Policy Research Working Paper 9273 Covariate Shocks and Child Undernutrition A Review of Evidence from Low- and Middle-Income Countries Zelalem Yilma Debebe Dhushyanth Raju South Asia Region Office of the Chief Economist June 2020 Policy Research Working Paper 9273 Abstract Unexpected adverse events that affect areas or populations countries, the pathways through which the effect operates, widely (covariate shocks) can have major consequences for and the relationship between the timing of a child’s expo- the welfare of a society. Although the negative effects on sure to a covariate shock and the effect on child nutrition households, especially among the poor, are well established status. The paper also examines whether public interven- in the economics literature, fewer studies have focused on tions can help to mitigate any negative effect and whether how natural, economic, and social covariate shocks affect the effect of covariate shocks can persist in the long term. individual welfare and particularly child nutrition status. Based on findings from the synthesis of evidence, the paper This paper reviews the evidence on the effect of covariate presents considerations and options for public policy and shocks on child nutrition status in low- and middle-income future research. This paper is a product of the Office of the Chief Economist, South Asia Region. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at zdebebe@worldbank.org and draju2@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Covariate Shocks and Child Undernutrition: A Review of Evidence from Low- and Middle-Income Countries Zelalem Yilma Debebe Dhushyanth Raju JEL codes: O15, I15, I12 Keywords: Covariate shocks, child nutrition, developing countries Debebe, World Bank, zdebebe@worldbank.org; Raju, World Bank, draju2@worldbank.org. The authors thank Rekha Menon, Martin Rama, and Hans Timmer for their support. The authors gratefully acknowledge funding by the United Kingdom Department for International Development and the European Commission, through the South Asia Food and Nutrition Security Initiative (grant number TF0A5366) administered by the World Bank. I. Introduction A super typhoon in the Philippines. Drought in India. A financial crisis sweeping East Asia. Unexpected adverse events that affect areas or populations widely (covariate shocks) have been shown to have a negative effect on the welfare of households, especially among the poor. Studies often examine the effect of shocks on welfare by assessing the extent to which they affect the ability of households to maintain their normal consumption level, often referred to as a test of consumption insurance (Baez et al. 2015; Kazianga and Udry 2006; Dercon 2004). Although findings differ across studies and shock types, the consensus is that households are only partially able to insure their consumption. The evidence also suggests that the poor, where undernutrition is primarily concentrated, are less able to guard their consumption against shocks (Jalan and Ravallion 1999). This is primarily because the poor typically do not have access to credit and liquidity options that are available to wealthier households during periods of crisis, such as the ability to tap into savings or to borrow from a bank. And common coping responses for the poor, such as turning to relatives for assistance, become unfeasible when shocks hit entire social networks (Morduch 1995; Townsend 1995). Climate change is expected to increase the frequency and severity of shocks, posing a threat to food security and population health around the globe. Even as changes to weather patterns are already occurring, rising levels of greenhouse gases are projected to further increase the variability in rainfall, contributing to both water shortages and drought, and severe rainfall and flooding. Temperatures are also expected to rise and the frequency of heat waves to increase. These weather- related shocks are likely to negatively impact the nutrition status of individuals by affecting both direct determinants of nutrition (inadequate dietary intake and disease) and underlying factors (e.g., poor socioeconomic status and environmental conditions). For example, extreme weather events could disrupt various parts of the food supply chain. Water and temperature stress may reduce crop production and caloric availability. Weather-related shocks may damage crop storage facilities and lead to contamination and spoilage, including through aflatoxins, poisonous carcinogens produced by mold. And extreme weather could hinder transport and distribution of food from areas of surplus to areas of shortage, impacting availability and prices. At the same time, weather-related shocks could alter the disease environment and increase infectious diseases (including diarrheal diseases), which can decrease nutrient absorption and increase nutrition needs (Fanzo et al. 2017). The consequence of these accumulated factors could be a worsening of nutrition outcomes, especially for young children, the focus of this review. Understanding the effects of covariate shocks on child nutrition status and designing appropriate public policy responses are vital given that transitory shocks could have lasting effects. While most studies examine whether households can guard their consumption against shocks, a number look at the effect of these shocks on child nutrition status and adult stature. Focus on this topic comes amid mounting evidence linking child nutrition status to adult outcomes including height, education attainment, income and assets, and offspring birth weight (Victora et al. 2008). Evidence of the effect of early-life shocks on adult height has been documented in India (Ho 2015), Ethiopia (Dercon and Porter 2014), and Indonesia (Maccini and Yang 2009), among other developing countries. In Zimbabwe, Alderman et al. (2006) find early-life exposure to the 1982–84 drought and civil war results in significant reduction in height-for-age persisting into adolescence, leading 2 to a permanent loss of stature of 2.3 cm on average. Other long-term negative effects include a delay in starting school by 3.7 months and a decline in completed years of schooling by 0.4 years. According to the authors, these effects altogether result in an average reduction in lifetime earnings of 7 percent. Being in a vulnerable age group at the time of a shock can affect a child’s lifetime welfare whether or not the household's income recovers from the shock. This paper reviews 29 studies and summarizes their findings on the effect of covariate shocks on the nutrition status of children below age five in developing countries. The paper also reviews the findings on some specific sub-questions, namely the pathways through which shocks may affect child nutrition status, the relationship between the timing of exposure to a shock and the shock’s effect on child nutrition status, the potential mitigating role of public interventions on the negative effect of shocks on child nutrition status, and on the long-run persistence of effects and potential catch-up growth. The different child nutrition indicators examined in the review comprise height-for-age z-scores (HAZ), the probability of stunting status, weight-for-age z-scores (WAZ), the probability of underweight status, weight-for-height z-scores (WHZ), and the probability of wasting status. 1 The review also examines child micronutrient deficiency, which can impair growth and development, weaken immunity, and increase the severity of illness. Covariate shocks comprise extreme events such as natural disasters, civil war, and economic crisis, as well as anomalies in rainfall and temperature compared to historical means. Given the diversity of shocks and outcome measures in the reviewed papers, we opt for a narrative review of the evidence rather than a formal meta analysis. To our knowledge, this is the first attempt to systematically review the evidence on covariate shocks and child nutrition status in developing countries. Studies that test for household-level consumption insurance have traditionally served as the basis for public policy recommendations. However, a stronger case can be made for public intervention based on evidence on the intrahousehold effects of covariate shocks, and particularly the effect on the nutrition status of children, given the high, long-term economic costs of poor child nutrition status to both the individual and society. By summarizing what has been researched and what has not, where the evidence is more consistent and where it is less so, the paper aims to provide guidance to researchers. Based on the patterns in the findings and insights gleaned from the review, the paper also aims to inform the choice and design of public interventions that can help mitigate the negative effect of shocks on child nutrition status, especially in developing countries. The remainder of the paper is organized as follows. Section II describes the literature search and results. Section III provides definitions of shocks in the reviewed studies. Section IV discusses findings on the effect of covariate shocks on child nutrition status. Section V discusses the potential 1 Stunting, wasting, and underweight statuses are defined as HAZ, WHZ, and WAZ less than 2 standard deviations (SD) below the relevant WHO Growth Standards medians, respectively. Stunting status (low height-for-age) is considered to reflect chronic undernutrition due to inadequate food intake over the long term, or repeated or chronic illness. Wasting status (low weight-for-height) is considered to reflect acute undernutrition due to a recent drop in food intake or recent acute illness. Underweight status (low weight-for-age) is a hybrid measure of height-for-age and weight-for-height, and thus considered to reflect both acute and chronic undernutrition (de Onis and Blossner 2003). 3 pathways through which shocks can affect child nutrition as well as the findings on pathways. Section VI discusses the findings on the relationship between the timing of exposure to a shock and its effect on child growth and development. Section VII discusses findings on the potential role of public policy in mitigating the negative effect of shocks on child nutrition status. Section VIII discusses findings on the long-term persistence of negative effects and on potential catch-up growth. Section IX concludes by summarizing findings and outlining implications for public policy and future research. II. Literature search and results We searched for peer-reviewed journal articles, working papers or reports on multiple websites, databases and repositories, and publication series, including Google Scholar, ScienceDirect, and JSTOR. Search words comprised “shocks” combined with, for example, “child nutrition,” “stunting,” “wasting,” “weight-for-height z-score,” or “height-for-age z-score.” We then reviewed the resulting papers and adopted a snowballing approach using their reference sections. In addition, we consulted with some of the authors of the papers to identify any additional relevant papers. We concluded the search for papers in December 2018. To reduce the effect of publication bias in our review, we included all relevant papers irrespective of publication status. However, we excluded papers that examine idiosyncratic shocks rather than covariate shocks and those that do not attempt to explicitly estimate the effect of shocks on child nutrition measures. 2 We identified 29 relevant studies, including 20 that have been published in peer-reviewed journals, three working papers, and six unpublished reports. See table 1 for the list of the studies. The 29 studies span four regions, with 10 in South Asia, six in Southeast Asia, nine in Sub-Saharan Africa, and four in Latin America. The distribution of the studies by year suggests that the topic is rather new; the median year of publication is 2012. III. Definition of shocks The reviewed studies differ in how they define a shock in their empirical analysis. Some studies construct a continuous variable for the shock, mostly involving deviations in weather (rainfall and temperature) from historical means, while others examine an extreme event (such as a drought event) and construct either a binary or continuous variable based on extent of exposure. Table 2 summarizes the shock variables across the reviewed studies, describing their main features, namely whether the shock variable is continuous or represents an extreme event and, in the case the variable reflects an extreme event, whether the extreme event is commonly thought of as such (i.e., “objective”) or if it is arbitrarily defined by the researcher (i.e., “subjective”). 2 We excluded Bengtsson (2010) and Mulmi et al. (2016). Bengtsson estimates the elasticity of weight to transitory income changes using rainfall as an instrument, but does not examine rainfall shock per se. Mulmi et al. examine the effect of exposure on different levels of Normalized Difference in Vegetation Index (NDVI), a measure of vegetation density. As such, the study essentially explains seasonality effects rather than shocks. 