The Use of Mixture Random-Effect Models in Meta-Analysis: A Follow-Up Simulation Study

• In Event: Advancing Methods for Meta-Analysis in Education Research

Abstract

Mixture models can be used to detect concealed moderator variables in meta-analysis, and the aim of this study is to explore their performance in this field. In a previous -pilot- simulation study, we have observed that goodness-of-fit index DIC do not perform well at selecting the correct number of underlying clusters. Therefore, in this new simulation we have explored the performance of other goodness-of-fit indexes, such as BIC and AIC, and we have applied mixture models using the SEM-based proposed by Cheung (2008). Results show that BIC is the best index to select the correct number of clusters, and that parameter estimates are accurate under most conditions, except when the between-studies variability is large and cluster means are too similar

Authors

• Belén Fernández-Castilla, University of Leuven

• Jung Yeon Park, University of Leuven

• Wim Van den Noortgate, KU Leuven