Outline
Our results are structured in four parts: (1) descriptive results designed to outline patterns in wasting, stunting and food inflation; (2) regression results for child wasting and stunting; and (3) sensitivity tests and extensions.
Descriptive Results
Our sample of 1.271 million children 0–59 months of age in 44 LMICs shows large variation in wasting and stunting prevalence across countries. Approximately 13% of the sample suffers from wasting (< -2 WHZ) while 5% are severely wasted (< -3WHZ) (Supplemental Table S1). Stunting is much more common, with 35% of the sample stunted (HAZ<-2) and 15% severely stunted (HAZ<-3). Wasting and stunting prevalence vary by region and by child age, with important implications for regression analyses.35
Figure 2 reports smoothed regression estimates of wasting and stunting by child age for DHS sub-regions with the highest child undernutrition burdens, all of which are in sub-Saharan Africa and Asia. In Panel A we observe large variations in wasting prevalence across regions that seem only weakly associated with overall regional economic development levels. For example, wasting prevalence is highest in South Asia and South-East Asia, in spite their higher economic development than African regions. The only exception is the high levels of wasting observed in the Sahel region, especially in the first 2 years of life. Wasting patterns also show interesting variations by child age, with almost 30% of South Asian children in our sample being born wasted, compared to ~ 20% in South-East Asia and in the Sahel and between 10 and 13% in other African regions. In the two Asian regions, wasting prevalence declines gradually from birth to 24 months of age when it reaches around 18–20% and stabilizes thereafter. For African regions (except the Sahel), similar declines are observed over time, especially after 12 months of age and wasting remains much less prevalent than in Asia throughout the full first 60 months of life. The Sahel shows quite different age-related patterns of wasting, with rapid increases during the first year (up to 28% by 10 months of age) and then falls sharply to around 10% by 36 months of age before levelling off like in other regions. Our results showing the highest burden of wasting between birth and 3–6 months are consistent with global evidence36 and are of great concern given the association between wasting and excess mortality risk in this age group.23
Stunting, which results from the cumulative effect of repeated or chronic cumulative nutritional insults, shows very different patterns (Panel B, Fig. 2). First, stunting prevalence does not show as much variation across high-burden regions as wasting does. In all Asian and African regions included in our sample, between 10 and 20% of children are born stunted, but stunting prevalence increases gradually from around 4 to 20 months of age, before levelling off thereafter. The process of becoming stunted therefore mostly occurs during the “first 1000 days” of life, from conception to approximately age 2 years.
Supplement Figure S1 uses the same regression smoothing technique (with 95% confidence intervals) to demonstrate heterogeneity in wasting prevalence by demographic, geographic and socioeconomic characteristics. Male and female children are born with similar wasting risks (Panel A), but wasting is significantly higher for boys compared to girls, which is likely related to the male fragility hypothesis.37 Panel B shows that wasting is much higher among rural children from birth all the way to age 5 years, with the difference varying between 3–6 percentage points, reflecting the many disadvantages that rural populations have in socioeconomic status and access to nutrition-relevant services, such as healthcare.38,39 Panel C focuses more explicitly on differences in socioeconomic status and shows higher prevalence of wasting among children from asset-poor households compared to those from non-poor households (see also Supplement Figure S2 for asset-poverty comparisons in India specifically). Finally, Panel D focuses on a key measure of rural wealth as well as potential resilience to food price shocks and confirms that children from rural-farm households are less likely to be wasted, typically by around 3 percentage points.
