Predictive Bias in Work and Educational Settings

Abstract

In this chapter we review the research and concept of predictive bias, present a new theory of when and how predictive bias will occur, and create a list of methodological and substantive confounds that complicate research on predictive bias. We note that educational and organizational researchers have often used different methods to operationalize the same concept in academic and work settings, respectively. We review these approaches and recommend a modified method based on the examination of regression plots and residuals. A new theory is presented that addresses how and when predictive bias would occur using the existing definition (i.e., the Cleary Model). Theoretically, we suggest that a performance determinants framework provides a good foundation for understanding how and when predictive bias can occur. We illustrate key concepts in our theory using existing research. We suggest that the nature of predictive bias is dependent on the reason a predictor is correlated with subsequent performance. We then use this theory to guide a review of previous research testing for the presence of predictive bias across a range of personnel selection and academic admissions tools. Although much of the literature indicates that professional selection tools are unbiased for native speakers of English in the U.S., the most troubling finding is the scarcity of information for many of the most popular selection and admissions tools.