthiest and those with better genes. Similarly, the gender-based analysis could be biased due to differentiated mortality risk for boys and girls. The medical research indeed documents that male foetuses are more vulnerable to shocks and at greater mortality risk than female foetuses [Catalano et al., 2006, Eriksson et al., 2010, Kraemer, 2000, Mizuno, 2000, Shettles, 1961]. Empirical studies also document how negative prenatal exposure could alter sex composition at birth [Almond et al., 2010, Almond & Mazumder, 2011, Dagnelie et al., 2018, Nilsson, 2017, Valente, 2015, Van Ewijk, 2011]. We cannot directly test the effect of the shock on prenatal death differential between boys and girls. We do not have information about miscarriages and prenatal deaths. However, following Van Ewijk [2011], we test the role of selection by estimating the exposure effect on the probability of being a male at ages 1, 5, 8 and 12. We do not find strong evidence for mortality selection. Food insecurity shocks in utero do not seem to translate into changes in the sex ratio (Table 4). Only at age 5, we find a positive coefficient significantly different from zero at 90 percent level of confidence. Such coefficient cannot explain the stronger detrimental impact for boys compared to girls at ages 8 and 12. So, the causal interpretation of our main results is not threatened by mortality selection in utero or after birth. Gender imbalances in cognitive development cannot be explained by selective mortality. Table 4 Effect of Exposure on the Probability of the Child Surveyed Is Male . Logit odds ratio . . (1) . (2) . (3) . (4) . . Age 1 . Age 5 . Age 8 . Age 12 . Panel A: Child surveyed is male, without controls Exposure-Std 1.108 1.112 1.108 1.105 (0.074) (0.076) (0.079) (0.078) Observations 1,875 1,793 1,768 1,754 Panel B: Child surveyed is male, with controls Exposure-Std 1.094 1.109* 1.106 1.098 (0.065) (0.070) (0.071) (0.071) Observations 1,846 1,770 1,745 1,731 Community FE Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes . Logit odds ratio . . (1) . (2) . (3) . (4) . . Age 1 . Age 5 . Age 8 . Age 12 . Panel A: Child surveyed is male, without controls Exposure-Std 1.108 1.112 1.108 1.105 (0.074) (0.076) (0.079) (0.078) Observations 1,875 1,793 1,768 1,754 Panel B: Child surveyed is male, with controls Exposure-Std 1.094 1.109* 1.106 1.098 (0.065) (0.070) (0.071) (0.071) Observations 1,846 1,770 1,745 1,731 Community FE Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is dummy indicating child born is boy. The main independent variable is standardised prenatal exposure to seasonal food insecurity (exposure in whole nine months). Panel A reports results estimated without controls, while panel B shows results estimated with the following control variables: number of older siblings, household wealth index, dummies for child ethnicity, prematurity, gender of household head and mother’s education. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 4 Effect of Exposure on the Probability of the Child Surveyed Is Male . Logit odds ratio . . (1) . (2) . (3) . (4) . . Age 1 . Age 5 . Age 8 . Age 12 . Panel A: Child surveyed is male, without controls Exposure-Std 1.108 1.112 1.108 1.105 (0.074) (0.076) (0.079) (0.078) Observations 1,875 1,793 1,768 1,754 Panel B: Child surveyed is male, with controls Exposure-Std 1.094 1.109* 1.106 1.098 (0.065) (0.070) (0.071) (0.071) Observations 1,846 1,770 1,745 1,731 Community FE Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes . Logit odds ratio . . (1) . (2) . (3) . (4) . . Age 1 . Age 5 . Age 8 . Age 12 . Panel A: Child surveyed is male, without controls Exposure-Std 1.108 1.112 1.108 1.105 (0.074) (0.076) (0.079) (0.078) Observations 1,875 1,793 1,768 1,754 Panel B: Child surveyed is male, with controls Exposure-Std 1.094 1.109* 1.106 1.098 (0.065) (0.070) (0.071) (0.071) Observations 1,846 1,770 1,745 1,731 Community FE Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is dummy indicating child born is boy. The main independent variable is standardised prenatal exposure to seasonal food insecurity (exposure in whole nine months). Panel A reports results estimated without controls, while panel B shows results estimated with the following control variables: number of older siblings, household wealth index, dummies for child ethnicity, prematurity, gender of household head and mother’s education. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab 4.2 Parental responses versus biological effects Parents may respond to in utero shocks by adapting their investment towards children either to compensate or reinforce children’s endowments. If investment responses are compensatory, the effect of prenatal food insecurity shock will tend to understate biological effects. However, parents may also decide to reinforce children’s endowment. In that case our baseline results may overestimate the true biological effect. Recent empirical studies reviewed in Almond & Mazumder [2013] indeed find that parental investments reinforce initial endowment differences. In our case, that would mean that parents discriminate against boys more vulnerable in utero. That would be quite surprising given the abundant report on gender discrimination against girls in Ethiopia [Ayalew, 2005]. On the contrary, compensatory investments would attenuate the established gender imbalances in the previous