This study exploits variation in the real price of food within DHS countries, as measured by changes in the ratio of the consumer price index (CPI) for food to the CPI for all consumer items. We variously term changes in this food CPI/total CPI ratio “food inflation” or “real food price increases”. For wasting regressions we measure real food price changes for the three months prior to the month of anthropometric measurement,40 but for stunting and child diet diversity regressions we use 12-month food price changes in the prenatal period and first and second years of life. In all our regression analyses we essentially exploit the fact that DHS rounds are conducted in both low food inflation and high food inflation periods, including a number of surveys conducted in the 2007–2011 period when international and domestic food prices were highly volatile. Figure 3 demonstrates this by reporting mean, minimum and maximum 12-month food price changes in the 130-round DHS sample. There are large temporal and cross-country variations in real food price changes through most of the period in question. There were also 38 DHS surveys conducted between 2007–2011 when international and domestic price volatility was very high, with striking instances of 3-month food inflation in Liberia (10% in April 2007), Bolivia (15% in June 2008), Kenya (18% in January 2009), Ethiopia (13.6% in mid 2011) and Uganda (20% in mid 2011). Figure S3 in the Supplement also reports a histogram of the distribution of 3-month real food inflation measure used for the wasting analysis.
Main regression results: Food inflation and child wasting and stunting
Figure 4 reports coefficient plots from weighted regressions that represent the predicted impact of 5% increases in the food/total CPI ratio for the past three months on wasting (black circles) and severe wasting risks (blue circles), with 95% confidence intervals. The regression approach follows a previous study on macroeconomic shocks and child wasting41 in controlling for DHS-based predictors of wasting, country fixed effects, and region-specific time trends, seasonality factors, and wasting-age dynamics. Regressions are weighted to be representative of this specific sample of DHS countries, and changes in the food/total CPI ratio are interacted with a country’s mean wasting prevalence across all its survey rounds to ensure that the impact of inflation is proportional to a country’s typical wasting prevalence. This is important because countries with very low and very high levels of wasting are unlikely to see the same absolute change in wasting risks from a given macroeconomic shock;41 for example, countries with low wasting rates will have very few children close to the − 3 z-score threshold for severe wasting, so even a large food price shock will have little impact on severe wasting prevalence in an absolute sense.
Due to the striking wasting-age patterns described above, the results are stratified by child age. In the full sample of children 0–59 months the regression coefficient is positive and highly statistically significant, implying that a 5% increase in the food/total CPI ratio – equivalent to around two standard deviations in this sample – predicts a 9% increase in the risk of wasting. This marginal effect varies from 11% for children 0–11 months of age to 6% for children 12–23 months of age. Coefficients for severe wasting are somewhat larger in magnitude but less precisely estimated. A 5% increase in real food prices predicts a 14% increase in severe wasting for children 0–59 months, with similar magnitudes of effects in the 11–23 and 24–59 month samples. The coefficient for the 0–11 month age group is not statistically significant in a two-sided test.
Given the large and statistically significant elasticity of wasting with respect to food inflation in the 0–11 month sample in Fig. 4, it seems plausible that food inflation increases the risk of low birthweight (or wasting at birth) by adversely affecting maternal nutrition during pregnancy. In Fig. 5 we therefore examine wasting risks for infants 0–5 months of age. We find that for both wasting and severe wasting, the coefficients are large and statistically significant, supporting the hypothesis that a prenatal maternal nutrition mechanism links food inflation to wasting at birth and in the first few months of life. This result is important because mortality rates among newborns and young infants are especially high, suggesting food inflation poses a major risk for infant mortality through a maternal malnutrition pathway.
Testing for heterogenous effects of food inflation on wasting by sociodemographic groups
Biological and economic theories suggest that the impacts of food inflation on undernutrition may be heterogeneous in other dimensions, and Supplement Figure S1 demonstrated significant disparities in wasting prevalence across demographic and socioeconomic strata. Table 1 therefore introduces interaction terms between food inflation and these strata for moderate/severe wasting, while Supplement Table S5 reports results for severe wasting.
In regression (1) we observe that food inflation’s wasting risks are around twice as large for boys as they are for girls. In regression (2) we find that children from urban households are less likely to become wasted after food inflation compared to their rural counterparts. While surprising – urban households are mostly net-food consumers highly dependent on markets – it is also true that urban children are less likely to have low WHZ scores to begin with, and also have better access to health and nutrition services, and more assets and parental education. In regression (3) we find that asset-poor children are much more likely to become wasted than non-poor children. A 5% increase in real food prices increased the risk of wasting by just 6% for non-poor children, but by 15% for asset-poor children.
