ENSO impacts child undernutrition in the global tropics

The study investigates how interannual climate variability driven by the El Niño Southern Oscillation (ENSO) affects child nutrition in the global tropics. Prior work has shown ENSO influences agriculture, economies, and health, but global, systematic health effects—particularly on undernutrition—are understudied. Because ENSO events are partially predictable months in advance, understanding their impact could enable proactive public health and food security responses. Existing analyses often focus on single countries or episodes and lack global scope. The authors aim to estimate the total influence of ENSO variability on children’s nutritional status across teleconnected regions, considering pathways such as agricultural productivity, infectious disease, and conflict. They leverage large-scale Demographic and Health Surveys (DHS) anthropometrics to quantify how ENSO states correspond to short-run nutritional shocks in children.

The paper situates its contribution within literature linking ENSO to agricultural yields, economic production, social stability, and infectious diseases. Prior studies document ENSO’s destabilizing effects on agriculture and associations with vector- and water-borne diseases and conflict. In the tropics, adverse yield impacts are acute due to proximity to crop temperature thresholds. Related research using DHS has linked weather variability to child nutrition in specific regions, but few studies estimate ENSO’s global health impacts or compare effects across precipitation teleconnections. The authors note ENSO forecasts’ predictive skill and the need for regionally comprehensive evidence to guide proactive interventions.

Data and sample: The authors compile over 1.25 million child anthropometric observations (ages 0–59 months) from 186 DHS surveys across 51 countries (1986–2018). Countries are included if teleconnected to ENSO via temperature: at least 50% of the population resides in pixels where local temperature significantly correlates with the 2-month lag of NINO3.4 SST for at least 3 months per year (using UDel 0.5° gridded data, 1950–2014). Precipitation teleconnections are also classified at the pixel level to distinguish regions where warmer ENSO corresponds to wet vs. dry anomalies. Anthropometrics are calculated following NCHS/CDC/WHO standards; improbable or missing values are excluded. Exposure assignment: ENSO state is measured by the NINO3.4 SST index (NOAA CPC). Treatment is the mean NINO3.4 between May and December of year t, assigned to children surveyed May of year t through April of year t+1, aligning with ENSO event evolution (avoiding misclassification across the “spring barrier”). Outcomes: Weight-for-age z-score (WAZ), weight-for-height z-score, BMI z-score, and binary indicators for underweight (WAZ < -2) and wasting (weight-for-height < -2). Econometric approach: OLS regressions estimate associations between child outcomes and contemporaneous ENSO state with the following controls: country fixed effects split by urban/rural; country-specific mother’s education (years), mother’s age at child’s birth, urban/rural indicator; UNICEF world region-specific linear trends in survey year; month-of-interview fixed effects. Standard errors are two-way clustered by interview year and first administrative unit. Observations are weighted to represent the average country (DHS sampling weights normalized within country and country-size weights), with alternative weights to represent the average child for global effects. Heterogeneity: Models include an interaction to allow different effects where precipitation is positively correlated with ENSO (>50% of area has ≥3 months of significant positive correlation) vs. neutral/negative regions. Additional analyses include flexible functional forms (local polynomial after residualization via Frisch-Waugh-Lovell), decadal and regional subsamples, lag structures to assess persistence (weight vs. stunting), alternative ENSO indices/definitions, alternative detrending (decade fixed effects), robustness to teleconnection definitions, survey timing checks, and alternative error adjustments and link functions for binary outcomes.