section. Following Adhvaryu & Nyshadham [2016] and using the YLE survey, we assess whether the behavioural response from parents is driven by food insecurity shock in utero. Specifically, we test the effect of the shock on parental investments at ages 8 and 12 to investigate parental response once the cognitive endowment is realised.19 We employ several parental investment outcomes. These include an indicator to school enrolment, the number of study hours at home (including extra tuition) and an indicator to whether a child is enrolled into a private or a public school, an indicator if parents paid for school fees or tuition (last 12 months), an indicator if parents paid any medical expenditure (last 12 months), the number of meals a child had in the last 24 hours and the total number of food variety a child experienced in the last 24 hours.20 Panel E in Table D.1 in online Appendix D reports the descriptive statistics of these variables. In Table 5, we explore the role of parental investments that are directly related to education that happened at age 8 and 12.21 Overall, with other outcomes, we confirm the conclusions by Miller [2017] that there is limited role for parental investment.22 One exception is the fact the shock decreases the odds of being enrolled in school for girls at age 12. Under-investment in girls’ education at age 12 would tend to attenuate the gender imbalances against boys found earlier. Such under-investment is not confirmed using the time available for study at home or the probability to be sent to a private school or expenditures on school fees or tuition (educational expenditures). Table 5 Childhood Parental Educational Investments . Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Enrolled in to school Exposure-Std 0.892* 0.652 0.836 0.726 1.015 0.359** (0.061) (0.188) (0.150) (0.364) (0.093) (0.184) Observations 1,629 1,398 749 666 742 437 Panel B: Study hour at home(including extra tuition) Exposure-Std -0.005 0.043 -0.003 0.034 -0.005 0.052 (0.025) (0.031) (0.043) (0.043) (0.036) (0.052) Observations 1,744 1,732 904 900 840 832 Panel C: In private school Exposure-Std 1.192 0.940 1.144 0.996 1.277 0.945 (0.278) (0.166) (0.356) (0.264) (0.653) (0.260) Observations 757 851 269 353 323 308 Panel D: Education expenditures (school fees or tuition) Exposure-Std 0.868 1.035 1.072 0.836 0.787 1.247 (0.101) (0.126) (0.222) (0.117) (0.129) (0.225) Observations 1,555 1,631 782 825 724 723 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes . Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Enrolled in to school Exposure-Std 0.892* 0.652 0.836 0.726 1.015 0.359** (0.061) (0.188) (0.150) (0.364) (0.093) (0.184) Observations 1,629 1,398 749 666 742 437 Panel B: Study hour at home(including extra tuition) Exposure-Std -0.005 0.043 -0.003 0.034 -0.005 0.052 (0.025) (0.031) (0.043) (0.043) (0.036) (0.052) Observations 1,744 1,732 904 900 840 832 Panel C: In private school Exposure-Std 1.192 0.940 1.144 0.996 1.277 0.945 (0.278) (0.166) (0.356) (0.264) (0.653) (0.260) Observations 757 851 269 353 323 308 Panel D: Education expenditures (school fees or tuition) Exposure-Std 0.868 1.035 1.072 0.836 0.787 1.247 (0.101) (0.126) (0.222) (0.117) (0.129) (0.225) Observations 1,555 1,631 782 825 724 723 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes For binary outcomes (indicators of school enrolment, type of school enrolled in to and educational expenditures), Logit odds ratio are reported. Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is indicator to school enrolment (panel A), study hours at home (including extra tuition) (panel B), and indicator to whether a child is enrolled in to private or public school (panel C), indicator if parents paid for school fees or tuition for the child (panel D). Controls include (X) household wealth index, number of older siblings and dummies for gender, gender of household head, mother’s education, child ethnicity and prematurity. In the school enrolment regressions many observations are dropped because in several communities all children reported being in school. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 5 Childhood Parental Educational Investments . Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Enrolled in to school Exposure-Std 0.892* 0.652 0.836 0.726 1.015 0.359** (0.061) (0.188) (0.150) (0.364) (0.093) (0.184) Observations 1,629 1,398 749 666 742 437 Panel B: Study hour at home(including extra tuition) Exposure-Std -0.005 0.043 -0.003 0.034 -0.005 0.052 (0.025) (0.031) (0.043) (0.043) (0.036) (0.052) Observations 1,744 1,732 904 900 840 832 Panel C: In private school Exposure-Std 1.192 0.940 1.144 0.996 1.277 0.945 (0.278) (0.166) (0.356) (0.264) (0.653) (0.260) Observations 757 851 269 353 323 308 Panel D: Education expenditures (school fees or tuition) Exposure-Std 0.868 1.035 1.072 0.836 0.787 1.247 (0.101) (0.126) (0.222) (0.117) (0.129) (0.225) Observations 1,555 1,631 782 825 724 723 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes . Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Enrolled in to school Exposure-Std 0.892* 0.652 0.836 0.726 1.015 0.359** (0.061) (0.188) (0.150) (0.364) (0.093) (0.184) Observations 1,629 1,398 749 666 742 437 Panel B: Study hour at home(including extra tuition) Exposure-Std -0.005 0.043 -0.003 0.034 -0.005 0.052 (0.025) (0.031) (0.043) (0.043) (0.036) (0.052) Observations 1,744 1,732 904 900 840 832 Panel C: In private school Exposure-Std 1.192 0.940 1.144 0.996 1.277 0.945 (0.278) (0.166) (0.356) (0.264) (0.653) (0.260) Observations 757 851 269 353 323 308 Panel D: Education expenditures (school fees or tuition) Exposure-Std 0.868 1.035 1.072 0.836 0.787 1.247 (0.101) (0.126) (0.222) (0.117) (0.129) (0.225) Observations 1,555 1,631 782 825 724 723 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes For binary outcomes (indicators of school enrolment, type of school enrolled in to and educational expenditures), Logit odds ratio are reported. Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is indicator to school enrolment (panel A), study hours at home (including extra tuition) (panel B), and indicator to whether a child is enrolled in to private or public school (panel C), indicator if parents paid for school fees or tuition for the child (panel D). Controls include (X) household wealth index, number of older siblings and dummies for gender, gender of household head, mother’s education, child ethnicity and prematurity. In the school enrolment regressions many observations are dropped because in several communities all children reported being in school. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 6 reports results from parental health and nutritional investments such as medical expenditures; meal frequency or the food variety. In this case too, we find little evidence that parents respond to the shock through health and nutritional investments. At age 12, in utero exposure to food insecurity decreases the number of meal frequency, but with no apparent significant difference between boys and girls.23 Table 6 Childhood Parental Health and Nutritional Investments Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Medical expenditures Exposure-Std 1.145 1.140 1.097 1.214 1.203 1.094 (0.133) (0.163) (0.192) (0.328) (0.235) (0.283) Observations 1,746 1,734 905 901 841 829 Panel B: Meal frequency in the last 24 hours Exposure-Std 0.010 -0.052* -0.004 -0.055 0.019 -0.064 (0.021) (0.027) (0.040) (0.048) (0.035) (0.041) Observations 1,746 1,728 905 897 841 831 Panel C: Food variety in the last 24 hours Exposure-Std -0.054 -0.030 -0.092 0.041 -0.040 -0.077 (0.054) (0.073) (0.137) (0.124) (0.144) (0.078) Observations 1,745 1,727 905 897 840 830 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Medical expenditures Exposure-Std 1.145 1.140 1.097 1.214 1.203 1.094 (0.133) (0.163) (0.192) (0.328) (0.235) (0.283) Observations 1,746 1,734 905 901 841 829 Panel B: Meal frequency in the last 24 hours Exposure-Std 0.010 -0.052* -0.004 -0.055 0.019 -0.064 (0.021) (0.027) (0.040) (0.048) (0.035) (0.041) Observations 1,746 1,728 905 897 841 831 Panel C: Food variety in the last 24 hours Exposure-Std -0.054 -0.030 -0.092 0.041 -0.040 -0.077 (0.054) (0.073) (0.137) (0.124) (0.144) (0.078) Observations 1,745 1,727 905 897 840 830 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes For binary outcomes (indicator of if parents paid any medical expenditure to the child), Logit odds ratio are reported. Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is indicator if parents paid any medical expenditure to the child (panel A), the number of meals a child ate in the last 24 hours (panel B) and total number of food variety a child ate in the last 24 hours (panel C). Controls include (X) household wealth index, number of older siblings and dummies for gender, gender of household head, mother’s education, child ethnicity and prematurity. In the school enrolment regressions many observations are dropped because in several communities all children reported being in school. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab Table 6 Childhood Parental Health and Nutritional Investments Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Medical expenditures Exposure-Std 1.145 1.140 1.097 1.214 1.203 1.094 (0.133) (0.163) (0.192) (0.328) (0.235) (0.283) Observations 1,746 1,734 905 901 841 829 Panel B: Meal frequency in the last 24 hours Exposure-Std 0.010 -0.052* -0.004 -0.055 0.019 -0.064 (0.021) (0.027) (0.040) (0.048) (0.035) (0.041) Observations 1,746 1,728 905 897 841 831 Panel C: Food variety in the last 24 hours Exposure-Std -0.054 -0.030 -0.092 0.041 -0.040 -0.077 (0.054) (0.073) (0.137) (0.124) (0.144) (0.078) Observations 1,745 1,727 905 897 840 830 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes Full sample . Boys . Girls . . (1) . (2) . (3) . (4) . (5) . (6) . . Age 8 . Age 12 . Age 8 . Age 12 . Age 8 . Age 12 . Panel A: Medical expenditures Exposure-Std 1.145 1.140 1.097 1.214 1.203 1.094 (0.133) (0.163) (0.192) (0.328) (0.235) (0.283) Observations 1,746 1,734 905 901 841 829 Panel B: Meal frequency in the last 24 hours Exposure-Std 0.010 -0.052* -0.004 -0.055 0.019 -0.064 (0.021) (0.027) (0.040) (0.048) (0.035) (0.041) Observations 1,746 1,728 905 897 841 831 Panel C: Food variety in the last 24 hours Exposure-Std -0.054 -0.030 -0.092 0.041 -0.040 -0.077 (0.054) (0.073) (0.137) (0.124) (0.144) (0.078) Observations 1,745 1,727 905 897 840 830 Community FE Yes Yes Yes Yes Yes Yes Birth Month FE Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes For binary outcomes (indicator of if parents paid any medical expenditure to the child), Logit odds ratio are reported. Robust standard errors (clustered at the community