An Exact Bayesian Model for Meta-Analysis of the Standardized Mean Difference with Its Simultaneous Credible...

Area of Science:

• Behavioral Research
• Statistical Methodology
• Meta-Analysis

Background:

• Bayesian methodology is gaining traction in behavioral research due to its probabilistic inference.
• However, its application in meta-analysis is limited, with existing Bayesian hierarchical models lacking thorough performance evaluation.
• Conventional models show significant issues when applied to meta-analysis of standardized mean difference.

Purpose of the Study:

• To evaluate the performance of two common Bayesian meta-analysis models for standardized mean difference.
• To introduce a novel Bayesian model addressing identified limitations and enhancing current Bayesian meta-analysis.
• To develop a computational approach for simultaneous credible intervals of summary effect and heterogeneity.

Main Methods:

• Evaluation of two existing Bayesian meta-analysis models for standardized mean difference.
• Development and introduction of a new Bayesian model with enhanced features.
• Simulation studies comparing the new model against existing ones under realistic conditions.
• Application of a novel computational approach for joint credible intervals.

Main Results:

• The new Bayesian model significantly outperforms existing models in meta-analysis of standardized mean difference.
• The proposed model demonstrates enhanced statistical properties and addresses limitations of current approaches.
• The computational approach effectively captures joint uncertainty in summary effect and heterogeneity.

Conclusions:

• The novel Bayesian meta-analysis model offers improved performance and robustness for behavioral research.
• The developed approach strengthens inferences regarding summary effects by addressing parameter uncertainty.
• This work advances Bayesian meta-analysis methodology, offering practical benefits for researchers.