The Usefulness of Directed Acyclic Graphs: What Can DAGs Contribute to a Residual Approach to Weight-Related Income Discrimination?

Abstract

This paper provides one of the first empirical applications of directed acyclic graphs (DAGs) on a research question typical for the social sciences: wage discrimination. Besides a substantial interest in the weight wage penalty we ask whether DAGs help to improve the widely applied residual approach to discrimination. Using the German Socio-economic Panel (GSOEP) we find that body composition is associated with wages and that the effects of fat mass and fat free mass are markedly stronger for females than for males. Further we show that DAGs help to identify covariates which should and should not be adjusted for and to reduce the statistical model without losing information with regard to the estimation of the effect of interest. However, DAGs do not necessarily ensure that the central assumption of the residual approach, selection on observables, holds.