level) in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is indicator if parents paid any medical expenditure to the child (panel A), the number of meals a child ate in the last 24 hours (panel B) and total number of food variety a child ate in the last 24 hours (panel C). Controls include (X) household wealth index, number of older siblings and dummies for gender, gender of household head, mother’s education, child ethnicity and prematurity. In the school enrolment regressions many observations are dropped because in several communities all children reported being in school. Note that the odds ratio are interpreted as follows: if the odds are greater than one, they indicate positive effects; but if the odds are less than one, they are interpreted as negative effects. Open in new tab 5 Conclusions We examine the effect of in utero seasonal food insecurity on childhood cognitive development and grade-for-age. We exploit a unique dataset from the YLE. We estimate the effect of variation in the number of days of exposure to prenatal food insecurity on these outcomes, controlling for community and birth month fixed effects together with child and household characteristics. The inclusion of community and month of birth fixed effects means our estimations are unlikely to be affected by seasonality effects, or unobserved heterogeneity at the community level. We find that at age 8, maths are adversely affected by in utero exposure to seasonal food insecurity, but only for boys. At age 12, gender imbalances exacerbate. At age 12, a standard deviation increase in food insecurity in utero decreases maths scores by about one third of a standard deviation for boys, almost three times the decrease observed for girls. Moreover, at age 12, we find that food insecurity in utero decreases the odds of being on the correct grade, but only for boys. Based on the lack of selective mortality in utero and after birth and weak evidence for differentiated parental investment, we conjecture that scarring effects, particularly fierce for male foetuses, accumulate overtime. Such detrimental impacts are likely to have long-term consequences on socio-economic outcomes. Policy interventions that address seasonal food insecurity and programs that target pregnant women to enhance their resilience to seasonal food shortages could protect the development of children and minimise the long-term economic cost. Social safety net or cash transfer programs together with nutrition and micro-nutrient supplementation programs are obvious policy options. In Ethiopia, starting from 2005, the Productive Safety Net Programme (PSNP) aims at addressing seasonal food insecurity. Unfortunately, our data do not allow us to investigate the mitigating effect of the PSNP since the sampled children were in utero between 2000 and 2002, before the implementation of the PSNP. Understanding how specific programs build resilience to seasonal food insecurity is a path for future research. Finally, our paper faces other data limitations. While the focus on seasonal food insecurity seems is arguably policy relevant, the lack of weather data does not allow us to benchmark our results against large-scale weather shocks. Similarly, another limitation is the indirect nature of our evidence behind the gender imbalances. Data on birth weights, miscarriages, stillbirth and neonatal mortality would help us better understand the mechanisms potentially explaining gender imbalances to seasonal food insecurity. Supplementary material Supplementary material is available at Journal of African Economies online. Acknowledgments We thank Ian Walker, Kalle Hirvonen, Maria Navarro Paniagua and Colin Green for their comments. All errors and opinions expressed remain our own. The data used in this publication come from Young Lives, a 15-year study of the changing nature of childhood poverty in Ethiopia, India, Peru and Vietnam (www.younglives.org.uk). Young Lives is funded by UK aid from the Department for International Development (DFID). The views expressed here are those of the authors. They are not necessarily those of Young Lives, the University of Oxford, DFID or other funders. Footnotes 1 " The data may not be representative of the country since the information is obtained from chronically food-insecure woredas (districts). 2 " The literature on the long-term effect of in utero shocks has relied on rare and extreme events such as famine, war and terrorist attacks. In addition to the likely fiercer selection in utero, it has been difficult to distinguish the nutritional impact of shocks from the psychological stress associated with the shock [Currie & Vogl, 2013]. We are not able to distinguish between these insults but in our case, similar to Miller [2017] and Nilsson [2017], we are more likely to directly capture the nutritional impact of shocks. 3 " Since childhood height is correlated with later cognitive development outcomes [Case & Paxson, 2008], our choice of outcomes (maths score and grade-for-age) also complements the analysis in Miller [2017]. 4 " The vulnerability of boys in utero is consistent with the medical literature [Catalano et al., 2006, Eriksson et al., 2010, Kraemer, 2000, Mizuno, 2000, Shettles, 1961]. 5 " See http://www.younglives.org.uk/content/sampling-and-attrition for details. 