Clarification to Kish-Gephart et al. (2010)

Abstract

Reports an error in "Bad apples, bad cases, and bad barrels: Meta-analytic evidence about sources of unethical decisions at work" by Jennifer J. Kish-Gephart, David A. Harrison and Linda Klebe Treviño ( Journal of Applied Psychology , 2010[Jan], Vol 95[1], 1-31). In the recent article “Bad Apples, Bad Cases, and Bad Barrels: Meta-Analytic Evidence About Sources of Unethical Decisions at Work,” by Jennifer J. Kish-Gephart, David A. Harrison, and Linda Klebe Treviño ( Journal of Applied Psychology , 2010, Vol. 95, No. 1, pp. 1–31), we quantitatively summarized and investigated moderators of the effects of individual, situational, and organizational variables on unethical behaviors and intentions. We used two meta-analytic procedures—one based on artifact corrections (Hunter & Schmidt, 2004) and the other based on hierarchical linear modeling (Erez, Bloom, & Wells, 1996). Following previous descriptions (Erez et al., 1996; Hedges & Vevea, 1998), we may have created confusion by characterizing the Hunter & Schmidt meta-analytic procedure as a “fixed effects” approach. As with Erez et al. (1996), the core assumption of Hunter and Schmidt (2004) is based on the random effects idea that observed correlations are sampled from a population distribution of heterogeneous effects. The Hunter and Schmidt model decomposes the observed distribution of effect sizes into a component of true population variation and artifact variation. Specifically, subtracting an estimate of artifact variation from the observed distribution provides an estimate of true population variation. However, Hedges and Vevea (1998) noted that the study effect sizes in the Hunter and Schmidt model are weighted with fixed effects weights (see also Viechtbauer, 2005). In the Erez et al. random effects model, true variation is estimated through an iterative estimation procedure; its modeling assumptions use iterative sampling weights that create different point estimates and confidence intervals for effect size distributions. Approaches commonly used to test for a priori moderators also differ across these methods. Most applications of Hunter and Schmidt for identifying moderators follow a successive, subgrouping process (see Hunter & Schmidt, 1990), whereas the Erez et al. and related procedures might adopt a weighted regression. In both cases, significant moderator effects are fixed effects, whereas remaining true variance not due to statistical artifacts reflects random effects. This is called a mixed effects model (see Overton, 1998). It is important to note that, as we reported in Tables 2–4, both procedures led to the same substantive findings about unethical choices in organizations. We encourage meta-analytic researchers to further explore differences among these methods and to evaluate when they materially affect research conclusions. (The abstract of the original article appeared in record 2010-00343-001 .) (PsycInfo Database Record (c) 2025 APA, all rights reserved)