Six Entrenched Misconceptions About Meta-Analysis Models

Meta-analysis has become central to evidence-based medicine, yet persistent misconceptions continue to distort how models are selected and interpreted. This paper identifies and clarifies six entrenched misconceptions that have shaped the practice of meta-analysis for decades. It first challenges the belief that the choice of parameter assumption determines whether inference can extend beyond the included studies, emphasizing that conditional or unconditional inference arises from the analyst's scientific objective-not from the model or its assumptions. Second, it corrects the notion that the fixed-effect (FE) model is the only model under the common parameters assumption (aCP), noting that several modern models within this framework can accommodate heterogeneity. Third, it dispels the idea that only random-effects (RE) models can address heterogeneity, showing that this property exists under any parameter assumption. Fourth, it refutes the practice of letting observed heterogeneity dictate model choice, arguing that parameter assumptions and inferential purpose must guide decisions instead. Fifth, it challenges the claim that RE estimators best handle overdispersion, demonstrating persistent error-estimation flaws and inflated type I error rates. Finally, it contests the view that heterogeneity renders aCP-based models unrealistic, highlighting that recent aCP models handle such diversity effectively. By disentangling these misconceptions, the paper proposes a purpose-driven, assumption-aware framework for model selection that prioritizes conceptual clarity, analytical validity, and reproducibility in evidence synthesis.