4 Eight of the 29 studies (or 28 percent) construct continuous shock variables. Almost all of these define shocks in terms of deviation in rainfall or temperature from the corresponding historical mean (i.e., rainfall or temperature shocks) during a certain period of a child’s life. The exception is Grace et al. (2012), who define a shock in terms of the variance of growing-season rainfall and temperature across the child’s life. While six of these eight studies construct measures for rainfall shocks only, Rabassa et al. (2014) and Grace et al. (2012) construct measures for rainfall as well as temperature shocks. Altogether, these eight studies will be our basis to gauge if a decrease in rainfall relative to the historical mean (i.e., unusually low rainfall) or an increase in temperature relative to the historical mean (i.e., unusually high temperature) worsens child nutrition status. Note that, under this definition of shocks, the expected sign of rainfall shock in child HAZ, WAZ, and WHZ regressions is positive while that of temperature shock is negative. By construction, these studies assume linear effects on child nutrition status where an increase in the deviation of rainfall (i.e., unusually high rainfall) is taken to be good. Arguably, it is only within a certain range that higher rainfall could have a positive effect on nutrition status. A large majority of the studies (21 of 29 studies, or 72 percent) construct a measure of an extreme event. In six of the 21 studies, extreme event is arbitrarily defined. Definitions include a “positive” shock; a “negative” shock; or either of the two, where positive and negative shocks are only meant to refer to the direction of the extreme event but both are adverse shocks. A positive shock is defined in several ways, including rainfall or temperature exceeding the 90th percentile of its historical distribution (Ho 2015), rainfall being 20 percent above its historical mean (Wang et al. 2009), and rainfall and temperature being one standard deviation (1 SD) above the historical mean (Skoufias and Vinha 2012). A negative shock is defined in one of the following ways: rainfall being below the 20th percentile (Mendiratta 2015), rainfall or temperature being less than the 10th percentile of the historical distribution (Ho 2015), rainfall or temperature being 1 SD below the historical mean (Skoufias and Vinha 2012), and rainfall being less than 75 percent of the historical district mean (Kumar et al. 2016). 3 One study (Jensen 2000) does not distinguish between the direction of the shock, defining the shock as rainfall being more or less than 1 SD from the historical mean. The expected sign of these sets of shocks (extreme events) in child HAZ, WAZ, and WHZ regressions is negative, irrespective of whether the shock is labeled “positive” or “negative.” The other 15 studies among the 21 construct measures that reflect extreme events that are not based on the researcher’s arbitrary choice of the threshold above or below which an event is termed extreme. These studies investigate three types of shocks: natural, economic, and social. Exposure to civil war (in Rwanda and Zimbabwe) during a child’s critical age, which we classify as a social shock, is examined in two of these studies. Another two studies examine the effect of the 1997–98 East Asian financial crisis. The other 11 studies examine different types of weather-related or other natural shocks, with flood the most commonly studied. While some in this subset of studies use a 3 Mendiratta (2015) constructs a binary variable for the shock such that it equals one if rainfall is above the 20th percentile of the historical mean, or zero if otherwise. As the author does in interpreting the coefficient, we define movement from one to zero as a negative shock. 5 binary variable to capture exposure to the extreme event, others exploit variation in extent of exposure to the extreme event. IV. Effects of shocks In this section, we discuss the evidence from reviewed studies on the effect of shocks on child nutrition status, organized by three groups of indicators: child weight-for-height z-scores (WHZ) or the probability of wasting status (which are generally viewed as reflecting acute undernutrition); child weight-for-age z-scores (WAZ) or the probability of underweight status (which are generally viewed as reflecting both acute and chronic undernutrition); and child height-for-age z-scores (HAZ) or the probability of stunting status (which are generally viewed as reflecting chronic undernutrition). We conclude the section by discussing the limited evidence on the effect of shocks on micronutrient deficiencies among children. We assess the evidence by gathering all estimates of effects from each study across shock types, country samples, and outcome measures examined. However, we do not include estimates of effects for subsamples or estimates of effects that serve as robustness checks. Child weight-for-height z-scores or wasting status The collective evidence is weak on whether shocks worsen child acute undernutrition, as measured by child WHZ or the probability of wasting status. Eight of the 29 reviewed studies provide relevant estimates. These studies together provide 16 estimates. Of these estimates, six are statistically significant in the direction that indicates worsening acute undernutrition status (see table 3 for a summary of our findings). In terms of significant effects, for example, Jensen (2000) finds that rainfall shocks in Côte d’Ivoire (whether positive or negative in nature) have a significant, positive effect on the probability of wasting for boys, by 3.8 percentage points. (The effect for girls of 3.1 percentage points was insignificant.) Jensen also finds that the full WHZ distributions for boys and girls worsen in areas affected by rainfall shocks, and that much of the worsening in the WHZ distributions occurs above the WHZ cutoff for wasting status. Thus, examining the effect on the probability of wasting would have concealed the full extent and nature of the effect on WHZ. Beaz and Santos (2007) find that Hurricane Mitch had a significant, negative effect on WHZ and a significant, positive effect on the probability of wasting among Nicaraguan children. Their difference-in-differences ordinary least squares (OLS) regression with a full set of controls yields an effect on WHZ of –0.501 SD and an effect on wasting of 7.3 percentage points. In terms of insignificant effects, Frankenberg and Thomas (2017) compare the cohort that was in utero in Indonesia during the 2004 Indian Ocean tsunami to a cohort born prior to the event, and do not find that the tsunami shock had a negative effect on WHZ. As the authors note, the weights of children in all cohorts fell after the natural disaster; thus, the selected older cohort appears to be a weak counterfactual. The authors also do not find that the 1997–98 East Asian financial crisis had a negative effect on WHZ of Indonesian children. However, they do find that households 6 adjusted their consumption behavior in ways that are consistent with insulating children from the negative effects of the economic crisis. Portner (2010) examines the effects of different types of natural shocks–frost, hurricanes, storms, heavy rains, and flooding–on WHZ of Guatemalan children. Estimating the effect of each shock in a separate regression, the author finds a significant, negative effect of frost on WHZ of 0.08 SD, but does not find the same for the other shocks. As Portner notes, the examined shocks occur frequently. Consequently, households may have taken measures to effectively mitigate and cope with these shocks. Child weight-for-age z-scores or underweight status The collective evidence is strongly consistent that shocks worsen child undernutrition as measured by WAZ or the probability of underweight status. Eight of the 29 reviewed studies provide relevant estimates. These studies provide 10 estimates in total, of which eight are statistically significant in the direction that indicates worsening undernutrition status (see table 4 for a summary of our findings). 4 Among studies that find statistically significant effects, Kumar et al. (2016) find that early-life drought exposure reduces child WAZ by nearly 0.1 SD in India. The authors also find that the probability of being underweight increases by about 1.8 percentage points if the child is exposed to drought in his or her year of birth, and the probability of being severely underweight increases by about 1.5 percentage points. Mendiratta (2015) finds that Indian boys who experience a negative rainfall shock during their first year of life on average observe a reduction in WAZ of 0.22 SD, while Indian girls see a reduction of 0.28 SD. And Ahmed (2015) finds that a decrease in rainfall by 1 SD during the South West monsoon season, which is equivalent to 130 mm of rainfall, reduces child WAZ by 0.63 SD in India. In terms of statistically insignificant effects, Block et al. (2004) examine the effect of Indonesia’s drought and financial crisis of 1997–98 using time-age-cohort decomposition analyses employing high-frequency nutrition data. They find that child WAZ remained constant throughout the crisis period despite a rapid increase in food prices and reduction in household consumption of rice, eggs, and dark, leafy vegetables. However, the authors note that the absence of effect on child WAZ was the result of an intrahousehold resource reallocation and a coping response that protected child caloric intake at the expense of their food quality. Deuchert and Felfe (2015) exploit variation in Super Typhoon Mike’s effect on housing damage in 1990 in the Philippines to estimate the effect on child WAZ and find that it was insignificant. Children whose housing was not affected were a weak counterfactual as the disaster likely also impacted the local economy, among other things. Moreover, the study notes that households that experienced housing damages were more likely to have received disaster relief, which might have masked the true effect of the shock. Child height-for-age z-scores or stunting status 4 Bengtsson (2007), not included in this review, uses rainfall variation as an instrumental variable for income changes and finds that, for children below age 10 in rural Tanzania, a decrease in income by 10 percent increases the probability of being malnourished (defined as child WAZ less than 1 SD below the international reference median) by 7 percent. 7 The collective evidence is also strongly consistent that shocks increase chronic child undernutrition, as measured by child HAZ or the probability of stunting status. Twenty-one of the 29 reviewed studies estimate the effects of shocks on either child HAZ or the probability of child stunting. These studies provide a total of 33 relevant estimates, of which 24 are statistically significant in the direction that indicates a negative effect of shocks on nutrition status (see table 5 for a summary of our findings). To reiterate, the expected effect is that child HAZ declines due to exposure to an extreme shock or due to a decline in rainfall or an increase in temperature from historical means. The other nine estimates show either a nonsignificant (7/9), significant but opposite-signed (1/9), or inconclusive (1/9) association. 5 In terms of significant effects, Mendiratta (2015) exploits change in rainfall in a district over time and finds that boys who experience an extreme negative rainfall shock during their first year of life see a decline in HAZ by 0.32 SD, while girls see a decline in HAZ by 0.37 SD. Using a different definition of shock, Rabassa et al. (2014) find that a 10-percentage point deviation of rainfall in the last two agricultural seasons is associated with a child HAZ of 0.07–0.1 SD, indicating that unusually low rainfall increases chronic child undernutrition. The authors also find that this magnitude of deviation in temperature reduces child HAZ by 0.7–0.8 SD. Portner (2010) finds that every additional heavy rain, storm, frost, and flood event experienced six months preceding the survey is associated with a decline of 0.13, 0.11, 0.09, and 0.06 SD in child HAZ, respectively. With regards to social shocks, in Rwanda, Akresh et al. (2011) find that an additional month of exposure to civil war lowers child HAZ by 0.11 SD. Using qualitative data, the authors argue that this effect is due to increased theft of livestock or crops and exposure to water- and vector-borne diseases, with internal displacement driving this exposure to disease. In terms of insignificant effects, Grace et al. (2012) examine the association of current HAZ among Kenyan children aged one to five years with the variance in growing-season rainfall and temperature across each year of the child’s life. They find that neither the variance in temperature nor the variance in rainfall is significantly associated with current child HAZ. The lack of a significant association could be a result of the way the shock variable is defined. The youngest (one-year-olds), who are the most vulnerable to shocks, would not have experienced this variability as the number of rainfall and temperature observations that are used in the calculation of the variance depends on how old the child is. Deuchert and Felfe (2015) also do not find a significant effect of Super Typhoon Mike, which hit the Philippines in 1990. Their analysis is on a sample of children whose average age at the time of the disaster is six years. As such, it might indeed be too late for the shock to affect growth trajectories for most of these children. Micronutrient deficiency 5 “Inconclusive” refers to Skoufias and Vinha’s (2012) finding of significant effects in both directions for different subsamples. While we would have preferred to summarize the estimate from a full-sample regression, the authors do not report such an estimate. 