• Main association: A 1°C increase in May–December mean NINO3.4 is associated with a decrease in WAZ of approximately 0.025σ to 0.03σ (Table 1: -0.0251, SE 0.0105, p=0.023; flexible forms show consistent negative association). - Other contemporaneous nutrition measures: Weight-for-height z-score decreases by 0.0377/°C (p=0.0065) and BMI z-score by 0.0381/°C (p=0.0067). - WHO thresholds: Underweight prevalence increases by 0.588 percentage points per 1°C (0.00588, SE 0.00264, p=0.033). Wasting increases by 0.319 p.p./°C (0.00319, SE 0.00251) but is not statistically significant (p=0.213). - Precipitation heterogeneity: In areas where warmer ENSO leads to wet anomalies (≈6.4% of sample), effects reverse: interaction for WAZ is +0.0733 (p=0.0024), and for underweight is -0.0195 (p=0.00037), implying improved nutrition with El Niño–associated rainfall increases. - Distributional evidence: Detrended WAZ distributions differ significantly between El Niño and La Niña years (p<0.001), with worse outcomes during El Niño in dry/neutral regions. - Temporal and spatial consistency: Effects are similar across decades (1980s–2010s) and UNICEF regions; estimates are statistically indistinguishable from the main effect. - Persistence: No persistent lagged effects on weight-based measures, consistent with weight recovery when conditions improve; however, stunting (height deficits) persists years after ENSO-related shocks, indicating scarring. - 2015 El Niño quantification: With a 1.92°C NINO3.4 anomaly, underweight risk increased by 1.9 percentage points, implying approximately 5.9 million additional underweight children among 311 million under-5s in sample countries. Offsetting this effect would require interventions on the order of 134 million children for multiple micronutrient supplementation (CI 75–193M) or 72 million with complementary foods (CI 33–105M) or 72 million with nutrition education (CI 26–118M). - Average-country effect (without separating wet/dry): Warmer ENSO leads to a 0.040σ/°C reduction in WAZ (p=0.02) and a 1 percentage point increase in underweight prevalence (p<0.01).

The findings demonstrate a consistent negative association between warm ENSO conditions and child nutritional status across the global tropics, primarily in regions where El Niño brings drier conditions. The reversal of effects in wet-anomaly regions underscores precipitation as a key mediator, suggesting agricultural production as a principal pathway linking ENSO to nutrition, though other channels (infectious disease, flooding, conflict) may contribute contextually. The lack of attenuation across decades implies limited adaptation capacity within the income range of sampled countries, despite globalization and rising incomes. ENSO’s predictable nature implies that episodes like 2015 can erase substantial annual progress toward SDG targets (e.g., eliminating one year of gains in reducing underweight prevalence). Incorporating ENSO forecasts into humanitarian planning and early warning systems could enable proactive nutrition support and resource allocation, mitigating episodic food insecurity driven by climate variability.

This study provides global-scale empirical evidence that ENSO systematically affects child nutrition, with warmer El Niño-like conditions worsening short-run nutritional status in most teleconnected tropical regions and leaving lasting impacts on child height. Effects are robust across specifications, ages, regions, decades, and ENSO definitions, and they reverse where ENSO increases rainfall. The work highlights precipitation and likely agricultural production as key mediators and quantifies the magnitude of impacts during the 2015 El Niño. Policy implications include using ENSO forecasts to trigger anticipatory nutrition and humanitarian interventions. Future research should refine understanding of causal mechanisms across contexts, evaluate targeted adaptation strategies, and develop operational early warning systems to decouple nutrition outcomes from ENSO variability.

• Migration and selection: DHS data selectively report migration, limiting the ability to account for ENSO-induced migration in or out of surveyed areas. Severe events may differentially influence who is surveyed due to mortality, illness, or access constraints. - Survey operations: ENSO-related shocks (e.g., conflict or disasters) could affect DHS fieldwork quality or timing; however, analyses suggest survey timing is not systematically affected by ENSO, and seasonality controls are included. - Measurement error: Height is harder to measure accurately than weight in population surveys, potentially attenuating estimates for wasting. - Generalizability: The sample covers teleconnected countries with available DHS data (51 countries) and may not represent non-teleconnected or data-sparse regions. - Residual confounding: Although extensive fixed effects, detrending, and controls are used, unobserved factors correlated with ENSO cannot be entirely ruled out; nonetheless, robustness checks and placebo tests support the findings.