The misuse of regression-based x-Scores as dependent variables

Abstract

Researchers often use regression-based x-Scores (e.g., conservatism C-Score, misstatement F-Score) from a stage 1 model as a dependent variable in stage 2. We argue that this x-Score analysis can cause coefficient biases and interpretation problems because (1) x-Score does not capture new sources of variation, and (2) the estimates often hinge on unacknowledged technical assumptions. Instead, we recommend that researchers include the test variables and the relevant controls in stage 1, obviating the need for an x-Score. In replication analyses, some important published findings change after we remove the coefficient bias caused by the use of x-Score as a dependent variable.