8 Attempts to protect caloric intake of children from shocks could come at the expense of quality of food, leading to micronutrient deficiency. Only two of the 29 reviewed studies examine the effect of shocks on micronutrient deficiency (anemia). In Indonesia, Block et al. (2004) find that the drought and financial crisis of 1997–98 increased the prevalence of anemia among children, although, as discussed above, child WAZ remained constant throughout the crisis period. The authors report that the prevalence of child anemia increased from 52 percent to 68 percent as households reallocated resources to buffer child caloric intake at the expense of quality of food. The results are consistent with substantial declines recorded in the consumption of eggs and dark green vegetables in Indonesia during this period. In contrast, Kumar et al. (2016) find that drought in India did not have an effect on the likelihood of children being anemic. As noted earlier, this is despite a significant effect the shock had on the probability of being underweight. V. Pathways behind effects Theory There are at least three potential pathways discussed in the literature through which shocks can affect child nutrition status. Income pathway. The first pathway relates to the loss of income induced by a shock. Extreme weather events could worsen child nutrition by reducing agricultural output, harming transportation infrastructure, disrupting distribution of food from areas of surplus to areas of shortage, increasing food prices, and constraining household budgets. This impact could be especially felt among those who were barely consuming the required amount in the pre-shock period. In the case of negative rainfall shocks, which could lead to water shortages or drought, this transmission mechanism is exacerbated by an excessive reliance on rainfed agriculture. This type of agriculture relies on rainfall for watering crops and is pervasive in developing countries, in part because of the infeasibility of implementing large-scale irrigation in some topographies. Within bounds, excess rainfall may or may not increase production depending on crop type, season, and geographical zone, but extreme positive rainfall shocks, which could result in flooding, would most likely reduce agricultural production. Because many markets tend to be less integrated in developing countries due to poor infrastructure, rainfall shocks and subsequent declines in output would likely translate into higher food prices. The adverse effect of this is clear for net consumers of food while its effect on net producers of food depends on the relative magnitude of the decline in agricultural output and increase in food prices. For consumers, credit market constraints may limit the ability of households to smooth consumption over time. Further, the covariate nature of such shocks may limit risk sharing with extended family members, friends, or neighbors. The result is that households may be forced to adopt less optimal coping responses, such as reducing the quality and quantity of food available to household members or reallocating food to productive family members and away from children. Economic, social, and natural shocks may also affect the government’s fiscal health and lead to unfavorable public policy measures, such as withdrawal of food transfers or subsidies. 9 Disease environment pathway. The second pathway relates to potential changes in the disease environment caused by shocks and to an inability to seek appropriate care due to demand and supply constraints. Specific environmental conditions are conducive to the survival and reproduction of parasitic species, and extreme weather changes may make areas where they traditionally could not survive more conducive to the spread of vector- and waterborne diseases and also contribute to an increase in their prevalence. For example, malaria and dengue fever, which are spread by mosquitoes, become more widespread with increased rainfall and humidity. Negative rainfall shocks may create stagnant water in river streams, which can become a breeding ground for vectors that spread disease. Such shocks could also increase fecal-oral transmission of gastrointestinal diseases due to poor hygiene resulting from limited water supply. Positive rainfall shocks, meanwhile, can lead to contamination of water sources and increase the prevalence of diarrhea and possibly vector-borne diseases. Positive temperature shocks may shorten pathogens’ incubation periods and increase rates of bacterial proliferation, contributing to a higher prevalence of diarrhea. Depending on how long a child received maternal antibodies through breastfeeding, acute diarrhea reduces the body’s capacity to take in and retain essential nutrients from consumed foods. Early life is a critical period for long-term child growth and development. Stress and poor nutrition of mothers induced by disasters and crises can adversely affect the health and nutrition of their children during this period. Moreover, disasters and crises may have a macro consequence, inducing a decline in public health spending, which could further constrain households’ ability to cope with an unfavorable disease environment. Some disasters and crises (for example, civil war, flooding, fiscal crisis) could directly affect access to and availability and operations of health facilities, making dire conditions worse. Childcare pathway. The third pathway relates to the effect of shocks on mother’s time use, or the time she spends caring for a child versus engaging in the labor market. One of the major determinants of child nutrition status is the quality of care a child receives, especially during the early years. Different types of shocks may affect female labor force participation differently. Shocks that increase the demand for labor, such as excess rainfall, may pull mothers away from domestic activities and into economic activities. The resulting effect on child nutrition status would depend on the size of the negative effect of reduced childcare (e.g., less breastfeeding, missed vaccination visits) weighed against the size of the positive effect of higher household income (and hence consumption) from female labor force participation. Evidence The reviewed studies follow three approaches to gain insight into the pathways that link negative shocks to child nutrition status. The first approach entails estimating the effect of a positive and/or negative rainfall shock by introducing both current and lagged rainfall shocks in regressions (Tiwari et al. 2013; Rabassa et al. 2014; Skoufias and Vinha 2012). Current shocks are interpreted as capturing mainly the disease environment pathway, while lagged shocks are interpreted as capturing mainly the income pathway, the assumption being that there is a time lag between a rainfall shock and its income effect. 10 The second approach entails estimating the effect of shocks on time spent on domestic chores and on the probability of suffering from diseases, receiving vaccination, terminating breastfeeding, and receiving treatment conditional on illness (Mendiratta 2015; Ahmed 2015; Wang et al. 2009; Tiwari et al. 2013; Portner 2010). The third approach entails estimating certain differential effects. A subset of the studies explores whether and to what extent the effect of shocks varies by (pre-shock) measures of socioeconomic status and availability of infrastructure such as irrigation, health, and transport (Frankenberg and Thomas 2017; Ahmed 2016; Mendiratta 2015; Ahmed 2015; Shively 2017; Rabassa et al. 2014; Hoddinott and Kinsey 2001; Foster 1995; Akresh et al. 2012). Finding stronger negative effects for those who are poor before the shock would constitute evidence for the income pathway. Studies employing the first approach point to the importance of the income pathway. For example, Tiwari et al. (2013) find that a 10-percent increase in rainfall from historical norms during the most recent monsoon generates a negative disease environment effect of no more than 0.02 SD and a positive income effect of 0.17 SD in Nepal. Rabassa et al. (2014) also find for Nigeria that a 10- percent increase in rainfall in the current season from the historical mean leads to a decline in child WHZ by 0.034 SD while the same magnitude of rainfall shock in the last completed rainy season increased child WHZ by 0.087 SD. The authors’ further finding of positive association of incidence of child diarrhea in the two weeks prior to the survey with current season rainfall shock and not with lagged rainfall shock provides additional support for this interpretation. In a similar line of thinking, Skoufias and Vinha (2012) interpret the negative effect of an extreme positive rainfall shock lagged by one period and the positive effect of an extreme negative rainfall shock lagged by one period on child HAZ in Guatemala as effects of the disease environment pathway. They also estimate the effect of shocks lagged by two periods, which they interpret as capturing mainly the income pathway, but do not find a clear pattern. The similarity of the approach to the previous two studies is in that the more recent shock is interpreted as the disease environment effect but the assumption, in Skoufias and Vinha (2012), that even shocks lagged by one period pick up the disease environment pathway appears less convincing. Findings from the few existing studies that explicitly examine the effect of shocks on the pathways provide supportive evidence for the income and disease environment pathways but not for the childcare/health-behavior pathway. Most of the reviewed studies estimate reduced-form regressions and do not explicitly investigate the theoretical pathways laid out earlier. But there are at least seven outcomes examined in the few studies that do so: (1) agricultural wages, (2) food prices, (3) the probability that a child suffered from diarrhea, fever, or cough in the two weeks prior to the survey, (4) the probability of termination of breastfeeding, (5) the probability of medical treatment conditional on sickness, (6) the probability of a child being vaccinated, and (7) time spent on domestic chores including childcare. Supportive evidence is available only for the first three of these. With respect to the income pathway, in India, Ahmed (2015) finds that a 0.5 SD decrease in rainfall during a drought period reduces the average adult male daily wage rate for tilling a farm by Rs 57.5. As a consequence of reduced yield, a negative rainfall shock is also found to be associated 11 with higher prices for staple food items such as rice and wheat. As discussed earlier, the combined effect of reduced agricultural wages and higher prices is reflected in lower child nutrition status. Mendiratta (2015) also finds for India that all five of the country’s major staple crops are significantly associated with rainfall shocks, indicating that the income of agricultural households is sensitive to rainfall. The importance of the income pathway is also implicit in Kumar et al. (2016), who find that household assets decline during the harvest season immediately following a deficit monsoon season in Pakistan. Direct evidence of the role of the disease environment pathway comes from three studies that look at the association of rainfall and other weather shocks with diarrheal incidence and illness symptoms such as fever and cough (Tiwari et al. 2013; Portner 2010; Wang et al. 2009). In Nepal, Tiwari et al. (2013) examine the association of current rainfall shock with diarrhea episodes in the two weeks preceding the survey and find a significant, positive association. In a cross-country study of 19 Sub-Saharan African countries, Wang et al. (2009) also find that prolonged rainfall shock (i.e., excess rainfall for two months in a wet season) is significantly associated with higher incidence of diarrhea. The incidence of diarrhea is 2.6 percentage points higher among these children than among children who experienced zero months of excess rainfall in the wet season. 6 In contrast, Portner (2010) does not find significant association in Guatemala between any of the five types of weather shocks he examines and the probability that a child suffered from diarrhea in the two weeks prior to the survey. However, he finds fever and cough among children are more likely, especially as a result of heavy rain. An additional episode of heavy rain in the six months prior to the survey is found to increase the probability of child fever and cough by 25 and 8 percentage points, respectively. Measurement error in the timing of shocks vis-à-vis when these symptoms were collected in the survey is expected to result in attenuation bias. Considering this, the findings collectively suggest the importance of the disease environment pathway. With respect to the childcare pathway, in India, Mendiratta (2015) hypothesized that negative rainfall shocks that reduce returns on farming could lead to a reallocation of mother’s time away from agriculture and domestic chores towards market activities. Reduced time spent on domestic chores leads to less time spent on childcare, which can undermine child nutrition status. However, the study does not find a significant association of negative rainfall shock with time spent on domestic chores. Moreover, using a more direct measure of childcare–the risk of termination of breastfeeding for those who are 0–36 months old–the study does not find supporting evidence of shocks leading to poor childcare. This is despite the positive effect of the shock on the supply of mothers’ labor in market activities. Also, restricting to the sample for which interviews were conducted during the wet season, the shock is found to have no effect on the probability that a child who was ill in the days prior to the survey sought medical attention, suggesting that this pathway plays relatively little role compared to the income and disease environment pathways. This conclusion is also supported by Ahmed (2015), who does not find a significant association between rainfall anomalies and the probability that a child is vaccinated, or between rainfall anomalies and the duration that a child is nursed by the mother. This is despite the strong effect of 6 Although not introduced as a shock, an increase in average maximum temperature is also found to be associated with higher incidence of diarrhea in this study, while minimum temperature is associated with lower diarrhea rates, further suggesting climatic variations can affect the disease environment. 