Mortality Selection Bias and the Underestimation of Health Inequalities in Later Life: Evidence from India

Health inequalities between socioeconomic groups tend to narrow at older ages, in studies using conventional cross-sectional data. This apparent convergence, however, is largely an artifact of mortality selection bias: the systematic dropout of the most disadvantaged individuals from the population before they reach old age. Our study quantifies the extent of this bias in India and provides corrected estimates of age-specific health inequality using pooled data from three rounds of the National Sample Survey, comprising 1,278,372 individuals and 6,630 deaths. We document the wealth-mortality gradient using survival analysis and concentration indices, and apply Inverse Probability Weighting (IPW) to adjust health outcomes for differential survival, creating a counterfactual population that accounts for selection on observed characteristics. Finally, we implement Deaton-Paxson age-period-cohort (APC) models with and without selection adjustment. Our analysis finds evidence of a wealth gradient in mortality among working-age adults (Hazard Ratio (HR) for rich vs. poor: 0.576, p 0.10). Unadjusted age-health profiles flatten at older ages. However, after IPW correction, every age coefficient increases relative to unadjusted estimates, indicating systematic underestimation of poor health at all ages. Mortality selection bias significantly underestimates the true extent of poor health among lower SES groups in India, masking up to 83% of the health burden among young adults. Correcting this bias provides a more accurate picture of population health with direct implications for resource allocation and targeting interventions toward vulnerable populations.