Using g alternatives to reduce subgroup differences when predicting US public‐sector performance

Abstract

PurposeThe purpose of this paper is to bring attention to the issues of validity and subgroup differences of selection devices currently being used in the public sector.Design/methodology/approachAn attempt is made to identify unfair hiring practices, particularly important within the public sector, as this area of employment is characterized by a unique set of circumstances. Among them, economic constraints, the social burden to ensure fair treatment among applicants and incumbents, and an increasingly higher expectation of quality service among customers in the public sector. This paper also explores the effectiveness of two strategies for reducing subgroup differences while maintaining or increasing criterion‐related validity.FindingsThe findings of this study are important and answer some central questions. First, g and job knowledge were the best individual predictors of overall performance criteria; second, the g, alternative, and full models all significantly predicted the performance criteria, with the alternative model predicting more variance than the g model; third, the alternative model had more incremental validity over the g model than the g model had over the alternative model; the alternative model also produced less subgroup differences for Black–White comparisons than the g model. The Native American‐White differences were larger for the alternative model compared to the g model, but these differences are considered small effects and were non‐significant in the statistical sense. The Hispanic‐White differences were also somewhat larger for the alternative model when compared to the g model; however, this result is probably unreliable due to a very small Hispanic sample size and is a small effect. Thus, the alternative model will predict performance well for similar public sector samples while producing generally smaller subgroup differences.Originality/valueThere is little extant published research examining the validity and ethnic group score differences of alternate predictors used in the US public sector and the current effort seeks to provide empirical evidence to fill this void.