Global aid for nutrition-specific and nutrition-sensitive interventions and proportion of stunted children across low- and middle-income countries: does aid matter?

Abstract

Development assistance for health (DAH) has increased dramatically over the past two decades, and this increase has led to a debate on the benefits and perverse effects of scaling-up vs scaling back DAH, and the type of interventions DAH should support. Nutrition remains a contested category viewed as essential to achieving primary healthcare objectives but as falling outside of the direct ambit of the health system. Thus, despite the increase in DAH, it continues to remain an underfunded area and little is known about the relationship between aid for nutrition-specific and nutrition-sensitive interventions and the proportion of stunted children across low- and middle-income countries. We hypothesize that as nutrition-specific aid targets local needs of countries and is less fungible than nutrition-sensitive aid, it will contribute more to a reduction in the proportion of stunted children, with the steepest gains among countries that have the highest burden of malnutrition. We use fixed-effects regressions to examine the relationship between the proportion of stunted children and aid for nutrition interventions (specific and sensitive) to 116 low- and middle-income countries (2002–16). We construct our panel using the Creditor Reporting System, Institute of Health Metrics and Evaluation, Food and Agriculture Organization, World Health Organization and World Development indicators databases. We find a one-dollar increase in per capita nutrition-specific aid is associated with a reduction in the proportion of stunted children by 0.004 (P < 0.05). When stratified by burden of malnutrition, a one-dollar increase in per capita nutrition-specific aid to countries with the highest burden of malnutrition is associated with sharper reductions in the proportion of stunted children (0.013, P < 0.01). We also find a significant association for per capita nutrition-sensitive aid and proportion of stunted children when per capita aid for nutrition is lagged by 3 and 4 years (0.0002, P < 0.05), suggesting a long-run association between nutrition-sensitive aid and proportion of stunted children. Our findings suggest that in spite of criticisms that development assistance fails to adequately reach its intended beneficiaries, aid for nutrition has been successful at reducing the proportion of stunted children. Our findings imply a need to scale-up nutrition funding and improve targeting of aid.

Development assistance for health, nutrition, child mortality, developing countries, aid

Key Messages

• With rising concerns about the potential perverse effects of aid, we do not know whether increased aid for nutrition will yield the intended effects on health, nor do we know what types of nutrition aid (aid towards nutrition-specific or nutrition-sensitive interventions) will be most beneficial.

• We find that nutrition-specific and nutrition-sensitive aid lead to reductions in the proportion of stunted children. Furthermore, with an increase in per capita nutrition-specific aid, countries with the highest burden of malnutrition experience sharper reductions in the proportion of stunted children.

• Our findings suggest a need to scale-up aid for nutrition and improve targeting of aid. At present, aid allocations appear to benefit countries with smaller populations. Moreover, some countries that do not have the highest burden of malnutrition receive more per capita nutrition-specific aid as compared to countries with the highest burden of malnutrition.

Introduction

Child stunting, low height-for-age, is an anthropometric indicator that measures the overall wellbeing of children. Children who are two standard deviations below the median height-for-age of the child growth standards are considered to be stunted (WHO, 2014). Stunting is attributed to infections, poor nutritional intake before and after birth and poses a major development challenge (Semba et al., 2008). It leads to 3.1 million deaths annually, reduces the earning capacity and long-term cognitive potential of individuals (Shekar, 2014). Most of the stunted children are concentrated across low- and middle-income countries (35.2% and 22.4%, respectively) (UNICEF, 2018).

In order to address early childhood stunting, and improve other maternal and child health indicators, attention is recently being given to improve nutrition. The UN Secretary-General Ban Ki Moon recently noted, ‘Nutrition is both a maker and a marker of development. Improved nutrition is the platform for progress in health, education, employment, empowerment of women and the reduction of poverty and inequality, and can lay the foundation for peaceful, secure and stable societies’ (SUN, 2015). Nutrition has a dedicated Sustainable Development Goal (number 2) that aims to end hunger, achieve food security and improved nutrition and the promotion of sustainable agriculture. In addition, at least 12 of the 17 goals contain indicators that are highly relevant to nutrition (SUN, 2015). Furthermore, the World Health Assembly recently declared 2016–25 as the era of nutrition (Ruel and Alderman, 2013).

Nutrition interventions can broadly be classified into two categories: nutrition-specific and nutrition-sensitive interventions. Nutrition-specific interventions refer to interventions that address the immediate determinants of foetal and child nutrition and development. These include Vitamin A and zinc supplementation, exclusive breastfeeding, dietary diversity promotion and food fortification (IFPRI, 2016). Nutrition-sensitive interventions influence the underlying determinants of nutrition. For example, water, sanitation and hygiene; child protection; schooling; early child development; maternal mental health; agriculture and food security; health and family planning services; social safety nets; and women’s empowerment (Ruel and Alderman, 2013). Research highlights that improved nutrition is critical for improved early childhood development (Bhutta et al., 2013), reducing infectious disease burden (Fogel, 1993; Null et al., 2018) and improving treatment adherence (Kenworthy, 2017).

