Clarifying Misconceptions About Differential Validity and Prediction Bias Research

• In Event: Topics in Validity Evaluation

Abstract

The study of differential prediction and validity continues to receive attention from test developers, psychometricians, and applied researchers. Cleary (1968) proposed a method for evaluating prediction bias hypotheses. This paper shows that application of the Cleary model yields ambiguous interpretations of subgroup intercept and slope differences. A criterion-predictor factor latent variable model is used to clarify the interpretation of observed group differences in correlations and prediction equations. Results suggest that test-developers and decision-makers should be concerned about measurement invariance as opposed to subgroup differences in observed or latent prediction equations. Applied researchers should abandon the comparison of group correlations and prediction equations and instead adopt latent variable models to assess measurement invariance, subgroup latent distribution differences, and latent prediction bias.

Author

• Steven Andrew Culpepper, University of Illinois at Urbana-Champaign