Determinants, detection and amelioration of adverse impact in personnel selection procedures: Issues, evidence and lessons learned

Mean subgroup (gender, ethnic/cultural, and age) differences are summarized across studies for several predictor domains - cognitive ability, personal ity and physical ability - at both broadly and more narrowly defined constr uct levels, with some surprising results. Research clearly indicates that t he setting, the sample, the construct and the level of construct specificit y can all, either individually or in combination, moderate the magnitude of differences between groups. Employers using tests in employment settings n eed to assess accurately the requirements of work. When the exact nature of the work is specified, the appropriate predictors may or may not have adve rse impact against some groups. The possible causes and remedies for advers e impact (measurement method, culture, test coaching, test-taker perception s, stereotype threat and criterion conceptualization) are also summarized. Each of these factors can contribute to subgroup differences, and some appe ar to contribute significantly to subgroup differences on cognitive ability tests, where Black-White mean differences are most pronounced. Statistical methods for detecting differential prediction, test fairness and construct equivalence are described and evaluated, as are statistical/mathematical s trategies for reducing adverse impact (test-score banding and predictor/cri terion weighting strategies).