1. Introduction
This study explores the capacity of foreign aid to protect children from the devastating consequences of stunting, or impaired growth. Stunting refers to being too short for one’s age due to chronic malnutrition and repeated infections during early life. Stunted children are unlikely to reach their potential height and their brains are unlikely to reach their full capacity. They therefore perform less well in school and fall sick more often than other children do. As adults, they tend to display lower levels of human capital and productive capacity, and more often suffer from chronic diseases (WHO, 2015; UNICEF/WHO/World Bank Group, 2021). Stunting is thus a strong marker of general child health and child development, with important long-term consequences for the individual, and, in countries with high prevalence, for society at large.
Globally, close to 150 million children under age 5 suffered from stunting in 2020. This represented a 27 percent reduction compared with the year 2000 (UNICEF/WHO/World Bank Group, 2021). However, these figures are from before the Covid-19 pandemic and the Russian invasion of Ukraine, both of which have resulted in soaring energy prices and global shortages of grain and fertilizer. In addition, weather related shocks due to global warming increasingly disrupt food production. Thus, after decades of progress in fighting malnutrition and food insecurity, manifested by the “Zero Hunger” sustainable development goal (SDG2) in Agenda 2030 (UN, 2015), there has been a serious setback. Against this backdrop, the UN now warns of a global food crisis that could last for years and “tip tens of millions of people over the edge into food insecurity” (UN, 2022), and headlines speak of “the coming food catastrophe” (The Economist, 2022).
In view of these developments, investigating the capacity of foreign aid to prevent stunting in poor countries is key. This study focuses on Malawi, which is the country with the most complete geo-referenced record of aid projects from a broad range of donors, allowing for detailed and disaggregated analysis of local aid flows. Malawi is also one of the countries in the world with the highest prevalence of stunting (more on this in Section 3).
First, we ask whether aid projects, defined in a broad sense, help to reduce stunting in the local area. Next, we disaggregate the overall aid treatment and investigate when the potential treatment effects kick in and for how long they last, and how they vary depending on treatment intensity, type of project, and donor. Finally, we evaluate the robustness and heterogeneity of the estimated effects and explore possible mechanisms underlying the results.
To address these questions, we geographically match spatial data on aid projects in Malawi spanning the period 1998–2016, with respondents from four waves of the Demographic and Health Survey (DHS), conducted 2000–2016. Our benchmark sample consists of 26,604 children under the age of 5, matched with 778 aid project sites of 22 different donors.
Drawing on the anthropometric DHS data, we compute sex- and age-standardized height-for-age z-scores (HAZ scores), giving the number of standard deviations (SD) by which the observed child’s height-for-age differs from the mean of a child of the same age in a reference population. We use three outcome variables: 1) stunting (HAZ<=-2), 2) extreme stunting (HAZ<=-3), and 3) the continuous HAZ score. The latter is important since studies have shown that associations between growth faltering and risk of death or poor cognitive outcomes exist along a HAZ continuum, without a notable inflection point at –2 SD (Perumal et al., 2018). Furthermore, over dispersion, or measurement error, would render comparisons of stunting rates based on specific cutoffs less reliable (Ghosh et al., 2020).
We want our main explanatory variable to capture aid exposure, or treatment, during a critical period during the child’s early life. In the benchmark specification, we classify children as treated if they were born the same year as, or up to 3 years after, the start of a project located within 10 km of the survey cluster. In further estimations, we break down our treatment variable into multiple indicators depending on the number of years from project start to the birth of the child, the number of projects meeting the treatment criterion within the cutoff distance, the sectoral division of projects, and, finally, the donor in focus.
To identify the effect of aid, we rely on spatial and temporal variation in aid project coverage and survey rollout, coupled with variation in the year of birth of the child in relation to project start. In the main analysis, we start from the full sample, and then narrow down the control group in steps to ensure comparability with our treatment group. In the full sample estimations, we compare treated and untreated children within districts and within 55×55 km grid cells. In a next step, we restrict the sample to include only “ever-treated” clusters, consisting of survey clusters with a past, present, or future aid project within 10 km at the time of the survey, thus comparing only children living in areas that donors and the government have, at some point, deemed suitable for aid project localization. Finally, we restrict the sample to children born 0–3 years after project start (treated) and children born 2–4 years prior to project start (untreated) in ever-treated clusters.
