An Illustration of the Consequences of Meta-Analysis Model Choice

Abstract

Fixed- and random-effects models represent two different approaches to analyzing and understanding data with meta-analysis. The current article describes the results of a two-part study to illustrate the effect of choice of meta-analytic model on study conclusions. Part 1 illustrates the effect of model choice by analyzing data simulated to conform to either fixed- or random-effects scenarios with both fixed- and random-effects methods of data analysis. Part 2 uses two published meta-analyses to show that methodological choices, in this case mainly the choice of fixed- or random-effects models, affect estimates both of mean effect size and of the random-effects variance component (REVC). Overall, results suggest that random-effects procedures represent the best initial choice when conducting a meta-analysis.