While, there is increased interest in emphasizing nutrition interventions, how they will be financed remains questionable. Development assistance for health (DAH) has often been used for financing health interventions and there is a robust literature on the motivations for DAH, the impact of DAH on health outcomes and arguments both in favour and against DAH. Some studies suggest that health aid is often given to countries with waning health statuses (Lee and Lim, 2014), low per capita income (Lane and Glassman, 2007), high needs (Ottersen et al., 2018) and high disease burdens (Lane and Glassman, 2007; Ravishankar et al., 2009; Grépin et al., 2018), especially levels of human immunodeficiency virus (HIV)/acquired immune deficiency syndrome (Stepping, 2016). However, others find that aid has little to do with need or health indicators (Michaud and Murray, 1994) and is driven largely by political interests (Fielding, 2011), trade motivations (Alesina and Dollar, 2000), gross domestic product (GDP) (Dollar and Levin, 2006), level of democracy, quality of institutions of recipient countries (Dollar and Levin, 2006), civil rights (Trumbull and Wall, 1994), transparency (Alesina and Weder, 2002), human rights (Lebovic and Voeten, 2009), internal conflict within countries (Dietrich, 2011), donors commercial interest, colonial history and proximity to countries (Alesina and Dollar, 2000).

While some have welcomed aid infusions into developing economies (Sachs, 2005), others have raised significant concerns about whether aid has successfully contributed to improving health in low- and middle-income countries, or, conversely, whether aid in fact has compounded and reinforced health system weaknesses through a variety of pathways (Fukuyama, 2004; Easterly, 2006; Moyo, 2009; Deaton, 2013). Critics of development assistance argue that aid does not reach the poor as it often flows to regions within low- and middle-income countries with the richest people and does not favour areas with the poorest people (Briggs, 2017). Other arguments against aid highlight that it and mainly helps to increase the size of the government (Boone, 1995) and can be diverted to benefit the elites who are linked to those in power (Yontcheva and Masud, 2005; Easterly, 2006; Deaton, 2013). Aid thereby fosters corruption, economic dependency and reduces countries from taking responsibility for providing for their citizens (Moyo, 2009). Additionally, aid may prevent countries from developing and strengthening public systems as donor funding is funnelled through the private sector thereby undermining needed improvements to institutions, public infrastructure and legal capacity of the public sector (Fukuyama, 2004). Finally, aid is not always targeted at the right priorities and there is often a mismatch between national priorities and global priorities as donor priorities may be based more on politics and fads rather than either evidence of disease burden or bottom-up demands (Ollila, 2005; Shiffman, 2006; Dionne, 2018). Dionne (2018), for instance, argues that this mismatch stems from principle-agent problems inherent in the donor–recipient relationship, which force implementing agents to face pressure from competing principals (locals and donors).

Previous research has identified mixed effects of the impact of aid on child health outcomes such as stunting, mortality and disease burdens. Some studies have found that an increase in health aid leads to a reduction in child mortality rates (Mishra and Newhouse, 2009; Mary and Gomez y Paloma, 2015), while others do not find a significant relationship (Williamson, 2008; Mukherjee and Kizhakethalackal, 2013). Studies also find that aid leads to a reduction in prevalence of diarrhoea (Pickbourn and Ndikumana, 2018), diseases such as malaria (Marty et al., 2017) and HIV (Hsiao and Emdin, 2014). However, other studies report mixed effects of aid on disease prevalence (Odokonyero et al., 2018). Studies focusing on child stunting, primarily focus on analysing the relationship between food aid and childhood stunting and most find that food aid leads to a reduction in child stunting (Yamano et al., 2005; Porter, 2010; Mary et al., 2018).

