Three-level meta-analysis of dependent effect sizes

Area of Science:

• Statistics
• Biostatistics
• Psychometrics

Background:

• Meta-analytic methods commonly assume independent effect sizes, which is often not the case in practice.
• Dependence in effect sizes arises from various sources within and across studies.
• Existing multilevel extensions for meta-analysis are not widely known or utilized.

Purpose of the Study:

• To describe and illustrate three-level extensions of mixed-effects meta-analytic models.
• To present a three-level model for handling multiple effect sizes from the same sample within studies.
• To investigate the performance of the proposed three-level model in a simulation study.

Main Methods:

• Developed and applied three-level extensions of mixed-effects meta-analytic models.
• Focused on accounting for dependence in effect sizes within and across studies.
• Utilized a simulation study to empirically evaluate the model's performance.

Main Results:

• The three-level model effectively accounts for various sources of dependence in effect sizes.
• The proposed approach for handling multiple effect sizes from the same sample is simple and avoids imputing unknown covariances.
• Simulation results demonstrated that the three-level model yields unbiased estimates of treatment effects and standard errors.

Conclusions:

• Three-level meta-analytic models offer a robust solution for dependent effect sizes.
• The presented three-level model is a valuable tool for researchers dealing with complex data structures in meta-analysis.
• The model's validity and unbiasedness were confirmed through simulation, supporting its practical application.