6 " The survey also collects similar information for the older cohort, born around 1994–1995. These children were 7–8 years old during the first round survey in 2002. We do not have birth information such as prematurity for this cohort that are essential for computing our exposure variables. Thus, this cohort cannot be exploited for our main analysis. We will nonetheless use the information about this cohort to assess the relationship between cognitive development and long-term education outcomes to shed light on the long-run significance of our results. 7 " Our main outcome variables are measured in round 3 and round 4 of the YLE surveys. It is important to note that round 3 surveys were implemented between November and February of 2009/10 and round 4 surveys were implemented between November and February of 2013/14. As discussed below, these months are relatively food secure months as they correspond to the post-harvest period. 8 " For binary outcomes, logistic regressions are used instead. 9 " Similar to Miller [2017], date of birth for each child is calculated using age of child in days and the date of interview from the first survey round. 10 " In online Appendix D.2, we discuss the importance of standardisation within communities, together with other functional assumptions (e.g., linearity). Moreover, we show the robustness of our results to using non-standardised treatment variable. 11 " Sibling fixed effects cannot be used in our paper since the YLE surveys did not collect sibling information for our outcomes of interest. 12 " We do not report wild bootstrap p-values for the grade-for-age outcome as the results on this outcome are estimated using logistic regression. 13 " We also report results from other cognitive development measures collected by the YLE study: the Early Grade Reading Assessment (EGRA) and the Peabody Picture Vocabulary Test (PPVT). The EGRA is orally assessed only at age 8. It is implemented to measure the most basic skills for literacy acquisition in the early grades. It involves recognising letters of the alphabet, reading simple words, understanding sentences and paragraphs and listening with comprehension. The PPVT is a widely used test of receptive vocabulary. The descriptive statistics of these variables are presented in Panel A of Table D.1 in online Appendix D.5. These tests are adapted to different languages spoken in the country. Difficulty levels may have changed during translation, and as a result, it is recommended that comparison must be within languages [Cueto & Leon, 2012, Singh, 2015]. We cannot limit our data to a certain language in the country since the geographical concentration of languages in Ethiopia would cancel out the variation in the exposure variable. As a result, we are cautious about interpreting results from these two tests. 14 " Using food insecurity information from 2002 community survey confirms our results but with much lower magnitude. Results are discussed in online Appendix D.2. 15 " We describe the reliability of the community-level food insecurity information in the construction of our in utero exposure in online Appendix C. 16 " We also investigate the impact by trimester, finding stronger effects for boys in the first and second trimesters. For the sake of presentation, we only discuss these results in online Appendix D based on Table D.10. 17 " Detailed results of Table 3 including control variables are provided in Tables D.2 and D.3 of online Appendix D. Online Appendix D provides and discusses additional robustness checks. Gender imbalances in the effect of exposure on other tests are also apparent in Table D.4 of online Appendix D. Exposure decreases reading at age 8, significantly more for boys. Though we find no significant effect on PPVT, exposure has unexpected and positive effect on girl’s PPVT score at age 12. We do not, however, interpret further results from these two outcomes given the lack of accuracy of the cognitive tests in Ethiopia (see footnote 11). Although efficiency might be affected in some cases, our results are largely robust to not standardising the measure of exposure to food insecurity in utero (more subject to high-leverage communities, see Table D.11), to using round 1 food insecurity information (similar to Miller [2017]) to capture seasonal food insecurity at the community level (Table D.13), to non-linear effects in exposure (Table D.14), to relaxing the restriction of including only children for which we observe the outcomes of interest at all age (round) stages in the samples of regressions using the maths score as a dependent variable (Table D.15). These results are further discussed in online Appendix D. 18 " Those mechanisms can equally be seen as threats to the general identification but help us to understand the gender imbalances. Other identification threats, with no obvious gender bias, may affect the magnitude of our coefficients. In online Appendix D, we therefore also examine how our results may be threatened by (1) fertility selection; (2) reporting errors; (3) the existence of other mechanisms; (4) attrition and missing data; and (5) exposure to seasonal food insecurity after birth. Some of these identification threats are also tested on gender-stratified samples to assess their possible consequences on the consistency of our results on the gender imbalances of seasonal food security in utero. 