While the literature, provides evidence of the relationship between various child health outcomes and health aid or overall aid, the relationship between childhood stunting and aid remains understudied. Moreover, given the increasing emphasis on nutrition, little attention has been given to understanding if aid for nutrition-specific and -sensitive interventions can lead to improvements in child stunting. It is important to address this question as given the literature on the potential perverse effects of aid, including mistargeting of aid, we do not know whether increased aid for nutrition will yield the intended effects on health, nor do we know what types of nutrition aid (aid towards nutrition-specific or nutrition-sensitive interventions) will be most beneficial. Charting aid disbursements from the Organization for Economic Cooperation and Development’s Creditor Reporting System (OECD-CRS) database shows marked differences between aid for nutrition-specific and nutrition-sensitive interventions, between 2002 and 2016 with disbursements favouring nutrition-sensitive aid (OECD, 2018; see Figure 1). Governments spend a very small share of their budgets on nutrition-specific interventions, and although nutrition-sensitive official development assistance (ODA) has increased following the Nutrition for Growth Summit, it is difficult to estimate the total amount of funding due to data constraints (IFPRI, 2016). However, nutrition-specific interventions have a stronger evidence-base and constitute more discreet/targeted interventions. For instance, recent evidence suggests that scaling-up nutrition-specific interventions can reduce the prevalence of childhood stunting by 20%, reducing the number of children with stunted growth in the world by 33 million. It would also lead to a reduction in the under-five child mortality rate by 15%, saving ∼900 000 lives in 34 high burden countries (Bhutta et al., 2013). While a lot of lives are saved and the prevalence of childhood stunting is reduced, these statistics also mean that >80% of nutritional outcomes are due to other factors.

Figure 1

Proportion of stunted children and per capita nutrition-specific and nutrition-sensitive aid (2002–16). The figure shows trend in proportion of stunted children and nutrition-specific and nutrition-sensitive aid across the estimation sample (2002–16). Per capita aid values represent actual amounts ($s). The average values across the five imputed datasets are shown.

The present study assesses aid effectiveness in the area of nutrition by examining the effect of per capita aid for nutrition-specific and nutrition-sensitive interventions on the proportion of stunted children. We hypothesize that as nutrition-specific aid targets local needs and priorities of countries and is less fungible than nutrition-sensitive aid, it will contribute more to a reduction in the proportion of stunted children with steeper gains being recorded in countries that have the highest burden of malnutrition.

Methods

Country sample

We constructed a country-year database that covers 116 countries between 2002 and 2016. We restricted our database to low-income, lower-middle-income and upper-middle-income countries, for which at least 60% of observations were available for both per capita nutrition-specific aid and per capita nutrition-sensitive aid. We drew the income classification of countries in our sample from the World Bank (World Bank, 2017b) (see Supplementary Table A1).

Variables and data sources

We construct our database using several sources. Data on the proportion of stunted children were extracted from the Institute of Health Metrics and Evaluation (Global Burden of Disease Collaborative Network | GHDx, 2019). We extracted aid data from the OECD-CRS. While there are many different sources of aid data, we chose to work with the OECD-CRS data due to its comprehensive coverage (Grépin et al., 2012) and as we were interested in disaggregating nutrition aid into nutrition-specific aid and nutrition-sensitive aid. All the aid variables were downloaded from the OECD-CRS database between March 2019 and April 2019, in 2016 constant prices (USD millions), and reflect gross disbursements. We focus on aid disbursements (vs commitments) to capture the actual flow of funds and so our aid data series begins from 2002, as per the availability of these data (Grépin et al., 2012). Data on the per capita GDP and yield of cereals (total) were drawn from the World Development Indicators (World Bank, 2017a) and the Food and Agriculture Organization of the United Nations, respectively (FAOSTAT, 2019). We categorize countries into ‘high burden of malnutrition’ and otherwise by drawing on a recent classification used by Bhutta et al. in their paper, ‘Evidence-based interventions for improvement of maternal and child nutrition: what can be done and at what cost?’, published in The Lancet in 2013.

Our primary outcome of interest is ‘the proportion of stunted children’. It is measured as the proportion of children aged 0–59 months in a given population who fall two standard deviations below the World Health Organization 2006 height-for-age curve (Global Burden of Disease Collaborative Network | GHDx, 2019). Previous research suggests that children, who are undernourished, have higher stunting rates (Bhutta et al., 2013). Our main exposure variables are the ‘per capita nutrition-specific aid’ and ‘per capita nutrition-sensitive aid’ received by a country in a given year. We draw our classification of nutrition-specific and nutrition-sensitive aid using the CRS sector codes identified by Ickes et al. (2015) in, ‘Building a Stronger System for Tracking Nutrition-Sensitive Spending: A Methodology and Estimate of Global Spending for Nutrition-Sensitive Foreign Aid’. ‘Nutrition-specific aid’ is extracted using the CRS sector code 12240 (Ickes et al., 2015). It refers to aid which focuses on interventions that target the direct determinants of nutrition (IFPRI, 2016). Nutrition-sensitive interventions refer to assistance provided for health interventions that influence the underlying determinants of nutrition (Ruel and Alderman, 2013). Nutrition-sensitive aid is defined as the sum of ‘aid for health water and sanitation’ [CRS sector codes: aid for reproductive healthcare (13020), basic drinking water supply and basic sanitation (14030), basic healthcare (12220), health education (12261), infectious disease control (12250), health policy and administrative management (12110) and health personnel development (12281)]; ‘aid for gender empowerment’ [CRS sector code: women’s equality organizations and institutions (15170)]; and ‘aid for security and agriculture’ [CRS sector codes: agricultural policy and administrative management (31110), agricultural development (31120), agricultural inputs (31150), food crop production (31161), livestock (31163), agricultural extension (31166), agricultural education/training (31181), agricultural research (31182), agricultural services (31191), agricultural cooperatives (31194), fishing policy and administrative management (31310), fishery development (31320), rural development (43040), social welfare services (16010), food aid/food security programmes (52010) and humanitarian/emergency relief (72040)] (Ickes et al., 2015). We also control for ‘per capita total aid’, which is extracted using CRS sector code 1000, and for several confounding variable that can be associated with child stunting, per capita GDP (Bendavid and Bhattacharya, 2014) and yield of cereals (total) (FAOSTAT, 2019).

