Choosing between random effects models in meta‐analysis: Units of analysis and the generalizability of obtained results

In meta‐analysis, the choice between different models is recognized as important and as requiring empirical and theoretical justification. However, researchers are often unclear on how to make this choice. Each kind of model makes assumptions and brings implications regarding the generalizations that can be made from the data. The present article summarizes standard understandings of fixed versus random model concepts with special focus on a choice not often discussed: that between a weighted random effects model and an unweighted random effects model. The weighted effects approach is currently predominant as a random effects model, while the unweighted model, although once prevalent, is rarely used anymore. The present article revives the unweighted model as a legitimate, practical, and sometimes preferable alternative. The unweighted effects model allows the fullest generalization to future and unretrieved studies and is easiest to implement, although it is generally the most conservative and therefore typically has the least statistical power. Short description: This article summarizes the meaning and usage of fixed versus random effects models in meta‐analysis and brings attention to a type of random effects model (unweighted effects) that is less often used but that offers full generalization to future studies and is also easier to implement.