12 rainfall shocks on wages and food prices, as discussed earlier. To conclude, the lack of a detectable adverse effect on childcare and the relative magnitudes of the effects of lagged and current rainfall shocks presented earlier suggest that the income pathway dominates. Most findings of differential effects, such as those that show children from poor backgrounds are more vulnerable to the effects of shocks, provide support for the importance of the income pathway. In India, Mendiratta (2015) examines the effect of negative rainfall shock on HAZ and WAZ for children from households in poor and rich Indian states, as well as those from households belonging to the bottom 40 percent of the wealth index and to the upper 60 percent. The study does not find a significant effect on children in richer states, but among children in poor states, a negative rainfall shock experienced while in utero leads to a reduction in HAZ of 0.35 SD for boys and 0.42 SD for girls, and a reduction in WAZ of 0.16 SD for boys and 0.26 SD for girls. Similarly, while the study does not find an effect among children in the upper 60 percent, in the bottom 40 percent, in utero exposure to these shocks is found to reduce HAZ of boys and girls by 0.35 SD each, and WAZ of boys by 0.16 SD and WAZ of girls by 0.20 SD. Hoddinott and Kinsey (2001) also find differential effects by economic status in Zimbabwe. By stratifying the sample into two, based on pre-drought livestock holdings (above and below median holdings), the authors find that drought only affects the growth of children residing in poorer households. While children (12–24 months) from households with below-median livestock holdings experience a slowdown in growth of 2.2 cm, there is no significant effect for those from the better-off households. Similar patterns are reported by Akresh et al. (2011), who find that the effect of crop failure on girls’ HAZ is concentrated in poor households in Rwanda. Evidence on the vulnerability of lower social groups also comes from Guatemala, where Portner (2010) finds that indigenous children are most affected by the five weather shocks he examines compared to nonindigenous ethnic groups. Foster (1995) also finds for Bangladesh that children from landless households show a greater vulnerability to conditions created by flood while better-off households are partially effective in reducing the effect of flood on child weight. Although both landless and landowning households borrow to meet needs after the flood, the cost of borrowing varied across villages and economic strata, leading to differential effects. Such evidence of differential effects by pre-shock economic status suggests the importance of the income pathway. A few other studies either do not find strong differential effects along expected lines or find counterintuitive effects. For example, Ahmed (2016) does not find differential effects by land- ownership status but reports that, in the short run (when children reach on average 4 years of age), children living in uncemented houses (a proxy for low economic status) are more vulnerable to rainfall shocks compared to those living in cemented houses in Pakistan. This protective effect, however, does not carry through to the long run (when children attain on average 13 years of age). Ahmed (2015) also does not find differential effects of rainfall shocks by either wealth quintiles or landholding size in India. A limitation of such efforts to examine differential effects is that the measures of economic status may be endogenous to the shock. If better-off households have compromised their asset positions due to the shock, the use of a post-shock asset measure to examine differential effects will lead to an underestimation of true differential effects. Using pre-tsunami household consumption per capita, Frankenberg and Thomas (2017), however, also do not find that pre-tsunami economic 13 status is predictive of the magnitude of height deficit in Indonesian children two to three years after the tsunami. In fact, in a related study (Frankenberg et al. 2013), the authors report a counterintuitive result that the negative effect of the tsunami is stronger for Indonesian children from richer households (i.e., those with pre-tsunami household consumption per capita above the median). 7 Other findings of differential effects discussed in the following section also lend support to the income pathway. These findings include differential effects by availability of irrigation facilities (Ahmed 2016) and social welfare and humanitarian aid programs (Ahmed 2015; Yamano et al. 2005). In a similar fashion, Tiwari et al. (2013) do not find that source of drinking water and access to health facilities moderate the effect of rainfall shocks on child nutrition status in Nepal. Affirmative evidence would have provided support for the importance of the disease environment pathway. VI. Differential effects by shock exposure period The effect of shocks and their persistence could depend on the timing of exposure in the child’s life. While it is generally agreed that the first 1,000 days of a child’s life are critical for long-term development (Victora et al. 2008), it is less clear what particular age window is linked with the greatest vulnerability to shocks. Several of the studies reviewed here focus on a particular period of exposure to a shock (pre- pregnancy; in utero; first, second, and third trimesters; first, second, and third year after birth). Moreover, a subset of these studies estimates the effect of shocks in more than one period of exposure. We rely on patterns in the estimates of effects from these studies to discuss the relative importance of these age windows. We also discuss which period of exposure has more consistent evidence of persistent effects. For this, we rely on the five studies that examine the effect of early life shocks on adult height. There is fairly consistent evidence that shocks experienced in the first and second year of life affect child nutrition status and that this effect is persistent (see table 6 for a summary of the evidence). The evidence is less convincing for exposure in the child’s third year of life. For the effect of exposure to a shock in the first year of life, 17 of the 26 estimates gathered from the studies (or 65 percent) are statistically significant. Besides, shocks that hit in these periods of life appear to have persistent effects as illustrated by four of the six estimates of the effect on adult height. Of the 14 estimates gathered from the studies of the effect of exposure to a shock in the second year of life, 11 (79 percent) are statistically significant. In contrast, evidence is relatively weak for the effect of exposure to a shock in the third year of life. None of the four estimates on child HAZ and only one of the five estimates on adult height is statistically significant. The evidence on the effect of exposure to a shock while in utero is not as compelling as for the first two years after birth, particularly in terms of persistent effects. Overall, 10 of the 19 estimates gathered from the studies for exposure to a shock while in utero suggest adverse effects. None of the five estimates of the effects on adult height is statistically significant while all four estimates 7 Measurement error in consumption data could be behind this. 14 of the effects on child WAZ or the probability of being underweight, and six of the 10 estimates on child HAZ are statistically significant in the expected direction. The income effect of these shocks (which is expected to kick in later in the life of the child) might have been moderated by exclusive breastfeeding–one reason why the evidence is not as compelling despite the importance that is usually attached to this period of life. The result could also be due to an empirical challenge: The sample of children observed in the study may be biased if those who are hard hit by the shock do not survive. Frankenberg and Thomas (2017) provide similar reasoning for differential effects by trimesters of exposure. An unconducive environment may be more likely to stop fragile pregnancies in the first trimester than later on. The pattern emerging from zooming in on studies that estimate the effect of exposure to a shock while in utero as well as in the first, second, and third year after birth also shows more convincing evidence of post-birth vulnerability than while in utero. There are 16 pairs of estimates for the effect of a shock experienced in utero and in the first year after birth. Of these, eight (50 percent) are significant for exposure in utero while 11 (69 percent) are significant for exposure in the first year after birth. This pattern appears to be due to differences in the persistence of effects; excluding estimates for the effect of early life shocks on adult height, the same proportion of estimates are significant across these period of exposure (8/11). But even when exposure in both periods of life is found to be significant, the magnitude of effect appears to be slightly higher for exposure in the first year of life. 8 For example, in India, Mendiratta (2015) reports that the negative effect of a negative rainfall shock experienced in utero on child HAZ ranges from 0.21 to 0.23 SD depending on gender, while it is 0.32–0.37 SD for exposure in the first year. Further comparison among five pairs of estimates for exposure in utero and in the second year of life, and exposure in utero and in the third year of life shows that none of the five estimates for the effect of exposure while in utero on adult height is significant. Three of the estimates show a significant effect for exposure in the second year after birth, and one estimate does so for the third year after birth. Altogether, the evidence suggests weaker support for persistence in the effect of exposure while in utero than in the effect of exposure in the first few years after birth. The evidence is scant on the effect of exposure to a shock in the immediate pre-pregnancy period. Only two of the studies examine shocks experienced by mothers in the period preceding pregnancy and they find conflicting results. Ahmed (2016) finds a significant, positive association between higher-than-normal rainfall during the pre-pregnancy period and height-for-age of children of exposed mothers when they turn, on average, four and 13 years old. Considering the lagged income effect of rainfall shocks, this finding perhaps suggests the importance of prenatal household income and is consistent with the argument that the first 1,000 days are critical. In contrast, Maccini and Yang (2009) do not find an effect of pre-pregnancy rainfall shock on adult height. VII. Public policy and its potential moderating role 8 We cannot comment on the statistical significance of these differences as studies do not report on them. 15 The results discussed above raise the question as to whether and what public interventions can moderate the negative effect of shocks on child nutrition status. Only a few of the reviewed studies explore this question. One intervention of interest in the relevant subset of studies is education. Two studies from India find evidence that maternal education insulates child nutrition status from the negative effects of shocks. Mendiratta (2015) finds that children of mothers who have some schooling are completely protected from the effect of rainfall shocks while those whose mother has no schooling face a significant decline in HAZ (0.29–0.53 SD, depending on sex) and WAZ (0.25–0.38 SD). Ho (2016) also finds that those children whose mothers have no schooling are more likely to experience negative height effects from postnatal negative rainfall shocks. The remaining studies in the subset that explore the question fail to find confirmatory evidence. In their study for Mexico, Skoufias and Vinha (2012) report 48 estimates of the interaction of different types of shocks with an indicator for whether or not the mother completed primary school and find only 10 estimates that are significant. 9 Interacting rainfall shock variable with either the highest grade completed or with an indicator for at least primary education of the mother, Ahmed (2015) also does not find differential effects in India. In Pakistan too, Ahmed (2016) does not find differential effects of rainfall shocks by whether or not the mother is illiterate. 10 In Indonesia, Frankenberg et al. (2013) find that, regardless of whether the mother has more or less than six years of education in the pre-disaster period, there is a significant difference in height between tsunami-affected and unaffected cohorts. However, a year later, children of less educated mothers appear to catch up and later surpass older (unaffected) cohorts, while those from more-educated mothers catch up only later. 