Regression-based inequality decomposition: Pitfalls and a solution procedure

This paper explores pitfalls in regression-based inequality decompositions. A simple procedure is developed for rectifying these pitfalls. The procedure does not impose any restrictions on the underlying regression model and it can be applied to any inequality measure(s). Once combined with conventional decomposition methods or the Shapley value approach of Shorrocks (1999), what is being proposed becomes a most general and powerful framework for regression-based inequality decomposition. Empirical examples are provided to demonstrate the use of the procedure, and to contrast our results with those based on recent developments of Fields and Yoo (2000) and Morduch and Sicular (2002).