Effect modifiers

There are marked differences across our sample of countries in terms of malnutrition rates and nutrition-specific aid, with research suggesting that undernutrition can lead to high levels of stunting among children (Bhutta et al., 2013). Therefore, we test if increased aid for nutrition-specific interventions to countries with a higher burden of malnutrition can lead to a greater reduction in the proportion of stunted children. In other words if aid is targeted to match local needs, can it lead to greater gains? We also test the same effect for per capita nutrition-sensitive aid. A recent study, ‘Evidence-based interventions for improvement of maternal and child nutrition: what can be done and at what cost?’, published in The Lancet in 2013, identified 34 countries that had 94% of the burden of childhood malnutrition. These include, Guatemala, Chad, Burkina Faso, Côte d’Ivoire, Ghana, Nigeria, Cameroon, Congo, Angola, Zambia, South Africa, Mozambique, Malawi, Madagascar, Tanzania, Kenya, Mali, Niger, Sudan, Rwanda, Uganda, Ethiopia, Yemen, Egypt, Iraq, Afghanistan, Pakistan, India, Nepal, Bangladesh, Myanmar, Vietnam, Indonesia and Philippines (Bhutta et al., 2013). We draw on the classification used in this study to generate a binary variable that takes on a value of 1 if the country has a ‘high burden of malnutrition’ and 0 otherwise.

Analyses

We carry out multiple imputation in R, using the Amelia package (Honaker et al., 2011) to impute missing values for all variables included in our study sample (King et al., 2001) (see Supplementary Table A2 for percentage of missing values by variable). We report the mean, median, minimum and maximum values for our sample of countries (see Table 1; for descriptive statistics for the unimputed dataset, see Supplementary Table A3). Our primary specification is a fixed-effects model. The use of a fixed-effects model helps in removing variation due to unobserved characteristics of countries that do not change over time and also time-specific events that could influence the results of our analysis (Wooldridge, 2010). We run three main models and repeat them after lagging the per capita aid variables by 1, 2, 3 and 4 years. The first model estimates the relationship between per capita aid for nutrition-specific and nutrition-sensitive interventions and the proportion of stunted children. The second and third models test the moderating effect between countries with a ‘high burden of malnutrition’ and per capita nutrition-specific aid and per capita nutrition-sensitive aid, respectively, by an interaction term. We estimate the following models:

$${\mathrm{M}}_{i,t=}\hspace{0.17em}{\beta }_0+{\beta }_1\hspace{0.17em}(X_{i,t})+{\beta }_2(X_{i,t})+\theta \left(U_{i,t}\right)+{\alpha }_{\mathrm{i}}+{\tau }_t+{\epsilon }_{\mathrm{i},\mathrm{t}},$$

(Model I)

$${\mathrm{M}}_{i,t}={\beta }_0+{\beta }_1\left(X_{i,t}\right)+{\beta }_2\left(X_{i,t}\right)+\delta \left(X_{i,t}\right)*\left(D_{i,t}\right)+\theta \left(U_{i,t}\right)+{\alpha }_{\mathrm{i}}+{\tau }_{\mathrm{t}}+{\epsilon }_{\mathrm{i},\mathrm{t}},$$