19 " We focus on investment carried out at ages 8 and 12. On the one hand, parents at this stage can observe the realised cognition of their children to decide to reinforce or compensate it. On the other hand, it helps us understand whether differential investment at ages 8 and 12 could explain the difference in the observed effect of the shock on cognition between ages 8 and 12. 20 " Parents/children were asked 7 yes/no questions related to meal frequency for a child. Specifically, they were asked if the child ate any food before breakfast, breakfast, food between breakfast and midday meal, midday meal, food between midday meal and evening meal, evening meal and food after the evening meal. We computed meal frequency for each round (age) as the sum of these frequencies. We top coded at six meals. Moreover, parents/children were asked whether the child ate different types of foods in the last 24 hours. They were asked 17 (at age 8) and 15 (at age 12) yes/no questions. We computed food variety variables for each round (age) as the sum of these frequencies. We top coded at 10 food types. 21 " Nonetheless, in Table D.5 of online Appendix D, we also report results on investment on preschool, an educational investment that happened on or before age 5. We find no significant effect of exposure on preschool investment. 22 " In an unpublished paper, Fan & Porter [2018] investigate the relationship between cognitive ability and educational fees in Ethiopia. They find evidence of compensating effects when cognitive ability is instrumented by weather shocks at early age. However, comparability is limited given the differences in shocks, the timing of the shocks, their IV approach and the lack of heterogeneous analysis across gender. 23 " Gender-specific pre-natal investment is not expected since sex detection before birth is very uncommon in Ethiopia. We nonetheless test the impact of in utero exposure on pre-natal and neo-natal (BCG) investments. We do not find any significant impact of in utero exposure to food insecurity (see Table D.6 in online Appendix D). Moreover, similar to Yi et al. [2015], but at the cost of introducing endogeneity, we have also assessed how much our coefficients of interest may be altered by directly including proxies for possible parental responses into the main regressions. The results in Table D.7, Table D.8 and Table D.9 from online Appendix D confirm that our main results remain unchanged. Parental responses cannot explain the gender imbalances found in our main results. Furthermore, other sources of heterogeneity may explain why the effects accumulate overtime, indirectly shedding light on differentiated ability of households to deal with food insecurity in utero. Results from heterogeneity analysis based on wealth (Table D.16); access to market (Table D.17) and access to road (Table D.18) are commented and discussed in online Appendix D. References Abay , K. and Hirvonen , K. ( 2017 ). ‘ Does market access mitigate the impact of seasonality on child growth? Panel data evidence from Northern Ethiopia ’, The Journal of Development Studies , 53 ( 9 ): 1414 – 29 . Google Scholar Crossref Search ADS WorldCat Adhvaryu , A. and Nyshadham , A. ( 2016 ). ‘ Endowments at birth and parents’ investments in children ’, The Economic Journal , 126 ( 593 ): 781 – 820 . Google Scholar Crossref Search ADS PubMed WorldCat Alderman , H. , Hoddinott , J., and Kinsey , B. ( 2006 ). ‘ Long term consequences of early childhood malnutrition ’, Oxford Economic Papers , 58 ( 3 ): 450 – 74 . Google Scholar Crossref Search ADS WorldCat Alem , Y. and Söderbom , M. ( 2012 ). ‘ Household-level consumption in urban Ethiopia: the effects of a large food price shock ’, World Development , 40 ( 1 ): 146 – 62 . Google Scholar Crossref Search ADS WorldCat Almond , D. and Currie , J. ( 2011 ). ‘ Killing me softly: the fetal origins hypothesis ’, The Journal of Economic Perspectives , 25 ( 3 ): 153 – 72 . Google Scholar Crossref Search ADS PubMed WorldCat Almond D. , Edlund L., Li H., Zhang J. ( 2010 ) ‘ Long-term effects of early-life development: evidence from the 1959 to 1961 China famine ’, in The Economic Consequences of Demographic Change in East Asia, NBER-EASE Volume 19 . Chicago : University of Chicago Press , pp. 321 – 45 . Google Scholar Crossref Search ADS Google Scholar Google Preview WorldCat COPAC Almond , D. and Mazumder , B. ( 2013 ). ‘ Fetal origins and parental responses ’, Annual Review of Economics , 5 ( 1 ): 37 – 56 . Google Scholar Crossref Search ADS WorldCat Almond , D. , Mazumder , B., and Ewijk , R. ( 2015 ). ‘ In utero Ramadan exposure and children’s academic performance ’, The Economic Journal , 125 ( 589 ): 1501 – 33 . Google Scholar Crossref Search ADS WorldCat Almond , D. and Mazumder , B. A. ( 2011 ). ‘ Health capital and the prenatal environment: the effect of Ramadan observance during pregnancy ’, American Economic Journal: Applied Economics , 3 ( 4 ): 56 – 85 . Google Scholar Crossref Search ADS WorldCat Ayalew , T. ( 2005 ). ‘ Parental preference, heterogeneity, and human capital inequality ’, Economic Development and Cultural Change , 53 ( 2 ): 381 – 407 . Google Scholar Crossref Search ADS WorldCat Barker , D. J. ( 1990 ). ‘ The fetal and infant origins of adult disease ’, BMJ: British Medical Journal , 301 ( 6761 ): 1111 . Google Scholar Crossref Search ADS WorldCat Barnett , I. , Ariana , P., Petrou , S., Penny , M. E., Galab , S., Woldehanna , T., Escobal , J. A., Plugge , E., Boyden , J., et al. ( 2013 ). ‘ Cohort profile: the Young Lives study ’, International Journal of Epidemiology , 42 ( 3 ): 701 – 8 . Google Scholar Crossref Search ADS PubMed WorldCat Berg , G. J. , Pinger , P. R., and Schoch , J. ( 2016 ). ‘ Instrumental variable estimation of the causal effect of hunger early in life on health later in life ’, The Economic Journal , 126 ( 591 ): 465 – 506 . Google Scholar Crossref Search ADS WorldCat Black , S. E. , Devereux , P. J., and Salvanes , K. G. ( 2016 ). ‘ Does grief transfer across generations? Bereavements during pregnancy and child outcomes ’, American Economic Journal: Applied Economics , 8 ( 1 ): 193 – 223 . Google Scholar Crossref Search ADS WorldCat Buckles , K. S. and Hungerman , D. M. ( 2013 ). ‘ Season of birth and later outcomes: old questions, new answers ’, Review of Economics and Statistics , 95 ( 3 ): 711 – 24 . Google Scholar Crossref Search ADS PubMed WorldCat Cameron , A. C. , Gelbach , J. B., and Miller , D. L. ( 2008 ). ‘ Bootstrap-based improvements for inference with clustered errors ’, The Review of Economics and Statistics , 90 ( 3 ): 414 – 27 . Google Scholar Crossref Search ADS WorldCat Cameron , A. C. and Miller , D. L. ( 2015 ). ‘ A practitioner’s guide to cluster-robust inference ’, Journal of Human Resources , 50 ( 2 ): 317 – 72 . Google Scholar Crossref Search ADS WorldCat Case , A. and Paxson , C. ( 2008 ). Stature and status: height, ability, and labor market outcomes . Journal of Political Economy , 116 ( 3 ): 499 – 532 . Google Scholar Crossref Search ADS PubMed WorldCat Catalano , R. , Bruckner , T., Marks , A. R., and Eskenazi , B. ( 2006 ). ‘ Exogenous shocks to the human sex ratio: the case of September 11, 2001 in New York city ’, Human Reproduction , 21 ( 12 ): 3127 – 31 . Google Scholar Crossref Search ADS PubMed WorldCat Cueto , S. and Leon , J. ( 2012 ). Psychometric characteristics of cognitive development and achievement instruments in round 3 of young lives . Young Lives Technical Note 25, Young Lives, University of Oxford . Currie J. , Thomas D. ( 2001 ) ‘ Early test scores, school quality and SES: long-run effects on wage and employment outcomes ’, in Research in Labor Economics: Worker Wellbeing in a Changing Labor Market , vol. 20 . Bingley, UK : Emerald Group Publishing Limited , pp. 103 – 32 . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Currie , J. and Vogl , T. ( 2013 ). ‘ Early-life health and adult circumstance in developing countries ’, Annual Review of Economics , 5 ( 1 ): 1 – 36 . Google Scholar Crossref Search ADS WorldCat Dagnelie O. , De Luca G. D., Maystadt J.-F. ( 2018 ) ‘ Violence, selection and infant mortality in Congo ’, Journal of Health Economics , 59 : 153 – 77 . Google Scholar Crossref Search ADS PubMed WorldCat Dercon , S. ( 2004 ). ‘ Growth and shocks: evidence from rural Ethiopia ’, Journal of Development Economics , 74 ( 2 ): 309 – 29 . Google Scholar Crossref Search ADS WorldCat Dercon , S. and Krishnan , P. ( 2000 ). ‘ Vulnerability, seasonality and poverty in Ethiopia ’, The Journal of Development Studies , 36 ( 6 ): 25 – 53 . Google Scholar Crossref Search ADS WorldCat Dercon , S. and Porter , C. ( 2014 ). ‘ Live Aid revisited: long-term impacts of the 1984 Ethiopian famine on children ’, Journal of the European Economic Association , 12 ( 4 ): 927 – 48 . Google Scholar Crossref Search ADS WorldCat Eriksson , J. G. , Kajantie , E., Osmond , C., Thornburg , K., and Barker , D. J. ( 2010 ). ‘ Boys live dangerously in the womb ’, American Journal of Human Biology , 22 ( 3 ): 330 – 5 . Google Scholar Crossref Search ADS PubMed WorldCat Fan , W. and Porter , C. ( 2018 ). ‘ Reinforcement or Compensation? Parental Responses to Children’s Revealed Human Capital Levels ’, Technical report, Young Lives . FAO ( 2009 ) ‘ The State of Food Insecurity in the World: Economic Crises—Impacts and Lessons Learned ’, Technical report . Food and Agriculture Organization of the: United Nations (FAO) . Glewwe , P. , Jacoby , H. G., and King , E. M. ( 2001 ). ‘ Early childhood nutrition and academic achievement: a longitudinal analysis ’, Journal of Public Economics , 81 ( 3 ): 345 – 68 . Google Scholar Crossref Search ADS WorldCat Gørgens , T. , Meng , X., and Vaithianathan , R. ( 2012 ). ‘ Stunting and selection effects of famine: a case study of the Great Chinese Famine ’, Journal of Development Economics , 97 ( 1 ): 99 – 111 . Google Scholar Crossref Search ADS WorldCat Heckman , J. J. ( 2007 ). ‘ The economics, technology, and neuroscience of human capability formation ’, Proceedings of the National Academy of Sciences , 104 ( 33 ): 13250 – 5 . Google Scholar Crossref Search ADS WorldCat Heckman , J. J. and Masterov , D. V. ( 2007 ). ‘ The productivity argument for investing in young children ’. Applied Economic Perspectives and Policy , 29 ( 3 ): 446 – 93 . OpenURL Placeholder Text WorldCat Hernández-Julián , R. , Mansour , H., and Peters , C. ( 2014 ). ‘ The effects of intrauterine malnutrition on birth and fertility outcomes: evidence from the 1974 Bangladesh famine ’, Demography , 51 ( 5 ): 1775 – 96 . Google Scholar Crossref Search ADS PubMed WorldCat Hill R. V. , Porter C. ( 2017 ) ‘ Vulnerability to drought and food price shocks: evidence from Ethiopia ’, World Development , 96 : 65 – 77 . Google Scholar Crossref Search ADS WorldCat Hoddinott J. , Kumar