Disentangling effect size heterogeneity in meta-analysis: A latent mixture approach

Area of Science:

• Statistics
• Machine Learning
• Psychology

Background:

• Meta-analysis aims to synthesize research findings but struggles to explain heterogeneity in effect sizes.
• Identifying factors contributing to heterogeneity is challenging due to numerous subtle influences.

Purpose of the Study:

• To develop a novel latent mixture-based method using machine learning to disentangle effect-size heterogeneity in meta-analysis.
• To improve the statistical power of heterogeneity analysis compared to traditional moderator analysis methods.

Main Methods:

• Developed a novel latent mixture-based method leveraging theoretical machine learning advances.
• Conducted mathematical analysis and simulation studies to evaluate the method's performance.
• Applied the method to a real-world case study to demonstrate its utility.

Main Results:

• The proposed method demonstrates substantially higher statistical power than traditional moderator analysis when heterogeneity arises from multiple factors.
• The method effectively disentangles heterogeneity without requiring researchers to pre-specify potential contributing factors.
• A case study illustrated the method's capability in resolving long-standing literature inconsistencies.

Conclusions:

• The novel latent mixture-based method provides a powerful tool for analyzing effect-size heterogeneity in meta-analysis.
• This approach overcomes limitations of traditional methods by not needing a priori identification of moderators.
• The method offers a promising solution for addressing complex heterogeneity and inconsistencies in scientific literature.