(Model II, III)

where M_i,t is a measure of the proportion of stunted children in recipient country i in time period t, $\beta$₁ captures the coefficient for the independent variable per capita nutrition-specific aid, $\beta$₂ captures the coefficient for the independent variable per capita nutrition-sensitive aid, $\theta \mathrm{ }$is a vector of coefficients contained in the matrix U_i,t.,$\mathrm{ }\delta$ is the coefficient of the interaction term between per capita nutrition-specific aid and countries with a high burden of malnutrition in Model II and it is replaced by nutrition-sensitive aid in Model III, D_i,t is a dummy variable which takes on a value of one if the country belongs to the list of 34 countries with the highest burden of malnutrition, $\alpha$_i is a country fixed effect which captures the time-invariant characteristics of the recipient, $\tau$_t captures time-specific events common across countries and $\epsilon$_i,t is the error term. We cluster the standard errors by country. We repeat the same analysis with per capita aid variables lagged (Mishra and Newhouse, 2009) by 1, 2, 3 and 4 years to tease out the effect of time on the association between per capita aid for nutrition and proportion of stunted children. All analyses were done using STATA 14 (StataCorp, 2015).

Table 1

Descriptive statistics

	Mean	Median	Minimum	Maximum
Proportion of stunted children	0.27	0.27	0.01	0.59
Per capita aid variables
Per capita total aid	74.06	46.87	0.90	977.21
Per capita nutrition-specific aid	0.25	0.06	0	7.74
Per capita nutrition-sensitive aid	13.72	9.89	0.08	145.88
Per capita nutrition-specific aid by malnutrition levels of countries
Highest burden of malnutrition	0.25	0.08	2.65e−08	2.60
Other burden of malnutrition	0.25	0.05	0	7.74
Per capita nutrition-sensitive aid by malnutrition levels of countries
Highest burden of malnutrition	8.83	7.79	0.53	37.56
Other burden of malnutrition	15.78	10.89	0.08	145.88
Yield of cereals (total)	22 527	19 941.5	343	94 537
Gross domestic product (logged)	7.54	7.59	1.54	9.92

Notes: Per capita aid values represent actual amounts ($s). The average values across the five imputed datasets are shown.

Sensitivity analysis

We ran several sensitivity analyses. First, we re-run the regression models for the unimputed dataset. Second, we divided nutrition-sensitive aid into its subcomponents and ran regressions with these disaggregated variables. Third, we lagged the disaggregated per capita aid variables by several years across different models. Fourth, we ran the analysis using several additional controls [total fertility rate, percentage of women receiving prenatal care (Gillespie and van den Bold, 2017) and percentage of people in urban areas (Bendavid and Bhattacharya, 2014)]. Fifth, we use an alternate measure of child health, the under-five child mortality rate as our outcome variable. The under-five child mortality rate reflects the probability of a child dying by age 5 per 1000 live births (WHO, 2017). Data on annual under-five mortality rates were extracted from the Global Health Observatory (WHO, 2017). Finally, we also ran several additional models with lagged aid and under-five child mortality as our outcome variable.

Results

The proportion of stunted children ranges from 0.01 to 0.59. The average per capita nutrition-specific aid received by countries in our sample is $0.25, while the maximum is $7.74. Per capita nutrition-sensitive aid ranges from $0.008 to $145.88. The average per capita nutrition-specific aid received by the 34 countries with the highest burden of malnutrition is $0.25 with a maximum of $2.60. The average per capita nutrition-sensitive aid received by the 34 countries with the highest burden of malnutrition is $8.83 with a maximum of $37.56 (see Table 1 and Figure 2). For descriptive statistics for the unimputed dataset, see Supplementary Table A3.

Figure 2

Proportion of stunted children, per capita nutrition-specific aid and malnutrition levels for countries with high malnutrition levels. The scatterplot shows the average nutrition-specific aid and proportion of stunted children for countries across the estimation sample (2002–16). Per capita aid values represent actual amounts ($s). The average values across the five imputed datasets are shown.

In the fixed-effects regression model with the main effects (see Table 2, Model I and Figure 3), we find that increase in per capita aid for nutrition-specific interventions leads to a reduction in the proportion of stunted children. Specifically, a one-dollar increase in aid for nutrition-specific interventions is associated with a reduction in the proportion of stunted children by 0.004 (P < 0.05). We do not find a statistically significant association between per capita aid for nutrition-sensitive interventions and the proportion of stunted children. When we include an interaction term between per capita aid for nutrition-specific interventions and classification of countries by their burden of malnutrition, we find that an increase in per capita nutrition-specific aid to countries with a higher burden of malnutrition is associated with higher reductions in proportion of stunted children. A one-dollar increase in per capita nutrition-specific aid to countries with a high burden of malnutrition is more likely to be associated with a lower proportion of stunted children in these countries, 0.013 units, as compared to countries with a lower burden of malnutrition (P < 0.01) (see Table 2, Model II and Figure 3). We do not find significant effects for per capita nutrition-sensitive aid or per capita total aid to all sectors (see Table 2, Model III). However, we find that yield of cereals and per capita GDP have a significant relationship with the proportion of stunted children. An increase in GDP per capita and yield of cereals leads to a reduction in the proportion of stunted children (P < 0.05).