11 In a related study, Frankenberg and Thomas (2017) do not find that pre-tsunami paternal and maternal education predict the magnitude of the height deficit two to three years after the tsunami. Public interventions can increase the resilience of child nutrition status in the face of adverse shocks. In examining the effect of rainfall anomalies during the 2002 drought in India, Ahmed (2015) estimates separate regressions for communities with a preexisting social welfare program and for those without such a program during the drought period. She finds that children did not experience a reduction in WAZ in response to drought in communities with such a program. However, in communities without the program, a reduction in rainfall by 1 SD led to a 0.66 SD reduction in child WAZ. With respect to HAZ, five years after the drought, the effect on children in program communities was smaller than that for children in nonprogram communities (0.8 SD versus 1.0 SD). Moreover, after seven years, children from nonprogram communities who faced the shock were still found to have lower HAZ than those who did not face the shock, while the difference did not persist for children in program communities. 9 These significant estimates appear to be the result of spurious correlation, especially given that five of them are in the unexpected direction. 10 The author claims that this could partly be due to little variation in the measure of education status; 87 percent are illiterate. 11 This finding also holds for paternal education. The authors note that this may suggest that the negative impact of the disaster was stronger for children from richer households, relative to what they would have experienced in the absence of the tsunami, than was the case for children from poorer households. 16 In Bangladesh, Del Ninno and Lundberg (2005) also estimate whether public interventions, some of them designed to respond to the 1998 flood, have any effect on growth in child HAZ. Although the study does not directly test for the potential moderating effect of public interventions, it finds that residence in areas where there is a preexisting nutrition program is associated with higher growth in child HAZ in the post-flood period. The study also finds that among the general population, participation in ex-post relief programs does not help children grow faster, which the study argues is due to the small size of program transfers. The above studies do not control for endogenous placement of programs, such as through program targeting. For well-targeted interventions, this would imply the moderating effect of programs is underestimated. In studying the role of aid in Ethiopia, Yamano et al. (2005) address endogenous program placement using instrumental variables. They find that children in communities that receive the average amount of food aid in the sample (the equivalent of ETB 22.5 per capita) grow on average 2 cm faster in a six-month period than if no food aid were available. The authors state that food aid (in terms of free food distribution and food-for-work) compensates for about 46 percent of the negative effect of crop damage on child growth. These findings suggest that the presence of government assistance programs could partially offset the negative effects of shocks on child nutrition and growth. Irrigation facilities can help to offset the negative income effect of rainfall shocks. Ahmed (2016) studies the potential of irrigation canals to dampen the effect of rainfall shocks by interacting her rainfall shock variable with the percentage of cultivated area that is irrigated in a district. She finds that irrigation facilities for agriculture may be a successful strategy for protecting children against the negative, short-term consequences of rainfall shocks. 12 The study also finds that children are, on average, taller in districts with access to irrigation facilities as shown by the significant, positive coefficient of irrigated land ratio in height regressions. Studies that attempt to examine differential effects by availability of health and transport infrastructure either do not find an effect or are too weak methodologically to be conclusive. Rabassa et al. (2014) use information on how easy it is for women to access health facilities (i.e., whether distance is big problem or not) to study whether the interaction of this variable with current and lagged rainfall shock is significantly associated with child WHZ. While the sign of the coefficient is consistent with the interpretation that the presence of accessible health facilities moderates the role of the disease environment pathway, it is not statistically significant. Tiwari et al. (2013) also examine differential effects by self-reported access to health facilities and do not find them significant. Shively (2017) claims that health and transport infrastructure help to buffer children from the negative effects of rainfall shortfalls on nutrition status. However, the study only finds that road density and time to health facility are significantly associated with child WHZ and that their inclusion reduces the association of growing-season rainfall shock with child WHZ in Nepal. As such, the author’s conclusion is not based on a formal test of differential effects by health and transport infrastructure. 12 This study, however, finds that by age 13, this differential effect vanishes. 17 VIII. Potential catch-up growth The preponderance of the evidence suggests that transitory shocks lead to permanent height loss. Whether the negative growth effect of transitory shocks can be reversed may depend on the timing of the shock as well as on the effectiveness of public interventions. Ten of the 29 reviewed studies address this question of whether there are persistent effects or catch- up growth in one of the following three ways. The first approach estimates the effect of early life shocks on adult height. The second approach compares the effect of shocks experienced recently to those experienced some years ago. The third approach follows affected and unaffected cohorts over time and compares how the shortfall in child HAZ develops. Several studies that investigate the effect of early life shocks on adult height find persistent negative effects. In India, Ho (2015) finds that negative rainfall shocks in the first two years of life are associated with significantly increased odds of falling within the lowest height decile in adulthood, with an odds ratio of 1.44 for the first year and 1.41 for the second year. The same study finds that a positive temperature shock experienced six to nine months after birth is associated with higher odds of falling within the lowest height decile in adulthood. In Indonesia, Maccini and Yang (2009) find that females who experience higher rainfall of 20 percent in their year and location of birth attain 0.57 cm greater height in adulthood. Dercon and Porter (2014) study the long-term effect of the 1984 famine in Ethiopia and find that 20 years later, children who were 12 to 36 months old at the peak of the crisis are significantly shorter than their unaffected cohort peers, by at least 5.3 cm. Alderman et al. (2006) reach a similar conclusion studying the effects of Zimbabwe’s back-to-back drought of 1982–84 and its civil war on child HAZ and adolescent height. 13 Other studies find evidence of partial catch-up growth. In India, Ahmed (2015) explores the effect of rainfall shocks that occur during the first 17 months of a child’s life on his or her HAZ in the same year and five and seven years later. She finds that those who are exposed to plentiful rain during these early years grow up to be taller than those who are not exposed, but this positive effect declines over time, suggesting partial catch-up growth. In Pakistan, Ahmed (2016) also finds that children whose mother experience a reduction in rainfall by 1 SD during the pre-pregnancy period are 0.17 SD shorter by age four and 0.12 SD shorter by age 13, suggesting only partial catch-up growth. In contrast to these studies, three other studies suggest that there is complete catch-up growth after the effect of transitory rainfall shocks or other extreme events. In Nepal, Tiwari et al. (2013) study the effect of excess rainfall during a child’s second year of life across four cohorts (each corresponding to the number of past monsoon seasons they lived through) and find a significant association only for the two younger cohorts. Their estimates suggest that for every month elapsed after the child’s second monsoon, the positive effect of a 10-percent higher monsoon rainfall falls by 0.003 SD, and that by 37 months after the child’s second monsoon, the effect of the shock 13 Although they do not examine adult outcomes, Hoddinott and Kinsey (2001) also find that the cohort of children who were in their second year of life a year after Zimbabwe’s 1994–95 drought grow up to be 0.6 SD shorter than comparable children who were not affected by the shock. 18 becomes zero. In Indonesia, Frankenberg et al. (2013) examine the effect of the 2004 earthquake/tsunami two to five years after it occurred. They find that about two years after the tsunami, those children who were in their second trimester when the tsunami hit are 0.79 SD shorter, given age, than those born during the same season but three years earlier (the control cohort). However, this gap shrinks to 0.35 SD three years after the tsunami and by the fifth year, the affected cohort more than makes up for the initial deficits. In Nigeria, Rabassa et al. (2014) allow the effect of shocks to vary across three cohorts (ages 0– 12 months, 12–24 months, and 24–36 months), based on the number of rainy seasons children lived through. They find that shocks only in the most recent rainy season are significantly associated with child HAZ. For example, for the cohort of children aged 12–24 months, only the immediately preceding completed rainy season shock matters, but not shocks experienced two seasons before. These findings can be interpreted as evidence of catch-up growth. 14 IX. Conclusion Poor child nutrition status remains a challenge for many countries around the world. Especially in developing countries, any major exogenous shock–whether weather-related, economic, or social– could further challenge the ability of households to maintain normal levels of food consumption and to provide nutritious diets for their children. The frequency of covariate shocks is predicted to increase as climate change triggers more extreme weather events. Flood, drought, and heat waves all can disrupt food supply chains, intensify the incidence of diseases such as diarrhea, and affect labor markets, undermining child nutrition status. Because child nutrition can have an effect on adult health, education, and income, the costs do not just accrue to individuals but have potential broader, long-term ramifications for society. This paper reviews 29 studies and summarizes findings on the effect of various types of shocks on child nutrition status in developing countries. As part of the review, the paper considers findings on the pathways through which shocks may affect child nutrition status and on the relevance of timing of exposure on child nutrition status. It also summarizes findings on the potential role that public policy can play in alleviating the effect of shocks on child nutrition status, and on the persistence of effects and potential catch-up growth. By synthesizing the available evidence on these issues, this paper seeks to provide directions for future research and public policy. Our review shows that most of the studies examine the effect of shocks on indicators of chronic undernutrition (standardized height-for-age or stunting status), and a substantial majority show that shocks worsen undernutrition status. Our review also finds strongly consistent evidence that adverse shocks increase child undernutrition as measured by standardized weight-for-age or underweight status, although evidence on the effect on acute undernutrition (standardized weight- for-height or wasting status) is much less consistent. One reason that there is more consistent evidence of a negative effect on chronic undernutrition than on acute undernutrition may be because the latter measure is relatively easier to reverse, making it difficult to detect effects if not measured immediately after the shock. 