Measuring Within and Between Group Inequality in Early-Life Mortality Over Time: A Bayesian Approach with Application to India

Abstract: Most studies on inequality in infant and child mortality compare average mortality rates between large groups of births, for example, comparing births from different countries, income groups, ethnicities, or different times. These studies do not measure within-group disparities. The few studies that have measured within-group variability in infant and child mortality have used tools from the income inequality literature, such as Gini indices. We show that the latter are inappropriate for infant and child mortality. We develop novel tools that are appropriate for analyzing infant and child mortality inequality, including inequality measures, covariate adjustments, and ANOVA methods. We illustrate how to handle uncertainty about complex inference targets, including ensembles of probabilities and kernel density estimates. We illustrate our methodology using a large data set from India, where we estimate infant and child mortality risk for over 400,000 births using a Bayesian hierarchical model. We show that most of the variance in mortality risk exists within groups of births, not between them, and thus that within-group mortality needs to be taken into account when assessing inequality in infant and child mortality. Our approach has broad applicability to many health indicators.