ESG Frameworks and Primary-Care-Amenable Child Mortality: A Global Multi-Method Assessment

Environmental, Social and Governance frameworks are increasingly invoked in health policy, yet their three pillars are almost always studied in isolation and rarely against outcomes that primary care can influence. Using the World Bank Sovereign ESG Data Portal, this study decomposes the determinants of under-five mortality — the canonical primary-care-sensitive outcome and the target of SDG 3.2 — across the full ESG structure. Three equations share the same dependent variable and differ only in their regressor block: Environment (3,533 observations, 174 economies, 2000–2020), Social (2,032 observations, 99 economies, 2001–2023) and Governance (3,592 observations, 176 economies, 2000–2022). Each block is analysed three ways: panel econometrics comprising pooled OLS, one-way and two-way fixed effects, random effects, Between, WLS and dynamic GMM estimators; unsupervised taxonomy comparing six clustering algorithms across eleven internal validation indices; and supervised validation comparing six learners under both random and country-grouped cross-validation. Safely managed drinking water is the only predictor significant in all six static estimators including two-way fixed effects, with a ten-point coverage increase associated with 21% lower mortality within countries, while particulate exposure is robust on both the cross-country and within-country margins. Institutional quality becomes significant across all estimators only after the four Worldwide Governance Indicators, correlated between 0.80 and 0.94, are orthogonalised — a collinearity independently confirmed by unsupervised dimension reduction and by out-of-fold permutation importance. Within-country explanatory power ranks governance ahead of social and environmental blocks, while out-of-sample prediction ceilings run in the opposite order. Machine learning reproduces the econometric hierarchy, reveals a U-shaped renewable-energy relationship and a threshold in institutional capacity, and shows that random train-test splits overstate accuracy by up to 23 points of R2 through country memorisation. ESG determinants of child survival are pillar-specific in both magnitude and functional form: service access dominates, institutional quality matters above a capacity threshold, and environmental exposures act through structural conditions rather than year-on-year change.