Meta-Analysis and the Art of the Average
Frederick L. Oswald
Michigan State University
Rodney A. McCloy
Human Resources Research Organization (HumRRO)
Since its introduction into the organizational research literature 25 years ago (Schmidt & Hunter, 1977), meta-analysis has been the focus of numerous conceptual and methodological controversies and developments. Before we delve into some of those issues, let us first remember that meta-analysis can be seen as essentially two things: (a) statistically, it provides an estimate of the mean and variance of the effect size estimates across studies (e.g., correlations, d-values, odds ratios); and (b) more broadly, it provides a statistically and rationally driven process of identifying, gathering, coding, combining, and interpreting results across studies. Clearly these two characterizations of meta-analysis are interrelated: Conceptual and practical considerations inform the decision points regarding what research data to summarize and how to summarize them (Matt, 1989; Wanous, Sullivan, & Malinak, 1989), and in turn, the statistical results and substantive interpretation from meta-analysis inform further research and practice.
In the sections that follow, the popularity of meta-analysis in psychological research is briefly documented first; meta-analysis is clearly here to stay for a long time, at least in some form, even if not in the exact statistical forms we use today. Second, how the metaanalytic mean is computed is discussed and, consequently, how researchers and practitioners should interpret it. Third, several considerations when making statistical artifact corrections in a meta-analysis are reviewed—an important topic because so many