14 An alternative interpretation of this finding could be that the effect of shocks depends on the timing of exposure in the child’s life. 19 Separately, evidence on the effect of shocks on micronutrient deficiency is quite limited. Given the potential links between mortality and learning impacts and iron deficiency anemia, further exploration is warranted on the effect of shocks on micronutrient deficiency in different contexts. A decline in household income appears to be the most prominent pathway through which shocks undermine child undernutrition. Direct tests of three theoretical pathways in the studies provide supportive evidence for the income pathway as well as the disease environment pathway but not for the childcare pathway. Apart from the relatively few studies that directly test for different pathways, the case for the income pathway is supported by strong evidence of differential effects, specifically involving pre-shock indicators of economic status and the protective effects of irrigation facilities and social welfare programs. The relative magnitudes of the effects of current and lagged rainfall shocks, which arguably capture the immediate disease environment effects and future harvest effects, respectively, provide further support for this case. Timing also influences if and how exposure to a shock affects child nutrition status. Evidence suggests that the effect and persistence of shocks experienced during the first and second years of life are stronger than those experienced in utero or during the child’s third year. Given the delayed income effects of weather shocks, such effects most likely show vulnerability in the immediate, post-weaning period (second year of life) than when the child is exclusively breastfeeding (first year of life). Despite evidence that children are vulnerable to shocks experienced in utero, findings are absent that such exposure has persistent, long-term effects on height. Notwithstanding the empirical challenge of identifying such effects, one possible explanation is that individuals who experience shocks in utero still have enough window of opportunity for catch-up growth. Limited evidence suggests that social welfare programs and other interventions can help counter the negative repercussions of shocks on child nutrition status. Although researchers have conducted only a few studies on this topic, findings point to the positive role played by existing public nutrition and social welfare programs, as well as by interventions such as ex-post food aid. Overall, the evidence base is limited in terms of data that can help inform public policy choices to strengthen the resilience of child nutrition status in the aftermath of shocks. More evidence supports the hypothesis that transitory shocks have a permanent effect on child growth than the hypothesis that children completely catch up in terms of growth following a shock. In the literature at large, child nutrition status is found to be positively associated with performance on cognitive tests, school attainment, and labor market outcomes later in life (Case and Paxson 2010; Grantham-McGregor et al. 2007). This evidence suggests that transitory shocks could have substantial, long-term economic costs to both individuals and society more broadly. In light of these review findings, countries should take a targeted and strategic approach to post- disaster assistance, keeping in mind that transitory shocks can disrupt prospects for the long-term development of certain population groups. Evidence suggests that both household economic status (poverty status) and timing of shock exposure in a child’s life (ages 1–2 years) can exacerbate the degree and persistence of a shock’s effect. Instead of blanketing the whole community, governments with limited resources should foremost ensure that aid reaches the neediest populations, such as poor households with infants under age two. In resource-constrained settings, 20 more specific targeting can help to maximize the impact and long-term benefits of public interventions. Additional research on how households feed their children after experiencing a shock can help governments design more effective interventions. Existing findings suggest that financially affected households may attempt to protect a child’s gross caloric intake by purchasing cheaper, less nutritious food, but this conclusion is drawn from a limited evidence base. Further research on this topic could confirm this practice and help to inform the design of public interventions that better protect the vulnerable (such as through food-for-work programs or food supplementation programs) by putting a focus not just on gross caloric intake but also on quality of intake. Countries should redress weaknesses in the food supply chain that could exacerbate the negative effect of shocks. Evidence points to two main ways that shocks undermine child nutrition status: through a decline in household income and through an increase in consumer prices. To guard against these developments, governments can help to reduce dependence on rainfed agriculture and to ensure that food continues to be distributed from areas of surplus to areas of shortage. Other potential interventions include promoting diversified cropping and introducing weather insurance. Further studies could focus on whether such strategies in fact could help to disrupt the income pathway that links shocks to poor child nutrition status. Low-cost public health interventions can play an important role in alleviating negative effects on child nutrition status. Our review highlights that diarrhea is one pathway through which extreme weather events worsen child nutrition status. In the event of flood and heavy rain, the government could distribute oral rehydration salts and launch low-cost information campaigns that promote boiling water and practicing good sanitation and hygiene to reduce the incidence of disease and poor nutrition. For greater effectiveness, authorities should consider pairing such measures with interventions that make up for lost income. None of the reviewed studies specifically examines the protective effects of such interventions, and this issue presents another promising area for future research. Countries should invest in and implement weather-monitoring and early warning systems given the growing variability in weather conditions. Climate change is already increasing the intensity of rainfall and heat, contributing to more flooding, droughts, and heat waves. Mitigating global warming and its effects will require concerted global action, but countries can also take steps individually to better prepare and protect their citizens. For example, the government could warn households about upcoming heavy rainfall or flooding and provide clear instructions on how to protect children from contracting infectious diseases. Such measures do not require huge investments, but can play a vital role in saving crops, livestock, and lives, and in helping households protect against the worst damage of shocks, including their effect on children and on child nutrition and health. 21 References Ahmed, H. 2015. “Rainfall Shocks and Children’s Well-Being in India.” Unpublished. –––. 2016. “Rainfall Shocks and Child Health in Rural Pakistan.” Unpublished. Akresh, R., P. Verwimp, and T. Bundervoet. 2011. “Crop Failure and Child Stunting in Rwanda.” Economic Development and Cultural Change 59 (4): 777–810. Alderman, H., J. Hoddinott, and B. Kinsey. 2006. “Long Term Consequences of Early Childhood Malnutrition.” Oxford Economic Papers 58 (3): 450–74. Baez, J. E., L. Lucchetti, M. E. Genoni, and M. Salazar. 2015. “Gone with the Storm: Rainfall Shocks and Household Well-Being in Guatemala.” IZA Discussion Paper No. 8792, IZA Institute of Labor Economics, Bonn, Germany. Baez, J. E., and I. V. Santos. 2007. “Children’s Vulnerability to Weather Shocks: A Natural Disaster as a Natural Experiment.” Unpublished. Bengtsson, N. 2007. “How Responsive Is Body Weight to Transitory Income Changes? Evidence from Rural Tanzania.” Journal of Development Economics 92: 53–61. Block, S. A, L. Kiess, P. Webb, S. Kosen, R. Moench-Pfanner, M. W. Bloem, and C. P. Timmer. 2004. “Macro Shocks and Micro Outcomes: Child Nutrition During Indonesia’s Crisis.” Economics and Human Biology 2 (1): 21–44. Case, A., and C. Paxson. 2008. “Stature and Status: Height, Ability, and Labor Market Outcomes.” Journal of Political Economy 116: 499–532. Cornwell, K., and B. Inder. 2015. “Child Health and Rainfall in Early Life.” The Journal of Development Studies 51 (7): 865–80. Danysh, H., R. H. Gilman, J. C. Wells, W. K. Pan, B. Zaitchik, G. Gonzalvez, M. Alvarez, and W. Checkley. 2014. “El Nino Adversely Affected Childhood Stature and Lean Mass in Northern Peru.” Climate Change Responses 1 (1): 1–10. de Onis, Mercedes, and Monika Blossner. 2003. “The World Health Organization Global Database on Child Growth and Malnutrition: Methodology and Applications.” International Journal of Epidemiology 32 (4): 518–26. Del Ninno, C., and M. Lundberg. 2005. “Treading Water: The Long-Term Impact of the 1998 Flood on Nutrition in Bangladesh.” Economics and Human Biology 3 (1): 67–96. Dercon, S. 2004. “Growth and Shocks: Evidence from Rural Ethiopia.” Journal of Development Economics 74 (2): 309–29. Dercon, S., and C. Porter. 2014. “Live Aid Revisited: Long-Term Impacts of the 1984 Ethiopian Famine on Children.” Journal of the European Economic Association 12 (4): 927–48. Deuchert, E., and C. Felfe. 2015. “The Tempest: Short- and Long-Term Consequences of a Natural Disaster for Children’s Development.” European Economic Review 80: 280–94. Fanzo, J., R. McLaren, C. Davis, and J. Choufani. 2017. “Climate Change and Variability: What Are the Risks for Nutrition, Diets and Food Systems?” IFPRI Discussion Paper 01645, International Food Policy Research Institute, Washington, DC. Foster, A. D. 1995. “Prices, Credit Markets and Child Growth in Low-Income Rural Areas.” The Economic Journal 105 (430): 551–70. Frankenberg, E., J. Friedman, N. Ingwersen, and D. Thomas. 2013. “Child Height after a Natural Disaster.” Unpublished. Frankenberg, E., and D. Thomas. 2017. “Human Capital and Shocks: Evidence on Education, Health and Nutrition.” Unpublished. 22 Grace, K., F. Davenport, C. Funk, and A. M. Lerner. 2012. “Child Malnutrition and Climate in Sub-Saharan Africa: An Analysis of Recent Trends in Kenya,” Applied Geography 35 (1): 405–13. Grantham-McGregor, S., Y. B. Cheung, S. Cueto, P. Glewwe, L. Richter, B. Strupp, and the International Child Development Steering Group. 2007. “Developmental Potential in the First 5 Years for Children in Developing Countries.” Lancet 369: 60–70. Ho, J. Y. 2015. “Early Life Environmental Exposures and Height, Hypertension and Cardiovascular Risk Factors among Older Adults in India.” Biodemography and Social Biology 61 (2): 121–46. Hoddinott, J., and B. Kinsey. 2001. “Child Growth in the Time of Drought.” Oxford Bulletin of Economics and Statistics 63 (4): 409–36. IPCC (Intergovernmental Panel on Climate Change). 2014. “Climate Change 2014: Synthesis Report. Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change.” IPCC, Geneva, Switzerland. Jalan, J., and M. Ravallion. 1999. “Are the Poor Less Well Insured? Evidence on Vulnerability to Income Risk in Rural China.” Journal of Development Economics 58 (1): 61–81. Jensen, R. 2000. “Agricultural Volatility and Investments in Children.” American Economic Review 90 (2): 399–404. Kazianga, H., and C. Udry. 2006. “Consumption Smoothing? Livestock, Insurance and Drought in Rural Burkina Faso.” Journal of Development Economics 79 (2): 413–46. Kumar, S., R. Molitor, and S. Vollmer. 2016. “Drought and Early Child Health in Rural India.” Population and Development Review 42 (1): 53–68. Levitsky, D. A., and B. J. Strupp. 1995. “Malnutrition and the Brain: Changing Concepts, Changing Concerns.” Journal of Nutrition 125 (8 suppl): 2212S–2220S. Maccini, S., and D. Yang. 2009. “Under the Weather: Health, Schooling and Economic Consequences of Early-Life Rainfall.” American Economic Review 99 (3): 1006–26. McMichael, A. J., R. E. Woodruff, and S. Hales. 2006. “Climate Change and Human Health: Present and Future Risks.” Lancet 367 (9513): 859–69. Mendiratta, V. 2015. “Impact of Rainfall Shocks on Child Health: Evidence from India.” PSE Working Papers Number 2015-33, Paris School of Economics, Paris, France. Morduch, J. 1995. “Income Smoothing and Consumption Smoothing.” Journal of Economic Perspectives 9 (3): 103–14. Mulmi, P., S. A. Block, G. E. Shively, and W. A. Masters. 2016. “Climatic Conditions and Child Height: Sex-Specific Vulnerability and the Protective Effects of Sanitation and Food Markets in Nepal.” Economics and Human Biology 23: 63–75. Portner, C. C. 2010. “Natural Hazards and Child Health.” Unpublished. Rabassa, M., E. Skoufias, and H. Jacoby. 2014. “Weather and Child Health in Rural Nigeria.” Journal of African Economics 23 (4): 464–92. Rodriguez-Llanes, J. M., S. Ranjan-Dash, O. Degomme, A. Mukhopadhyay, and D. Guha-Sapir. 2011. “Child Malnutrition and Recurrent Flooding in Rural Eastern India: A Community- Based Survey.” BMJ Open 1 (2). Shively, G. E. 2017. “Infrastructure Mitigates the Sensitivity of Child Growth to Local Agricultural and Rainfall in Nepal and Uganda.” PNAS 114 (5): 903–8. Skoufias, E., and K. Vinha. 2012. “Climate Variability and Child Height in Rural Mexico.” Economics and Human Biology 10: 54–73. 23 Tiwari, S., H. G. Jacoby, and E. Skoufias. 2013. “Monsoon Babies: Rainfall Shocks and Child Nutrition in Nepal.” Policy Research Working Paper 6395, World Bank, Washington, DC. Townsend, R. M. 1995. “Consumption Insurance: An Evaluation of Risk-Bearing Systems in Low- Income Economies.” The Journal of Economic Perspectives 9 (3): 83–102. Victora, C. G., L. Adair, C. Fall, P. C. Hallal, R. Martorell, L. Richter, and H. S. Sachdev. 2008. “Maternal and Child Undernutrition: Consequences for Adult Health and Human Capital.” Lancet 371 (9609): 340–57. Wang, L., S. Kanji, and S. Bandyopadhyay. 2009. “The Health Impact of Extreme Events in Sub- Saharan Africa.” Policy Research Working Paper 4979, World Bank, Washington, DC. WHO (World Health Organization). 2015. “Climate and Health Country Profiles-2015: A Global Overview.” Report, World Health Organization, Geneva, Switzerland. Yamano, T., H. Alderman, and L. Christiaensen. 2005. “Child Growth, Shocks and Food Aid in Rural Ethiopia.” American Journal of Agricultural Economics 87 (2): 273–88. 