In regression (4) we turn to a rural sample to test whether food inflation’s impact on wasting depends on whether a household owns land (and can therefore produce food) or is landless (and therefore dependent on market purchases). The coefficient on farmland ownership is highly significant and reduces the impacts of food inflation on wasting by nearly half. In regression (5) we focus on the full sample again and test multiple interactions (except farmland ownership) and find that the interactions persist in the presence of each other, although the partial protection of living in an urban area is reduced when the poverty interaction is added. In regression (6) we focus on the rural sample and include all the interactions, including farmland ownership. All the interaction coefficients remain statistically significant, although the magnitude of the coefficient on poverty actually increases in absolute magnitude.
The presence of multiple statistically significant coefficients in regressions (5) and (6) implies additive effects. For example, the worst affected group are children from landless rural households that are also asset poor: regression (6) predicts that a 5% increase in the real price of food for this highly vulnerable group results in a 29 percent increase in the risk of wasting for boys and a 23% increase in the risk of wasting for girls. We also note that results for severe wasting are qualitatively very similar to those reported in Table 1 (See Supplement Table S5).
Table 1
Weighted multivariate linear probability models of wasting risks as a function of 5% increases in the real food price index over the past 3 months interacted with urban locality, gender, asset poverty and farmland ownership
|
(1)
|
(2)
|
(3)
|
(4)
|
(5)
|
(6)
|
Interaction variable
|
Girl child
|
Urban location
|
Asset-poverty
|
Owning farmland
|
Multiple
|
Multiple
|
Sample
|
Full
|
Full
|
Full
|
Rural
|
Full
|
Rural
|
Food inflation (base group)
|
0.11***
|
0.10***
|
0.06**
|
0.16***
|
0.10***
|
0.16***
|
|
(0.06–0.16)
|
(0.05–0.16)
|
(0.02–0.11)
|
(0.13–0.20)
|
(0.06–0.15)
|
(0.12–0.19)
|
Food inflation*girl
|
-0.05***
|
|
|
|
-0.05***
|
-0.06***
|
|
(-0.07 - -0.03)
|
|
|
|
(-0.07 - -0.03)
|
(-0.07 - -0.05)
|
Food inflation*urban
|
|
-0.06***
|
|
|
-0.04**
|
|
|
|
(-0.10 - -0.02)
|
|
|
(-0.07 - -0.01)
|
|
Food inflation*asset-poor
|
|
|
0.09**
|
|
0.08**
|
0.13***
|
|
|
|
(0.02–0.16)
|
|
(0.01–0.14)
|
(0.08–0.18)
|
Food inflation*farmland
|
|
|
|
-0.07***
|
|
-0.07***
|
|
|
|
|
(-0.09 - -0.04)
|
|
(-0.09 - -0.04)
|
Girl child
|
-0.02***
|
-0.02***
|
-0.02***
|
-0.02***
|
-0.02***
|
-0.02***
|
|
(-0.02 - -0.01)
|
(-0.02 - -0.01)
|
(-0.02 - -0.01)
|
(-0.02 - -0.01)
|
(-0.02 - -0.01)
|
(-0.02 - -0.01)
|
Urban locality
|
0.01***
|
0.01***
|
0.01***
|
|
|
|
|
(0.00–0.01)
|
(0.00–0.01)
|
(0.00–0.01)
|
|
|
|
Asset-poor
|
0.03***
|
0.03***
|
0.03***
|
0.03***
|
0.03***
|
0.04***
|
|
(0.02–0.03)
|
(0.02–0.03)
|
(0.02–0.04)
|
(0.03–0.04)
|
(0.02–0.04)
|
(0.03–0.05)
|
Farmland ownership
|
|
|
|
-0.00
|
|
-0.00
|
|
|
|
|
(-0.01–0.00)
|
|
(-0.01–0.00)
|
Observations
|
1,271,886
|
1,271,886
|
1,271,886
|
719,457
|
1,271,886
|
719,457
|
R-squared
|
0.06
|
0.06
|
0.06
|
0.05
|
0.05
|
0.05
|
Notes: 95% confidence intervals based on standard errors clustered at the country level are reported in parentheses. The regressions control for DHS-based predictors of wasting, country fixed effects, and region-specific time trends, seasonality factors, and wasting-age dynamics, and are weighted to be representative of this specific sample of DHS countries. Changes in the food/total CPI ratio are interacted with a country’s mean wasting prevalence across all its survey rounds to ensure that the impact of inflation is proportional to a country’s typical wasting prevalence. See Methods and Materials for more details. The full sample includes 1.271 children in 44 LMICs.
