Detecting Selection Bias in Meta-Analyses with Multiple Outcomes: A Simulation Study

This study explores the performance of classical methods for detecting publication bias-namely, Egger's regression test, Funnel Plot test, Begg's Rank Correlation and Trim and Fill method-in meta-analysis of studies that report multiple effects. Publication bias, outcome reporting bias, and a combination of these were generated. Egger's regression test and the Funnel Plot test were extended to three-level models, and possible cutoffs for the estimator of the Trim and Fill method were explored. Furthermore, we checked whether the combination of results of several methods yielded a better control of Type I error rates. Results show that no method works well across all conditions and that performance depends mainly on the population effect size value and the total variance.

Automated summary

The study found that classical methods for detecting publication bias perform differently under different conditions, depending on the population effect size and total variance.