Test Bias, Differential Prediction, and a Revised Approach for Determining the Suitability of a Predictor in a Selection Context

Abstract

The most commonly used and accepted model of assessing bias in a selection context is that proposed by Cleary in which predictor-criterion regression lines are tested for both slope and intercept equality. With this approach, any difference in intercepts or slopes is considered an indication of bias. We argue that differing regression lines intercepts is indicative of differential prediction but not test bias. We describe several fundamentally different potential causes of differences in groups’ regression line intercepts, many of which are unrelated to test properties. We argue that differential prediction because of such sources should not preclude the use of the test in selection contexts. We propose a new procedure to potentially identify the source of regression line differences and illustrate this framework using a job incumbent sample.