Figure 3

Per capita nutrition-specific aid and proportion of stunted children (2002–16), regression Models I and II. The figure shows the beta coefficients from Table 2, regression Models I and II. The circles and squares are the point estimates, while the bars represent the confidence intervals. The three aid variables are in per capita terms. Gross domestic product is logged. Year fixed effects included in the model not shown.

Table 2

Per capita nutrition-specific and nutrition-sensitive aid proportion of stunted children

	(1)	(2)	(3)
Per capita nutrition-specific aid	−0.004*	−0.001	−0.003*
	(0.002)	(0.001)	(0.001)
Per capita total aid	0.000	0.000	0.000
	(0.000)	(0.000)	(0.000)
Yield of cereals (total)	−0.000*	−0.000*	−0.000*
	(0.000)	(0.000)	(0.000)
Per capita GDP (logged)	−0.006*	−0.006*	−0.006*
	(0.003)	(0.003)	(0.003)
Per capita nutrition-sensitive aid	−0.000	−0.000	−0.000
	(0.000)	(0.000)	(0.000)
Other malnutrition countries # Per capita nutrition-specific aid		Ref
High malnutrition countries # Per capita nutrition-specific aid		−0.013**
		(0.005)
Other malnutrition countries # Per capita nutrition-sensitive aid			Ref
High malnutrition countries # Per capita nutrition-sensitive aid			−0.001
			(0.000)
Constant	0.376***	0.375***	0.377***
	(0.021)	(0.021)	(0.021)
Years	Yes	Yes	Yes
Number of id	116	116	116

Notes: Robust standard errors in parentheses; N = 1662, where *P < 0.05, **P < 0.01, ***P < 0.001. Fixed-effects model, year fixed effects included but not shown. Standard errors clustered by country. One hundred and sixteen low- and middle-income countries. Per capita nutrition-sensitive aid is the sum of per capita aid for health and water and sanitation, security and agriculture, gender empowerment and general budget support. Per capita total aid reflects aid for all sectors, per capita nutrition-specific aid is aid for basic nutrition.

In the fixed-effects regression models with lagged per capita aid variables (Mishra and Newhouse, 2009), we find that while shorter lags (1 and 2 years) lead to a significant association between per capita nutrition-specific aid and the proportion of stunted children, the effect does not hold out for per capita nutrition-specific aid when the length of the time lags is increased (see Table 3). When stratified by burden of malnutrition, for a 2-year time lag on the per capita aid variables, we find that a one-dollar increase in per capita nutrition-specific aid to countries with a high burden of malnutrition is more likely to be associated with a lower proportion of stunted children in these countries, 0.012 units, as compared to countries with a lower burden of malnutrition (P < 0.05) (see Table 3, Model 8). However, in contrast, per capita nutrition-sensitive aid shows a significant and inverse relationship with the proportion of stunted children when per capita aid is lagged by 3 and 4 years. A one-dollar increase in nutrition-sensitive aid is associated with a reduction in the proportion of stunted children by 0.0002 (P < 0.05) (see Table 3, Models 1–6; Supplementary Table A6a–d, Models 2, 5, 8).

Table 3

Fixed-effects models showing association between lagged per capita aid and proportion of stunted children

	4-year lag (aid variables)			3-year lag (aid variables)			2-year lag (aid variables)
	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)	(9)
Per capita nutrition-specific aid	−0.002	−0.002	−0.002	−0.003	−0.001	−0.003	−0.003*	−0.002	−0.003*
	(0.002)	(0.001)	(0.001)	(0.002)	(0.001)	(0.002)	(0.002)	(0.001)	(0.002)
Per capita nutrition-sensitive aid	−0.0002*	−0.0002*	−0.0002*	−0.0002*	−0.0002*	−0.0002*	−0.000	−0.000	−0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Yield of cereals (total)	−0.000*	−0.000*	−0.000*	−0.000*	−0.000*	−0.000*	−0.000*	−0.000*	−0.000*
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Per capita total aid	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000	0.000
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Per capita GDP (logged)	−0.004	−0.004	−0.004	−0.004	−0.004	−0.004	−0.004	−0.004	−0.004
	(0.003)	(0.003)	(0.003)	(0.004)	(0.004)	(0.004)	(0.003)	(0.003)	(0.003)
Other malnutrition countries # Per capita nutrition-specific aid		Ref			Ref			Ref
High malnutrition countries # Per capita nutrition-specific aid		−0.007			−0.011*			−0.012*
		(0.004)			(0.005)			(0.005)
Other malnutrition countries # Per capita nutrition-sensitive aid			Ref			Ref			Ref
High malnutrition countries # Per capita nutrition-sensitive aid			−0.000			−0.000			−0.000
			(0.000)			(0.000)			(0.000)
Constant	0.333***	0.333***	0.333***	0.346***	0.345***	0.345***	0.346***	0.346***	0.347***
	(0.025)	(0.025)	(0.025)	(0.027)	(0.027)	(0.027)	(0.024)	(0.024)	(0.024)
Years	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Observations	1198	1198	1198	1314	1314	1314	1430	1430	1430
Number of id	116	116	116	116	116	116	116	116	116