24 Table 1. Reviewed Studies, by Shock Type, Country, and Publication Status Study Shock type Country Publication status 1 Ahmed (2015) Rainfall India Unpublished 2 Ahmed (2016) Rainfall Pakistan Unpublished 3 Akresh et al. (2011) Civil war (1990–91) and crop Rwanda JA failure (1988–89) 4 Alderman, Hoddinott, and Kinsey Civil war and drought (1982–84) Zimbabwe JA (2006) 5 Baez and Santos (2007) Hurricane (1998) Nicaragua Unpublished 6 Block et al. (2004) Drought and financial crisis (1997– Indonesia JA 98) 7 Cornwell and Inder (2015) Rainfall Indonesia JA 8 Danysh et al. (2014) El Nino (extreme rainfall and flood, Peru JA 1997–98) 9 Del Ninno and Lundberg (2005) Severe flood (1998) Bangladesh JA 10 Dercon and Porter (2014) Famine (1984) Ethiopia JA 11 Deuchert and Felfe (2015) Natural disaster (super typhoon, Philippines JA 1990) 12 Foster (1995) Severe flood (1988) Bangladesh JA 13 Frankenberg, Friedman, Earthquake/tsunami (2004) Indonesia Unpublished Ingwersen, and Thomas (2013) 14 Frankenberg and Thomas (2017)* Financial crisis (1998) and Indonesia Unpublished earthquake/tsunami (2004) 15 Grace et al. (2012) Rainfall and temperature Kenya JA 16 Ho (2015) Rainfall and temperature India JA 17 Hoddinott and Kinsey (2001) Drought (1994–95) Zimbabwe JA 18 Jensen (2000) Rainfall Côte d’Ivoire JA 19 Kumar et al. (2016) Rainfall/drought India JA 20 Maccini and Yang (2009) Rainfall Indonesia JA 21 Mendiratta (2015) Rainfall India WP 22 Portner (2010) Natural hazards (frost, hurricane, Guatemala Unpublished storms, heavy rains, and flooding) 23 Rabassa, Skoufias, and Jacoby Rainfall and temperature Nigeria JA (2012) 24 Rodriguez-Llanes et al. (2011) Flood (2008) India JA 25 Shively (2017) Rainfall Nepal and JA Uganda 26 Skoufias and Vinha (2012) Climate variability (rainfall and Mexico JA temperature) 27 Tiwari, Jacoby, and Skoufias Rainfall Nepal WP (2013) 28 Wang et al. (2009) Rainfall and temperature SSA WP 29 Yamano et al. (2005) Crop failure (1995–96) Ethiopia JA Notes: JA = journal article, WP = working paper, SSA = Sub-Saharan Africa. * Part of the result on the effects of the tsunami/earthquake reported in Frankenberg and Thomas (2017) seems to be taken from a paper they coauthored with others (Frankenberg et al. 2013). Here, the two papers are presented separately because 1) the former study also looks at the effect of the 1997–98 financial crisis; and 2) in the case of the tsunami/earthquake shock, the former study also reports results for child WHZ in addition to child HAZ. 25 Table 2. Definition of Shocks Study Shock variable Variable features 1 Ahmed (2015) Standardized district rainfall deviation from historical Continuous mean (in specific months of the year 2002, the drought year) 2 Ahmed (2016) Standardized, district season-specific rainfall deviation Continuous from historical mean, using district crop calendars 3 Akresh et al. (2011) Exposure to civil war and crop failure (being born or alive Extreme event, during events in affected regions) objective 4 Alderman et al. (2006) Length of exposure to civil war and exposure to drought Extreme event, (1982–83 and 1983–84); children between 12–24 months objective 5 Baez and Santos (2007) Hurricane-affected municipality Extreme event, objective 6 Block et al. (2004) 1997–98 East Asian financial crisis Extreme event, objective 7 Cornwell and Inder Positive and negative rainfall deviation from month and Continuous (2015) district-specific historical mean 8 Danysh et al. (2014) El Nino year and residence in areas prone to flooding Extreme event, objective 9 Del Ninno and Flood exposure index at household level Extreme event, Lundberg (2005) objective 10 Dercon and Porter Self-reported drought during famine period (naming Sept. Extreme event, (2014) 1984–85 as worst year) and village rainfall deviation in objective* 1984–85 from 10-year mean 11 Deuchert and Felfe Housing damage caused by typhoon Extreme event, (2015) objective 12 Foster (1995) Flood year Extreme event, objective 13 Frankenberg and 1997–98 East Asian financial crisis Extreme event, Thomas (2017) objective 14 Frankenberg and 2004 tsunami/earthquake Extreme event, Thomas (2017); objective Frankenberg et al. (2013) 15 Grace et al. (2012) Variance in rainfall and temperature across a child’s life Continuous 16 Ho (2015) Positive rainfall shock (number of months that Extreme event, rainfall/temperature exceeded 90th percentile for a given subjective month and district history); negative rainfall shock (below 10th percentile) 17 Hoddinott and Kinsey Exposure to drought in 1994–95; children between 2–24 Extreme event, (2001) months old in 1995–96 objective 18 Jensen (2000) Rainfall in a geographical area being more/less than 1 SD Extreme event, above the mean (in 1986) subjective 19 Kumar et al. (2016) Drought year (district monsoon rainfall less than 75% of Extreme event, historical district mean) subjective 20 Maccini and Yang Rainfall deviation from historical mean (of birth district, Continuous (2009) birth season, and following season–not calendar years) 21 Mendiratta (2015) Rain below 20th percentile of historical distribution Extreme event, subjective 26 Table 2. Definition of Shocks Study Shock variable Variable features 22 Portner (2010) Number of natural disaster shocks experienced in the past Extreme event, six months of survey (drawn from newspapers) objective 23 Rabassa, et al. (2012) Deviations of rainfall and temperature from historical Continuous mean (of each area) 24 Rodriguez-Llanes et al. Residence in flooded village or not Extreme event, (2011) objective 25 Shively (2017) Standardized historical variation in rainfall Continuous 26 Skoufias and Vinha Positive shock when rainfall and temperature 1 SD above Extreme event, (2012) long-run mean, negative shock when 1 SD below mean subjective 27 Tiwari et al. (2013) Rainfall deviation from historical mean Continuous 28 Wang et al. (2009) Wet season rainfall 20% above 20-year mean for one Extreme event, month or more than one month (two variables) subjective 29 Yamano et al. (2005) Proportion of damaged crop plot area in a community Extreme event, objective Notes: SD = standard deviation. * Dercon and Porter’s (2014) study can also be classified as a “common deviation” from the norm as they instrument their household-level famine variable with village-level deviation of rainfall from historical mean. 27 Table 3. Summary of Evidence on the Effect of Shocks on Weight-for-Height Z-Scores and Wasting Status Study Country Shock Outcome measure Estimation method Significance and sign Remarks of effect 1 Baez and Santos Nicaragua Hurricane Mitch Wasting status DID, OLS regression, full Significant, positive (2007) (1998) WHZ set of controls Significant, negative (demographic variables, municipality fixed effects, and local public investment) 2 Frankenberg and Indonesia Indian Ocean WHZ Difference in means, Insignificant Thomas (2017) Tsunami (2004) unconditional East Asian Difference in means, Insignificant financial crisis unconditional (1997–98) 3 Jensen (2000) Côte Extreme (positive Wasting status for DID, unconditional Insignificant d'Ivoire or negative) girls rainfall Wasting status for Significant, positive Additionally, the study boys notes the effect on the distribution of WHZ. 4 Portner (2010) Guatemala Frost WHZ OLS regression Significant, negative Hurricanes controlling for sex, age, Insignificant Storms maternal/paternal Insignificant Heavy rain education, maternal Insignificant Flooding literacy, maternal age at Significant, positive This finding is contrary to child's birth, land expectations. ownership, survey year, and location FE 5 Rabassa et al. Nigeria Excess rainfall WHZ OLS regression Significant, positive Both effects are in the (2014) (positive controlling for gender, age (lagged shocks); expected direction. The deviation) FE, month of birth FE, significant, negative immediate effect captures state FE (contemporaneous the disease environment shocks) while the lagged effect captures the income effect. 6 Rodriguez-Llanes India Flood (2008) Wasting status Multivariate regression Insignificant The study compares et al. (2011) controlling for number of children in flood-affected children below age five, villages to those living in income, birth weight, non-affected villages. household size, occupation, religion, 28 Table 3. Summary of Evidence on the Effect of Shocks on Weight-for-Height Z-Scores and Wasting Status Study Country Shock Outcome measure Estimation method Significance and sign Remarks of effect caste, water source, storage for cooked food, immunization coverage, sex, and age 7 Shively (2017) Nepal Excess rainfall WHZ OLS regression Significant, positive (positive controlling for birth deviation) month, birth year, and ecological zone Uganda WHZ OLS regression Insignificant controlling for birth month, birth year, and ecological zone 8 Wang et al. (2009) Cross Extreme positive Wasting status Random effects regression Insignificant Despite the lack of country rainfall with full controls robustness of results to (SSA) (including per capita more controls, the authors income, year, climate actually interpret the zone, and country fixed results from OLS and effects) random effects regressions with fewer controls and conclude a significant positive effect. Notes: WHZ = weight-for-height z-score, DID = difference-in-differences, OLS = ordinary least squares, FE = fixed effects, SD = standard deviation, SSA = Sub-Saharan Africa. The table reports whether a significant effect (in either direction) is found in the respective study. It is possible that some of the studies where a significant effect is noted do not find a significant effect in the analyses of subsamples and analyses of different exposure periods. 29 Table 4. Summary of Evidence on the Effect of Shocks on Child Weight-for-Age Z-Scores or Underweight Status Study Country Shock Outcome measure Estimation method Significance and Remarks sign of effect 1 Ahmed (2015) India Excess rainfall WAZ OLS regression controlling for Significant, (positive age, sex, maternal height, positive deviation) education, wealth, household size 2 Block et al. (2004) Indonesia Indonesia WAZ Time-age-cohort Insignificant drought and decomposition (time effect financial crisis [crisis impact] conditional on (1997–98) age and cohort effects) 3 Deuchert and Felfe Philippines Super Typhoon WAZ OLS regression controlling for Insignificant (2015) Mike (1990) pre-shock characteristic of the child and the household (housing variables, gender, birth weight, number of siblings, parental education, wealth, etc.) 4 Foster (1995) Bangladesh Severe flood Change in age-by- Random effects IV regression Significant, The study does not (1988) sex standardized controlling for initial age, sex, negative directly test effect of weight interval between rounds flood but notes that growth in weight can be explained by variation in price of rice and disease (diarrhea) during and immediately after the 1988 flood. 5 Kumar et al. (2016) India Drought WAZ OLS controlling for year of Significant, birth, quarter of birth, age, negative Underweight status gender, birth order, mother Significant, age at birth, parental positive Severely education, year and month of Significant, underweight status interview, caste and tribal positive groups 6 Mendiratta (2015) India Extreme WAZ OLS regression controlling for Significant, The study finds positive negative rainfall birth season FE, district FE, negative coefficient regressing birth interval, birth order, WAZ on an indicator sibling size, maternal height variable which equals 1 and weight, education of if rainfall is above the 20th percentile of 30 Table 4. Summary of Evidence on the Effect of Shocks on Child Weight-for-Age Z-Scores or Underweight Status Study Country Shock Outcome measure Estimation method Significance and Remarks sign of effect parents, occupation, village historical distribution characteristics and zero otherwise. Since this cutoff is more appropriate to define a negative shock (below 20th percentile) and because the author interprets the results as such, here we define it as a negative shock. 7 Rodriguez-Llanes et India Flood (2008) Underweight status Logistic regression controlling Significant, The study compares al. (2011) for family monthly income, positive children in flood- number of children below 5 affected villages to those years of age, birth weight, living in non-affected household size, occupation, villages. religion, caste, source of drinking water, storage of cooked food, sex, age, and immunization coverage 8 Tiwari et al. (2013) Nepal Excess rainfall WAZ OLS regression with child- Significant, The study also finds (positive and mother-level positive negative coefficient for deviation) characteristics (age, birth current-period excess order, month of birth, year of rainfall, which authors birth, gender, parental interpret as capturing the occupation and education) and disease environment region- and survey-year FE effect. But the positive lagged rainfall effect dominates in absolute magnitude. Notes: WAZ = weight-for-age z score, OLS = ordinary least squares, IV = instrumental variables, FE = fixed effects. The table reports whether a significant effect (in either direction) is found in the respective study. It is possible that some of the studies where a significant effect is noted do not find a significant effect in the analyses of subsamples and analyses of different exposure periods. 31 Table 5. Summary of Evidence on the Effects of Shocks on Child Height-for-Age Z-Scores or Stunting Status Study Country Shock Outcome measure Estimation method Significance and sign Remarks of effect 1 Ahmed (2015) India Excess HAZ OLS regression Significant, positive This effect is significant rainfall controlling for age, sex, five and eight years later (positive maternal height, but not in the same year deviation) education, wealth, (perhaps because it is too household size soon to detect effect on height). 