In additional estimations we rely on sub-samples with variation in treatment status within clusters and across siblings within households, including cluster-by-year fixed effects and mother fixed effects, respectively. Furthermore, to ensure that treated and un-treated children are balanced on key covariates, we use coarsened exact matching (CEM) and run estimations based on a matched sample.
The empirical results consistently indicate a positive impact of early life aid exposure on child growth. The more we narrow down the comparison group to account for unobserved variation across time and space, the more pronounced the estimated treatment effect generally becomes. Treated children are around 2 percentage points less likely to be stunted in the least restrictive specification, compared with around 6 percentage points less likely to be stunted (4 in the case of severe stunting) in the most restrictive specification. Considering the continuous HAZ score, the corresponding effect sizes range from around 4 to 16 percent of a standard deviation. While no complete game changer, these effects are clearly not negligible.
As expected, there is significant treatment effect heterogeneity. First, we note that the positive treatment effects of aid projects on child growth materialize already for children born in the early project implementation phase, but do not remain for children born 4–5 years after project start. For children born within the treatment window, however, aid may help to protect against irreversible consequences of stunting that would otherwise have lasted a lifetime. With respect to treatment intensity, the results suggest no simple linear effect of the number of projects on our outcome variables of interest, but they nonetheless indicate that living near three or four projects fitting the treatment criteria has a stronger effect than living near one project. In terms of sectoral focus, we observe positive treatment effects for projects in the areas of rural development, infrastructure, vulnerability, and education, but somewhat surprisingly not for projects focusing on health, agriculture, and water and sanitation projects. Considering donor heterogeneity in the results, the treatment effects seem to be driven primarily by multilateral aid. This even though the bilateral donors – based on a key word search in the project activity descriptions provided by AidData – to a greater extent focus on more proximate determinants of stunting.
Our study contributes to the literature on the relationship between aid and health outcomes. To our knowledge, it is the first to use broad-based geocoded multi-donor aid data allowing for disaggregated analysis of the local effects of aid on impaired child growth in Africa.
Several earlier studies analyze the relationship between aid and various health outcomes (e.g., infant mortality, maternal mortality, life expectancy) at the country level. Some report a positive effect of aid (Arndt et al., 2015; Chauvet et al., 2013; Feeney and Ouattara, 2013; Gormanee et al., 2005; Gyimah-Brempong, 2015; Mishra and Newhouse, 2009; Pickbourn and Ndikumana, 2019; Taylor et al., 2013; Yogo and Mallaye, 2015), some find no relationship (Kizhakethalackal et al., 2013; Kosack and Tobin, 2006; Mukherjee and Kizhakkethalackal, 2013; Williamson, 2008; Wilson, 2011), and yet others find that the relationship depends on policy environment (Farag et al., 2013; Fielding, 2011). Two recent country-level studies estimate the impact of nutrition-related aid and agricultural aid on stunting. Khalid et al. (2019) find that interventions addressing immediate determinants of fetal and child nutrition reduce stunting, whereas no such treatment effects are observed for interventions influencing the underlying determinants of nutrition (such as water, sanitation and schooling). Mary et al. (2020) find moderate treatment effects of agricultural aid and larger effects of food aid.
While useful for uncovering broad patterns, the macro literature on aid effectiveness faces important challenges. First, it is difficult to establish causality. Receiving aid is associated with a multitude of country characteristics – known and unknown – that will tend to influence the estimates when seeking to establish the causal impact of aid (see, e.g., Bräutigam and Knack, 2004). Second, it is common to aggregate over aid flows that are provided for different purposes and thus should have different effects (see the discussion in Clemens et al., 2012 and Bourguignon and Gunning, 2016). Furthermore, the cross-country literature is not able to account for heterogeneity within countries. Many development projects target local development, arguably suggesting that they should be judged against location-specific outcomes (Findley et al., 2011). While (specific forms of) aid may have effects in targeted areas, these effects may not be sufficiently large to be measurable at country level or they may be obscured by omitted variable bias (Dreher and Lohmann, 2015). Against this background, we arguably need a finer lens when studying the effect of aid on child health outcomes.
