A modeling approach to decomposing changes in health concentration curves

Abstract

This paper proposes a decomposition approach for health concentration curves. Decomposing changes in health concentration curves gives additional insight compared to decomposing a single index such as the health concentration index. First, the results would be valid for a comprehensive set of indices. Second, and more importantly, it allows for identifying heterogeneous effects along socioeconomic ranks. We use inverse propensity weighting for the overall decomposition. We use multiple recentered influence function regressions on a grid of points to identify the impact of specific covariates. We weight these regressions by the inverse propensity score of the observations to correct for errors due to departure from linearity. The paper also derives the expressions of the recentered influence functions of the relative and absolute health concentration curves since the literature does not offer the expression of these recentered influence functions. We offer an empirical illustration using information on cigarette consumption from the National Health Interview Survey of 2000 and 2020.