NBER Publications by Joseph Lupton

This paper notes a potential problem in the method of Blinder and Oaxaca the most popular method in the literature for decomposing the mean difference between groups of a given variable into the portion attributable to differences in the distribution of some explanatory variables and differences in the conditional expectation functions. In its conventional application, the Blinder-Oaxaca method requires that a parametric assumption be made about the form of the conditional expectations function. We show that misspecification is likely to result in non-trivial errors in inference regarding the portion attributable to differences in the distribution of explanatory variables. A nonparametric alternative to the Blinder-Oaxaca method is proposed. Rather than specify an arbitrary functional fo...