Food inflation in the first 1000 days of life and subsequent stunting risks
Even relatively brief nutritional insults can have longer term consequences on child growth and development, especially insults that occur in the first 1000 days of life when children are highly vulnerable stunting rates climb precipitously in LMICs (Fig. 2, Panel B). We therefore tested whether food inflation in the prenatal period or the first or second years after birth is a longer-term risk factor for stunting in the 24–59 month period (after the first 1000 days). Figure 6 reports results from separate regressions for stunting and severe stunting. For both indicators we find that food inflation in the prenatal period or the first year after birth significantly elevates the risk of stunting in the 24–59 month period, while food inflation in the second year of life also has positive, but non statistically significant coefficients (in two-sided tests). For moderate/severe stunting the findings suggest that a 5% increase in food prices in the prenatal period increases the risk of stunting by 1.6 percent, and by 1.8 percent in the first year after birth. The point estimates are around twice as large for severe stunting but are still much lower than for wasting. The fact that food inflation in the prenatal period is a strong predictor of later-life stunting is consistent with the indirect evidence reported in Fig. 5 that inflation during pregnancy might affect intra-uterine growth and birthweight (proxied here by wasting during the first few months of life) We also tested for heterogeneous effects of food inflation by introducing the interaction terms used in Table 1, although there is no clear evidence that the same interaction effects hold for stunting (Supplement Table S7).
Assessing dietary diversity and symptoms of infections as likely mechanisms linking food inflation to wasting and stunting
Whilst wasting and stunting are affected by both diets and health, it is likely that the main mechanism linking food inflation to wasting is maternal nutrition and diets during pregnancy and the adequacy of infant and young child feeding practices and diets during postnatal life. The DHS allows measurement of minimum dietary diversity (MDD) (a proxy for diet quality) for children 6 months and older (although here we reverse the variable to measure inadequate dietary diversity). Adequate diet diversity captures whether a child consumed at least four of seven recommended food groups in the past 24 hours. Data on maternal diet diversity is not available in the surveys used, but child and maternal dietary diversity have been shown to be strongly correlated.42 Child dietary diversity does not capture quantities consumed, but the indicator has been shown to predict mean micronutrient adequacy43 and energy intake.44 In our dataset we use regression analysis to show that poor diet diversity predicts an increased risk of wasting by 1.2 percentage points for children 6–23 months, while it increases the risk of stunting by 4.7 points for children 18–23 months (Supplement Figure S4). Reported diarrhea and fever symptoms in the previous two weeks are also associated with wasting and stunting. For stunting one should probably interpret these associations as indicative of the recent dietary status or illness being reasonable proxies for longer term dietary status and exposure to disease.
Testing for impacts of total inflation on wasting and stunting
While relative increases in food prices predict increased risks of wasting and stunting, it is possible that non-food inflation could also adversely affect household welfare and child undernutrition. To test this, we added total inflation to the wasting and stunting regression models reported above. The results are presented in Supplement Figures S5 (wasting) and Figure S6 (stunting). Total inflation does not appear to be a statistically significant risk factor for either stunting or wasting. This may be because the poor spend less on non-food goods and services than on food, and because some big-ticket non-food expenditures – such as rent, public healthcare and schooling – are less subject to short term volatility and inflation, and also have less direct connection to diet quality or disease pathways.