A New Approach to Decomposition of a Bivariate Rank Dependent Index Using Recentered Influence Function Regression

Decomposition of a bivariate rank dependent index, such as the concentration index, is commonly used to explain socioeconomic inequalities in health. We introduce a new decomposition technique based on the recentered influence function that yields the marginal effects of covariates on the bivariate rank dependent index. This technique is simple to estimate and interpret. We show that our method, compared to the current standard procedure, relies on the imposition of fewer conditions and is therefore preferable both in a descriptive setting and in the estimation of causal effects given a suitable identification strategy. In an empirical application using Swedish Twin Registry data, we illustrate that this result bears out in practice. The two methods yield contradicting results due to differences in the conditions imposed. Using a within twin pair fixed effects identification strategy, our new method finds no evidence of a causal effect of education on income-related health inequality.