A Statistical Method for Synthesizing Mediation Analyses Using the Product of Coefficient Approach Across Multiple...

Area of Science:

• Statistics
• Biostatistics
• Psychometrics

Background:

• Mediation analysis requires substantial sample sizes for adequate statistical power.
• Combining data across trials is a practical solution to increase power for mediation analysis.
• Existing meta-analytic approaches may not be optimal for mediation analysis.

Purpose of the Study:

• To propose a novel method for estimating mediation effects by combining results from multiple trials.
• To develop an R program for implementing the proposed statistical method.
• To provide more accurate inference for mediation analysis compared to standard meta-analytic techniques.

Main Methods:

• A random effects model is used to combine regression coefficients (paths 'a' and 'b') and their standard errors from multiple trials.
• The method estimates marginal means for mediation paths and the between-trial variance-covariance matrix.
• A marginal likelihood approach with Monte Carlo confidence intervals is employed for inference.

Main Results:

• The proposed method provides accurate estimation of combined mediated effects and confidence intervals.
• This approach yields more precise inference than traditional meta-analytic methods.
• The R program facilitates the practical application of the method to real-world data.

Conclusions:

• This method offers a statistically sound and more accurate way to perform mediation analysis when combining data from multiple trials.
• The developed R program supports researchers in applying this advanced technique.
• Recommendations are provided for the optimal use of this method in various research settings.