A Comparison of Fixed-Effects and Random-Effects Models for Multivariate Meta-Analysis Using an SEM Approach

Area of Science:

• Statistics
• Psychometrics
• Biostatistics

Background:

• Meta-analysis synthesizes research findings.
• Structural equation modeling (SEM) is used for complex data.
• Choosing between fixed-effects (FE) and random-effects (RE) models is crucial.

Purpose of the Study:

• Compare FE and RE models in meta-analysis for multivariate effect sizes within SEM.
• Evaluate model performance under varying data conditions.
• Provide guidance on model selection based on study characteristics.

Main Methods:

• Monte Carlo simulations were employed.
• Performance characteristics of FE and RE models were examined.
• Simulations varied the number of studies and primary study sample sizes.

Main Results:

• In homogeneous cases, FE and RE models showed minimal differences.
• FE models offered better standard error estimation with few, small studies.
• In heterogeneous cases, FE models produced biased estimates and inflated Type I errors.
• RE models maintained unbiased estimates and controlled Type I errors effectively.

Conclusions:

• RE models are generally preferred for meta-analysis with a sufficient number of studies and large primary study sample sizes.
• FE models may be favored in meta-analyses with fewer studies and uniformly small sample sizes.
• The choice of model impacts the reliability of meta-analytic findings.