A new outlook on identification in detailed wage decompositions

Abstract

Some articles have shown that the conventional Oaxaca¿s wage decomposition methodology poses a serious problem when categorical variables are included: the contribution of each dummy variable to the wage discrimination may vary. This article shows that this problem originates from an inadequate interpretation of the models and of the contributions of dummy variables to discrimination, and that with an adequate interpretation of the model the problem is easily solved. We also show that the proposal made by Gardeazabal and Ugidos (2004) for the solution of this problem, widely employed in empirical literature, contains some important drawbacks.