Notes: Robust standard errors in parentheses. *P < 0.05, **P < 0.01, ***P < 0.001,. Fixed-effects model, year fixed effects included but not shown. Standard errors clustered by country. One hundred and sixteen low- and middle-income countries. Per capita nutrition-sensitive aid is the sum of: per capita aid for health, water and sanitation; security and agriculture; and gender empowerment. Per capita total aid reflects aid for all sectors, per capita nutrition-specific aid is aid for basic nutrition.

Sensitivity analysis

Our results for Table 2 were consistent with our analysis of the unimputed aid data (See Supplementary Table A5, Models 2, 5, 8). In terms of the lagged per capita aid variables, the results were also largely consistent for the unimputed data with several exceptions; nutrition-sensitive aid was only significant with a 4-year lag and yield of cereals was not significant in any of the regressions (see Supplementary Table A7a–d, Models 2, 5, 8). Including additional control variables (such as fertility rate, percentage of women receiving prenatal care and percentage of population in urban areas) led to significant results on the interaction term between per capita nutrition-specific aid and high burden of malnutrition countries. All other variables were insignificant (see Supplementary Table A8). In the regressions with disaggregated nutrition-sensitive aid, none of the nutrition-sensitive variables were significant (see Supplementary Table A9). In the models with per capita lagged aid, with the exception of the nutrition-sensitive variable—health, water and sanitation—in the model with the 4-year lag, all other nutrition-sensitive variables were insignificant (see Supplementary Table A10). In regressions with under-five mortality, an alternate measure of child health, as the dependent variable, we find that per capita nutrition-specific aid has a significant and inverse relationship with child mortality across all the models. The interaction term is between per capita nutrition-specific aid and countries stratified by burden of malnutrition was also significant and had an inverse relationship with the proportion of child stunting. Additionally, we find that while aggregate nutrition-sensitive aid is not significant, one type of nutrition-sensitive aid,

aid for health, water and sanitation, had a significant and inverse relationship with under-five child mortality (see Supplementary Tables A11 and A12a–d).

Discussion

We estimated the relationship between per capita nutrition-specific and nutrition-sensitive aid, and child stunting across 116 low- and middle-income countries from 2002 to 2016. We find that a one-dollar increase in per capita nutrition-specific aid is associated with a reduction in proportion of stunted children of child stunting by 0.004. Increases in nutrition-sensitive aid had a significant association with reducing the proportion of stunted children when aid was lagged by 3 and 4 years, suggesting that association exists but over the longer term. When stratified by burden of malnutrition, we find that as increase in per capita nutrition-specific aid to countries with the highest burden of malnutrition is associated with sharper reductions in the proportion of stunted children as compared to countries with a lower burden of malnutrition. Our findings suggest that in spite of criticisms that development assistance fails to adequately reach its intended beneficiaries, per capita aid for nutrition is associated with a reduction in the proportion of stunted children across countries. Our study adds to the literature by providing empirical evidence on the relationship between aid for nutrition across different countries, suggesting that aid targeted at specific health areas may be more beneficial than general development assistance and may not suffer from the same perverse effects, especially when matched with country need or burden.

To scale up, interventions concrete sources of funding are required yet a gap exists between the required and actual nutrition needs of populations (IFPRI, 2016). Previous research discusses the benefits (Sachs, 2005) and perverse effects (Easterly, 2006; Deaton, 2013) of DAH. Our findings in terms of our main hypotheses—per capita nutrition-specific and nutrition-sensitive aid leads to reduction in the proportion of stunted children—supports literature which argues that development assistance can have positive effects on health outcomes (Bendavid and Bhattacharya, 2014). Moreover, given the significant association between child stunting and both per capita nutrition-specific aid and per capita nutrition-sensitive aid, we find support for arguments which highlight that not only food and nutrients, but disease environments and overall living conditions can influence child health outcomes (Coffey et al., 2013). Our finding that GDP per capita is associated with a reduction in the proportion of stunted children further strengths our finding that adequate food is necessary but is not sufficient for treating and preventing childhood nutrition. Consequently, food, health and care and poverty reduction strategies (Alderman et al., 2014) are required to ensure positive outcomes.

