Blackwell Reference Online

Extract

A very narrow definition of differential validity is that the validity coefficient for one group differs significantly from the validity coefficient for another group. A broader, and a more robust, definition of differential validity takes into account differences in the respective prediction equations. In this case, differential validity refers to the possibility that different prediction equations can be generated for different subgroups of a population, as defined by factors such as race, sex, ethnic group, or religion. Slopes of the regression lines, intercepts, and standard errors of estimate are all considered under the umbrella of different prediction equations. To determine if differential validity exists between groups, one must initially generate separate prediction equations for each group (e.g., separate regression equations for black versus white applicants). Evidence for differential validity exists in the narrow case if the slopes of the regression lines differ significantly between the groups. In the broader case, tests of significance are also conducted on intercepts and standard errors of estimate. When significant differences are found for any of these factors, it suggests that information from the predictor should not be interpreted without considering subgroup membership. When evidence of differential validity exists, a score on the predictor is related to a ... log in or subscribe to read full text