In psychology and medicine, diagnostic and screening measures are often used to make decisions by comparing the observed score to a cut score. If these measures contain items that exhibit measurement bias, then there might be systematic inaccuracies of who gets “caught” by the selection process. Traditionally, approaches that flag items for bias do not provide guidance about how measurement bias affects the decisions from the overall measure. In previous work on the area of selection, Millsap and Kwok developed a procedure that described the impact of ignoring measurement bias as changes in test sensitivity and specificity across groups. Recently, the Millsap and Kwok procedure has been extended to handle discrete items and make less stringent distributional assumptions. In this chapter, we discuss a version of the Millsap and Kwok procedure that accommodates discrete items and illustrate the use of this approach to evaluate how measurement bias affects the sensitivity and specificity of a measure comprised of binary items.