We also find evidence which suggests a need to improve targeting of aid. First, we note that some countries that do not have the highest burden of malnutrition (according to the classification we use) receive more per capita nutrition-specific aid as compared to countries with the highest burden of malnutrition. For instance, Cabo Verde on average received high amounts of per capita nutrition-specific aid ($0.87), but is not considered to have a high burden of malnutrition, whereas India, one of the countries with the highest burden of malnutrition received $0.02 per capita (see Figure 2). Therefore, it appears that need alone does not drive aid allocations. Previous literature suggests that political and strategic interests (Maizels and Nissanke, 1984), human rights violations and government effectiveness (Bandyopadhyay and Wall, 2006) vs bottom-up needs of countries can also drive aid allocations. Moreover, evidence also suggests that aid often fails to reach the poorest individuals within a country (Briggs, 2017). Second, we also note a population bias inherent in how aid is allocated (Martinsen et al., 2018). Research suggests that donors may or may not give population a consideration when targeting aid (World Bank, 2016; Martinsen et al., 2018). We note wide variation in the per capita aid received by countries across the spectrum, where per capita aid is higher for countries with smaller populations. For example, countries such as Pakistan and India, which have larger populations and a high burden of malnutrition, receive less per capita nutrition-specific aid.

Our findings compliment existing literature, which discusses the benefits of scaling-up nutrition interventions. Previous research highlights that under nutrition, is one of the major causes of poor child health outcomes (Fogel, 1993). Nutrition affects health outcomes through two main pathways. First, poor nutrition increases the chances of an individual acquiring an infection and so healthier individuals are less likely to fall sick and have better health outcomes (Null et al., 2018). Second, nutritional status of children aids their recovery from illnesses (Fogel, 1993). Recent experimental evidence confirms the relationship between increased nutrition supplementation and better health outcomes among children, even when nutrition supplementation is implemented as a stand-alone intervention. However, in contrast nutrition-sensitive interventions such as improved water and sanitation when implemented alone do not lead to improved health outcomes (Ginsburg et al., 2015; Luby et al., 2018). Some studies show that specific types of nutrition-sensitive interventions, e.g. piped water, lead to reductions in child stunting (Headey and Giordano, 2019). Therefore, our findings support such literature, which highlights the importance of nutrition interventions in improving child health outcomes such as stunting.

The results of our study imply that there is a need to scale-up funding for nutrition-specific interventions. According to a recent report, donor spending on nutrition-specific interventions is stagnant at $1 billion, and donors will need to increase their financial commitments towards nutrition to achieve improvements in health outcomes (IFPRI, 2016). Effective action will also be required by the government to commit resources (Gillespie and van den Bold, 2017), build nutrition interventions into national plans and also develop laws and policies that enhance the status of women (Horton, 2008). At present, governments commit a very small proportion of funding for nutrition interventions. The average government expenditure on nutrition in low- and middle-income economies is ∼2.1% as compared to 33% for other sectors (health, agriculture, education and social protection). Within nutrition, funding for nutrition interventions is much lower. Previous research suggests that political commitment is important for ensuring the success of nutrition interventions. For example, in Senegal the establishment of the CLM (Cellule de Lutte contre la Malnutrition or Committee for the Fight against Malnutrition) was critical for diffusing nutrition information and attracting a larger number of stakeholders (Kampman et al., 2017). Consequently, greater political commitment is required to ensure successful outcomes.

Limitations

There are several limitations of our analysis. First, we use basic nutrition (CRS code: 12240) as our main explanatory variable. This variable does not fully capture all nutrition-specific interventions. While there is recognition in the literature of the inadequacy of existing nutrition-specific data (Mutuma et al., 2012; Ickes et al., 2015), a comprehensive source of time-series data does not exist. Second, we restricted our study sample to include the years after 2002, as we had more complete data coverage for the aid data for that period. However, we still had to use multiple imputations to fill in missing values for all variables of interest. Finally, we were unable to control for nutrition-specific policies across countries due to the absence of such data, better data are required for assessing nutrition gaps and policies across countries (Bhutta, 2016).

Conclusions

We find that per capita nutrition aid leads to a reduction in the proportion of stunted children across low- and middle-income countries, with the steepest gains being experienced by countries with the highest burden of malnutrition. Our findings point to the importance of global aid in reducing the proportion of stunted children when correctly targeted to meet local needs.

Conflict of interest statement. None declared.

Ethical approval. No ethical approval was required for this study.