At the micro level, there is a large literature evaluating the impact of specific interventions in a broad range of different areas on the nutritional status of children, with mixed findings. These include projects on nutritional supplements, feeding and/or behavioral change (e.g., Attanasio et al., 2014; Das et al., 2019; Attanasio et al., 2022a), conditional cash transfers (e.g., Cahyadi et al., 2020), nutrition-sensitive agriculture (for a review see Sharma et al., 2021) and antenatal care, water and sanitation and prevention and treatment of infectious diseases (for a recent overview see Vaivada et al., 2022).
Unlike impact evaluations, which focus on establishing the causal impact of specific interventions, we investigate the average impact of broad-based aid and aid broken down by sector and donor type. As illustrated by the so called ‘micro-macro paradox’ (Mosley, 1987), impacts of individual projects do not necessarily hold at a more aggregate level because of expenditure switching within the public sector, indirect effects on the private sector, or binding constraints (Rodrik, 2010). Sub-national analysis of geocoded aid and outcome data provides an intermediate perspective that can help bridge the micro-macro divide. Specifically, rather than estimating country-wide impacts of total aid, or analyzing the impact of specific interventions, it enables us to systematically estimate whether a multitude of aid projects have effects in the targeted areas on average, as well as to break down the analysis by donors and type of projects.
Our study contributes to the emerging literature evaluating sub-national effects of aid using geocoded aid and outcome data (e.g., Brazys et al., 2017; Civelli, et al., 2018; Isaksson and Kotsadam, 2018a,b; Dreher et al., 2019; Isaksson 2020; Isaksson and Durevall 2022). A few studies in this strand of literature focus on health outcomes. Odokonyero et al. (2018) find that health aid reduces the number of reported sick days of people living close to aid projects in Uganda. Two studies focus on Malawi. De and Becker (2015) find that health aid reduces workdays lost to illness and that water aid reduces the incidence of diarrhea. Marty et al. (2017) find that aid focusing on health infrastructure and parasitic disease control reduces malaria prevalence and improves self-reported healthcare quality. The above studies have in common that they primarily focus on health outcomes among adults.
To date, the literature evaluating sub-national effects of aid using geocoded aid and outcome data has seen relatively few attempts to explore the effects of aid on child health. Three papers find that aid helps to reduce infant mortality in the local area (Kotsadam et al., 2018, focusing on foreign aid to Nigeria; Wayoro and Ndikumana, 2020, focusing on World Bank aid to the Ivory Coast; and Widmer and Zurlinden, 2021, focusing on World Bank aid in a multi-country African sample). Rustad et al. (2020) study wasting, i.e., children being too thin for their height. Their results, based on a sample consisting of respondents from 16 African countries, suggest that aid helps reduce weight loss due to drought, but has little effect during normal meteorological conditions. Although both wasting and stunting are forms of malnutrition, they capture different conditions. Wasting is a marker of acute undernutrition, often indicating recent and severe weight loss. Stunting, on the other hand, captures linear growth faltering resulting from chronic or recurrent undernutrition due to inadequate dietary intakes and disease related nutrient loss (Wright et al., 2021). 1Hence, unlike Rustad et al., who consider whether aid helps to mitigate acute weight loss due to shocks, we investigate whether aid can help prevent the largely irreversible consequences of impaired child growth due to prolonged poor dietary and health conditions. 1Indeed, recent evidence suggests that most stunted children have never suffered wasting, and thus that the two conditions may have different causes (Wright et al., 2021). In line with this, Ngwira et al. (2017) find no association between wasting and stunting among Malawian children surveyed in 2010.
Source: AidData