N., Taffesse A. S., Tefera M. ( 2011 ) ‘ Evaluation of Ethiopia’s Food Security Program: Documenting progress in the implementation of the Productive Safety Nets Programme and the Household Asset Building Programme ’, in Unpublished report, International Food Policy Research Institute and Ethiopian Development Research Institute . Kraemer , S. ( 2000 ). ‘ The fragile male ’, BMJ: British Medical Journal , 321 ( 7276 ): 1609 . Google Scholar Crossref Search ADS WorldCat Lindeboom , M. , Portrait , F., and Van den Berg , G. J. ( 2010 ). ‘ Long-run effects on longevity of a nutritional shock early in life: the Dutch Potato famine of 1846–1847 ’, Journal of Health Economics , 29 ( 5 ): 617 – 29 . Google Scholar Crossref Search ADS PubMed WorldCat Lokshin , M. and Radyakin , S. ( 2012 ). ‘ Month of birth and children’s health in India ’, Journal of Human Resources , 47 ( 1 ): 174 – 203 . Google Scholar Crossref Search ADS WorldCat Luo , Z. , Mu , R., and Zhang , X. ( 2006 ). ‘ Famine and overweight in China ’, Applied Economic Perspectives and Policy , 28 ( 3 ): 296 – 304 . OpenURL Placeholder Text WorldCat Maluccio , J. A. , Hoddinott , J., Behrman , J. R., Martorell , R., Quisumbing , A. R., and Stein , A. D. ( 2009 ). ‘ The impact of improving nutrition during early childhood on education among Guatemalan adults ’, The Economic Journal , 119 ( 537 ): 734 – 63 . Google Scholar Crossref Search ADS WorldCat McEniry , M. and Palloni , A. ( 2010 ). ‘ Early life exposures and the occurrence and timing of heart disease among the older adult Puerto Rican population ’, Demography , 47 ( 1 ): 23 – 43 . Google Scholar Crossref Search ADS PubMed WorldCat Miller R. ( 2017 ) ‘ Childhood health and prenatal exposure to seasonal food scarcity in Ethiopia ’, World Development , 99 : 350 – 76 . Google Scholar Crossref Search ADS WorldCat Mizuno , R. ( 2000 ). ‘ The male/female ratio of fetal deaths and births in Japan ’, The Lancet , 356 ( 9231 ): 738 – 9 . Google Scholar Crossref Search ADS WorldCat Mu , R. and Zhang , X. ( 2011 ). ‘ Why does the Great Chinese Famine affect the male and female survivors differently? Mortality selection versus son preference ’, Economics & Human Biology , 9 ( 1 ): 92 – 105 . Google Scholar Crossref Search ADS WorldCat Neelsen , S. and Stratmann , T. ( 2011 ). ‘ Effects of prenatal and early life malnutrition: evidence from the Greek famine ’, Journal of Health Economics , 30 ( 3 ): 479 – 88 . Google Scholar Crossref Search ADS PubMed WorldCat Nilsson , J. P. ( 2017 ). ‘ Alcohol availability, prenatal conditions, and long-term economic outcomes ’, Journal of Political Economy , 125 ( 4 ): 1149 – 207 . Google Scholar Crossref Search ADS WorldCat Porter , C. ( 2012 ). ‘ Shocks, consumption and income diversification in rural Ethiopia ’, Journal of Development Studies , 48 ( 9 ): 1209 – 22 . Google Scholar Crossref Search ADS WorldCat Roseboom , T. J. , Painter , R. C., van Abeelen , A. F., Veenendaal , M. V., and de Rooij , S. R. ( 2011 ). ‘ Hungry in the womb: what are the consequences? Lessons from the Dutch famine ’, Maturitas , 70 ( 2 ): 141 – 5 . Google Scholar Crossref Search ADS PubMed WorldCat Shah , M. and Steinberg , B. M. ( 2017 ). ‘ Drought of opportunities: contemporaneous and long-term impacts of rainfall shocks on human capital ’, Journal of Political Economy , 125 ( 2 ): 527 – 61 . Google Scholar Crossref Search ADS WorldCat Shettles , L. B. ( 1961 ). ‘ Conception and birth sex ratios ’, Obstetrics & Gynecology , 18 ( 1 ): 122 – 30 . OpenURL Placeholder Text WorldCat Singh A. ( 2015 ) ‘ Private school effects in urban and rural India: panel estimates at primary and secondary school ages ’, Journal of Development Economics , 113 : 16 – 32 . Google Scholar Crossref Search ADS WorldCat Stifel D. , Minten B. ( 2017 ) ‘ Market access, welfare, and nutrition: evidence from Ethiopia ’, World Development , 90 : 229 – 41 . Google Scholar Crossref Search ADS WorldCat Umana-Aponte , M. ( 2011 ). Long-term effects of a nutritional shock: The 1980 famine of Karamoja, Uganda. Centre for Market and Public Organisation, University of Bristol, Working Paper, 11/258 . Valente C. ( 2015 ) ‘ Civil conflict, gender-specific fetal loss, and selection: a new test of the Trivers–Willard hypothesis ’, Journal of Health Economics , 39 : 31 – 50 . Google Scholar Crossref Search ADS PubMed WorldCat Van Ewijk , R. ( 2011 ). ‘ Long-term health effects on the next generation of Ramadan fasting during pregnancy ’, Journal of Health Economics , 30 ( 6 ): 1246 – 60 . Google Scholar Crossref Search ADS PubMed WorldCat WFP ( 2013 ). Your Tax Dollars Help Starve Children . https://www.wfp.org/stories/10-things-you-need-know-about-hunger-2013. [Online; accessed 10-December-2016]. Yi , J. , Heckman , J. J., Zhang , J., and Conti , G. ( 2015 ). ‘ Early health shocks, intra-household resource allocation and child outcomes ’, The Economic Journal , 125 ( 588 ): F347 – 71 . Google Scholar Crossref Search ADS PubMed WorldCat © The authors 2020. Published by Oxford University Press on behalf of the Institute of Mathematics and its Applications. All rights reserved. This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/open_access/funder_policies/chorus/standard_publication_model)

Journal

Journal of African EconomiesOxford University Press

Published: Mar 1, 2020

There are no references for this article.