|
Abstract:
|
This research evaluated inferential meta-analytic procedures in cases where few studies are available to precisely estimate potential population random effects. In a simulation experiment, we manipulated the number of available studies, the average study sample size, the size of the population random effect, and the mean population effect size. Within each cell of this design, we contrasted three common modeling strategies: always assuming fixed effects, always assuming random effects, and assuming fixed effects unless a significant heterogeneity test is observed. Results suggest that no modeling strategy is clearly preferable across all the studied conditions that are knowable in practice and therefore choice of strategy should be based and defended on the analyst's relative concern with Type I versus Type II errors.
|
