Residuals from two‐step research designs

A number of studies have begun to question the appropriateness of the widespread and pervasive usage of regressions in two‐step research designs in the accounting literature to obtain variables. The use of residuals from a first‐step model as a dependent variable in a second step has been used in contexts such as earnings management and unexpected audit fees. In this paper, I provide a discussion of the issues within the use of residuals from a first‐step model in a second step. Taking the case of discretionary accruals models, I demonstrate some of the model specification issues explicitly and implicitly considered within these criticisms. I then add a discussion over some concerns regarding the economic interpretation in the use of first‐step residuals.