Diagnosing When Evidence of Bias Is Problematic: Methodological Cookbooks and the Unfortunate Complexities of Reality

摘要

Meade and Tonidandel (2010) efficiently summarize a variety of factors that can influence results of moderated multiple regression (MMR) analyses that provide the evidentiary basis for judgments regarding the predictive bias of a test.~1 Their focal article serves as an important reminder for readers to (a) think carefully about factors that influence MMR statistics used to make judgments regarding predictive bias, and (b) perform analyses above and beyond MMR to help understand why a test may exhibit differential prediction across groups. As practitioners, we recognize Cleary's (1968) model in and of itself is insufficient to explain the sources of differential prediction, and that evidence of differential prediction is not sufficientto conclude there are problems with the test in question-despite the unfortunate semantics of labeling such findings as "evidence of predictive bias."