2 Ahmed (2016) Pakistan Excess HAZ OLS regression Significant, positive rainfall controlling for age, sex, (positive maternal height and deviation) education, assets, household dependency ratio 3 Akresh et al. (2011) Rwanda Civil war HAZ OLS regression Significant, negative (1990–91) controlling for sex, province FE, birth cohort FE, province-specific time trend 4 Alderman et al. Zimbabwe Drought HAZ Maternal FE (siblings Significant, negative (2006) (1982–84) difference) regressions Civil war controlling for sex, age, Significant, negative place of birth, year of initial measurement 5 Cornwell and Inder Indonesia Excess HAZ OLS regression Significant, positive Result is only for shocks (2015) rainfall controlling for age, sex, experienced at 1–3 months (positive district FE, mother's of life; result is deviation) height, household insignificant for other (≥0 values expenditure periods. only) Unusually Significant, negative Result is only for shocks low rainfall experienced during (negative gestation. In the case of deviation) unusually low rainfall, (≤0 values which is introduced as a only) continuous variable with negative values, the reported coefficient in the study is positive. In the 32 Table 5. Summary of Evidence on the Effects of Shocks on Child Height-for-Age Z-Scores or Stunting Status Study Country Shock Outcome measure Estimation method Significance and sign Remarks of effect table, we note “negative” to mean unusually low rainfall has negative effect on HAZ. Moreover, the study does not report estimates for the combined rural- urban sample, hence the findings reported here correspond to the signs where they are found significant. 6 Danysh et al. Peru 1997–98 El HAZ OLS regression Significant, negative The study shows that the (2014) Nino controlling for socio- further away after the El economic status, sex, Nino year that a child is likelihood of living in born, the lower his/her flood-prone household HAZ. The El Nino event was not purely a shock as it was forecasted six months before the onset of heavy rains and appropriate measures were taken by the government. 7 Del Nino and Bangladesh Severe flood Growth in HAZ OLS and IV random Significant, negative Authors first show the Lundberg (2005) (1998) effects (allowing for correlation between flood endogeneity of per capita and height in a post-flood consumption and period and then test their assistance received); hypothesis that flood- regressions controlling affected children would for sex, age, birth order, grow faster than those who previous illness, maternal were not affected (in order height, education, to catch up). The absence of whether spouse of differential catch-up growth household head, within the 15-month period household size, calories of the surveys is interpreted consumed, number of here as negative effect. children, community characteristics) 33 Table 5. Summary of Evidence on the Effects of Shocks on Child Height-for-Age Z-Scores or Stunting Status Study Country Shock Outcome measure Estimation method Significance and sign Remarks of effect 8 Deuchert and Felfe Philippines Super HAZ OLS regression Insignificant (2015) Typhoon controlling for pre-shock Mike (1990) characteristic of the child and the household (housing variables, gender, birth weight, number of siblings, parental education, wealth, etc.) 9 Frankenberg et al. Indonesia Indian Ocean HAZ OLS regressions Significant, negative A robustness check (2013) and tsunami controlling for sex, birth exploiting siblings (i.e., Frankenberg and (2004) order, month, year of maternal FE) in a Thomas (2017) interview subsample also shows negative effect. 10 Frankenberg and Indonesia East Asian HAZ Difference in means, Insignificant Thomas (2017) financial unconditional crisis (1997– 98) 11 Grace et al. (2012) Kenya Variance in HAZ Multilevel model with Insignificant rainfall across random effects a child's life controlling for livelihood Variance in zone, age, duration of Insignificant temperature breastfeeding, recent across a illness, birth weight, an child's life indicator if twin, sex, maternal height, water source, maternal education 12 Hoddinott and Zimbabwe Drought Growth rate in OLS regression Significant, negative Results are robust in Kinsey (2001) (1994–95) height controlling for initial maternal FE regression child height, sex, age at controlling for the same first observation, child-level characteristics. duration of observation, maternal height, age, and schooling, livestock 34 Table 5. Summary of Evidence on the Effects of Shocks on Child Height-for-Age Z-Scores or Stunting Status Study Country Shock Outcome measure Estimation method Significance and sign Remarks of effect holdings, soil type, village FE 13 Mendiratta (2015) India Extreme HAZ OLS regression Significant, negative The study finds positive negative controlling for birth coefficient regressing HAZ rainfall season fixed effects, on an indicator variable that district FE, birth interval, equals 1 if rainfall is above birth order, sibling size, the 20th percentile of maternal height and historical distribution and weight, parental zero otherwise. Since this education, occupation, cutoff is more appropriate village characteristics to define a negative shock (below 20th percentile) and because the author interprets the results as such, here we define it as negative shock. 14 Portner (2010) Guatemala Frost HAZ OLS regression Significant, negative Hurricanes controlling for sex, age, Insignificant Storms maternal/paternal Significant, negative Heavy rain education, maternal Significant, negative Flooding literacy, maternal age at Significant, negative child's birth, land ownership, survey year, location FE 15 Rabassa et al. Nigeria Excess HAZ OLS regression Significant, positive (2014) rainfall controlling for gender, (positive age FE, month of birth deviation) FE, state FE Excess OLS regression Significant, negative temperature controlling for gender, (positive age FE, month of birth deviation) FE, state FE 16 Rodriguez-Llanes India Flooding Stunting status Logistic regression Significant, positive The study compares et al. (2011) (2008) controlling for family children in flood-affected monthly income, number villages to those living in of children below 5 years nonaffected villages. of age, birth weight, 35 Table 5. Summary of Evidence on the Effects of Shocks on Child Height-for-Age Z-Scores or Stunting Status Study Country Shock Outcome measure Estimation method Significance and sign Remarks of effect household size, occupation, religion, caste, source of drinking water, storage of cooked food, sex, age, immunization coverage 17 Shively (2017) Nepal Excess HAZ OLS regression Significant, positive Uganda rainfall HAZ controlling for birth Insignificant (positive month, birth year, deviation) ecological zone 18 Skoufias and Vinha Mexico Extreme HAZ OLS regression Significant, negative Authors do not report (2012) positive controlling for household estimates for the whole rainfall composition, maternal sample but for different Extreme height and education, Significant, positive subsamples using one- negative child age, sex, birth period lagged and two- rainfall order, an indicator if periods lagged shocks, Extreme multiple birth, birth Insignificant which they interpret as positive weight, number of capturing the disease temperature siblings, access to water, environment and income Extreme toilet, kitchen, altitude Mixed effects effect, respectively. The negative negative HAZ effect of temperature extreme positive rainfall shock and positive effect of extreme negative rainfall shock are interpreted in the study as capturing the "disease environment" effect. For extreme negative temperature, we classify the result as mixed since the authors find significant effects in either direction for different subsamples (and because they do not report comparable estimates for the combined sample). 36 Table 5. Summary of Evidence on the Effects of Shocks on Child Height-for-Age Z-Scores or Stunting Status Study Country Shock Outcome measure Estimation method Significance and sign Remarks of effect 19 Tiwari et al. (2013) Nepal Excess HAZ OLS regression with Significant, positive rainfall child- and mother-level (positive characteristics (age, birth deviation) order, month of birth, year of birth, gender, mother/father occupation, education) and region and survey- year fixed effects (FE) 20 Wang et al. (2009) Cross Extreme Stunting status OLS regression with full Significant, positive The coefficient is, however, country (Sub- positive controls (including per insignificant in random Saharan rainfall capita income, year, effects regression. Africa) climate zone, and country fixed effects) 21 Yamano et al. Ethiopia Crop failure Growth in height IV regressions Significant, negative Authors examine the effect (2005) partly due to (instruments for food aid of crop damage, which is drought and initial height) partly due to drought and (1995–96) controlling for sex, age, partly due to idiosyncratic highest education, events (such as insect maternal age, household damage and crop disease). composition, assets, water source, elevation, population density, region FE Notes: HAZ = height-for-age z-score, OLS = ordinary least squares, IV = instrumental variables, FE = fixed effects. It is possible that some of the studies where a significant effect is noted do not find significant effect in the analyses of subsamples and analyses of different exposure periods. The table reports whether a significant effect (in either direction) is reported in the respective study. 37 Table 6. Effect of Shocks by Period of Exposure Country Shock Outcome Statistical significance of estimates by period of exposure (Yes/No) measure Pre- In utero Birth or first 2nd 3rd conception year after year year birth after after Entire Trimester birth birth period 1st 2nd 3rd 1 Ahmed (2015) India Excess rainfall HAZ No (positive deviation) 2 Ahmed (2016) Pakistan Excess rainfall HAZ Yes No No (positive deviation) 3 Alderman et al. Zimbabwe Drought (1982–84) HAZ Yes (2006) 4 Alderman et al. Zimbabwe Civil war Yes (2006) 5 Block et al. (2004) Indonesia Drought and financial HAZ Yes Yes crisis (1997–98) 6 Cornwell and Indonesia Excess rainfall HAZ No Yes Inder (2015) (positive deviation) 7 Cornwell and Indonesia Unusually low HAZ Yes No Inder (2015) rainfall (negative deviation) 8 Frankenberg and Indonesia 1997–98 East Asian HAZ No No Thomas (2017) financial crisis 9 Frankenberg and Indonesia 2004 HAZ No Yes Yes Thomas (2017) tsunami/earthquake 10 Hoddinott and Zimbabwe Drought (1994–95) HAZ Yes No Kinsey (2001) 11 Mendiratta (2015) India Extreme negative HAZ Yes Yes rainfall 12 Rabassa et al. Nigeria Excess rainfall HAZ Yes Yes No (2014) 13 Rabassa et al. Nigeria Excess temperature HAZ Yes Yes No (2014) 14 Shively (2017) Nepal Excess rainfall HAZ Yes Yes (positive deviation) 38 Table 6. Effect of Shocks by Period of Exposure Country Shock Outcome Statistical significance of estimates by period of exposure (Yes/No) measure Pre- In utero Birth or first 2nd 3rd conception year after year year birth after after Entire Trimester birth birth period 1st 2nd 3rd 15 Shively (2017) Uganda Excess rainfall HAZ No No (positive deviation) 16 Tiwari et al. Nepal Excess rainfall HAZ No Yes No (2013) (positive deviation) 1 Ahmed (2015) India Excess rainfall WAZ Yes (positive deviation) 2 Kumar et al. India Drought WAZ Yes Yes (2016) 3 Kumar et al. India Drought Underweight Yes Yes (2016) status 4 Kumar et al. India Drought Severe Yes Yes (2016) underweight status 5 Mendiratta (2015) India Extreme negative WAZ Yes Yes rainfall 6 Shively (2017) Nepal Excess rainfall WHZ Yes (positive deviation) 7 Shively (2017) Uganda Excess rainfall WHZ No (positive deviation) 8 Tiwari et al. Nepal Excess rainfall WAZ Yes Yes Yes (2013) (positive deviation) 1 Ahmed (2015) India Excess rainfall Adult height Yes (positive deviation) 2 Alderman et al. Zimbabwe Civil war and drought Adult height Yes (2006) (1982–84) 3 Dercon and Porter Ethiopia Famine (1984) Adult height No No Yes Yes (2014) 4 Ho (2015) India Negative rainfall Adult height No Yes Yes No shocks 39 Table 6. Effect of Shocks by Period of Exposure Country Shock Outcome Statistical significance of estimates by period of exposure (Yes/No) measure Pre- In utero Birth or first 2nd 3rd conception year after year year birth after after Entire Trimester birth birth period 1st 2nd 3rd 5 Ho (2015) India Positive rainfall Adult height No No No No shocks 6 Ho (2015) India Positive temperature Adult height No Yes Yes No shock 7 Maccini and Yang Indonesia Positive rainfall Adult height No No Yes No No (2009) deviation Proportion of estimates of effects found to be significant 1/2 8/16 0/1 1/1 1/1 17/26 11/14 2/10 Notes: “Yes” means that the study finds a significant effect from the shock based on exposure during the specified period. “No” means there is no detectable effect from that shock based on exposure during the specified period. Within each column, some of the studies make further distinctions. An empty cell means that the study does not address that period of life. In that case we note “Yes” if there is a significant effect for at least one of them. In the case of Rabassa et al. (2014), we note “Yes” if the study finds significant effect for at least one cohort studied. In the case of Cornwell and Inder (2015), where estimates are not provided for the full sample, we note “Yes” if the study finds an effect in either the rural or urban subsamples. Table rows are color coded to help distinguish studies and estimates by the outcome measure examined. 40