Reverse Regression and Orthogonal Regression in Employment Discrimination Analysis

In a recent review article, White and Piette provide an overview of the use of reverse regressions in discrimination-related litigation. They explain the technique, provide a model application, summarize its advantages and disadvantages, and identify litigation in which it has been used. We point out weaknesses in common uses of reverse regression, some of which might cause serious misinterpretations of the data. We suggest that typical interpretations of reverse-regression results are incorrect. We also question the practice of conducting both direct and reverse regressions when studying employment discrimination, since the two approaches make mutually-inconsistent assumptions about the nature of the stochastic error; these assumptions generally can not be corroborated from the data and bias the regressions’ results. We suggest orthogonal regression as a potential alternative to the problems associated with direct and reverse regressions in some cases. We then provide a model application of orthogonal regression to discrimination-related investigations.