The paper re-evaluates the use of random-effects models in meta-analysis, particularly in the presence of unexplained heterogeneity. It discusses the justification and interpretation of such models, focusing on estimation, prediction, and hypothesis testing. The authors argue that random-effects meta-analyses often fail to provide key results and explore the effectiveness of distribution-free, classical, and Bayesian approaches. They conclude that the Bayesian approach naturally allows for full uncertainty, especially in prediction, but it is not without computational intensity and sensitivity to prior judgments. The paper proposes a simple prediction interval for classical meta-analysis and offers extensions to Bayesian meta-analysis, using an example of studies on 'set shifting' ability in people with eating disorders. The authors emphasize the importance of addressing heterogeneity, mean effect, study-specific effects, prediction, and testing in random-effects meta-analysis, and provide detailed methods for each of these objectives under different assumptions about the distribution of random effects.The paper re-evaluates the use of random-effects models in meta-analysis, particularly in the presence of unexplained heterogeneity. It discusses the justification and interpretation of such models, focusing on estimation, prediction, and hypothesis testing. The authors argue that random-effects meta-analyses often fail to provide key results and explore the effectiveness of distribution-free, classical, and Bayesian approaches. They conclude that the Bayesian approach naturally allows for full uncertainty, especially in prediction, but it is not without computational intensity and sensitivity to prior judgments. The paper proposes a simple prediction interval for classical meta-analysis and offers extensions to Bayesian meta-analysis, using an example of studies on 'set shifting' ability in people with eating disorders. The authors emphasize the importance of addressing heterogeneity, mean effect, study-specific effects, prediction, and testing in random-effects meta-analysis, and provide detailed methods for each of these objectives under different assumptions about the distribution of random effects.