Additional results on the performance of location-scale models in meta-analysis: A simulation study

Abstract

Location-scale models are a useful tool in the field of meta-analysis since they allow the influence of moderator variables on the mean (location) and variance (scale) of the distribution of true effects to be studied at the same time. The implementation of location-scale models for meta-analysis was recently added to the metafor package for the R statistical software. In previous conference presentations, the results of a Monte Carlo simulation study comparing different estimation methods (maximum or restricted-maximum likelihood estimation), significance tests (Wald-type or likelihood-ratio tests), and methods for constructing confidence intervals for the scale coefficients (Wald-type and profile-likelihood intervals) were presented in terms of rejection rates and coverage probabilities. However, due to time constraints, other important results were not offered. With the present work, our goal is to present additional results regarding the bias and mean squared error, in the first place, of the estimates of the scale coefficients and, secondly, of the heterogeneity values for the levels of the moderator variable. Results are discussed with respect to the estimation method, the type of moderator variable (dichotomous or continuous), the heterogeneity values given to the levels of the moderator variable, the number of studies within the meta-analysis, and the average sample size of the included studies.