Random effects meta-analysis suits heterogeneous literatures with ill-specified treatments

Evidence

Whereas most past meta-analytic studies in social psychology have used fixed effects models (Johnson & Eagly, 2000), the nature of the contact research literature makes a random effects model more appropriate for our analysis. This model holds that part of the differences in effects across studies is essentially random and involves unidentifiable sources (Hedges, 1994; Hedges & Olkin, 1985; Lipsey & Wilson, 2001; Mosteller & Colditz, 1996; Raudenbush, 1994; Rosenthal, 1995). As Cook et al. (1992) concluded, this approach is “particularly attractive when considering (1) studies that are quite heterogeneous, (2) treatments that are ill-specified, and/or (3) effects that are complex and multidetermined” (p. 310). Because all three of these conditions hold for much of the intergroup contact literature, we used a random effects model. — Thomas F. Pettigrew, Linda R. Tropp, A Meta-Analytic Test of Intergroup…, 67

Categorization

• DEFINE: Clearly lays out when a random effects model beats a fixed effects model in meta-analysis — three concrete conditions that justify its use.
• XREF: Connects to research-methods knowledge about meta-analytic modeling choices, and to the different conditions (all three hold for intergroup contact literature) that signal random effects appropriateness.

Related

• Meta-analysis pools all studies to estimate effect sizes and moderators (similarity 0.86)
• Subjective literature reviews suffer selection biases that meta-analysis avoids. (similarity 0.83)
• A meta-analysis synthesized 515 studies with 250,089 participants across 38 nations. (similarity 0.81)
• No meta-analysis of intergroup contact research existed before this study. (similarity 0.81)
• Stronger contact effects in rigorous studies would confirm the contact-prejudice link. (similarity 0.80)
• Contact studies with cleaner control groups show larger effects. (similarity 0.79)
• Meta-analyses overestimate effect sizes because null results go unpublished. (similarity 0.79)
• Randomized contact experiments yield near-large prejudice reduction effects. (similarity 0.79)
• Prior contact research reviews each covered fewer than 60 studies. (similarity 0.79)
• Contact studies vary in how they generalize findings beyond immediate settings. (similarity 0.78)