Correcting for publication bias in a meta-analysis with the p-uniform* method

Area of Science:

• Statistics
• Biostatistics
• Meta-analysis methodology

Background:

• Publication bias is a significant threat to meta-analysis validity, often leading to overestimated effect sizes.
• Existing methods like p-uniform and random-effects models may not adequately address publication bias and between-study variance.

Purpose of the Study:

• To introduce and evaluate p-uniform*, a novel method designed to generalize and improve upon the p-uniform method for addressing publication bias in meta-analyses.
• To compare the statistical properties of p-uniform* against p-uniform, three-parameter selection models (3PSM), and random-effects models.

Main Methods:

• Development of p-uniform*, an enhanced statistical method for meta-analysis.
• Comparative analysis of statistical properties including efficiency, effect size estimation, and detection of between-study variance.
• Re-analysis of two published meta-analyses to demonstrate practical application and impact.

Main Results:

• P-uniform* demonstrated comparable and often superior performance to 3PSM, outperforming p-uniform and random-effects models in the presence of publication bias.
• Both p-uniform* and 3PSM showed good performance in estimating average effect size and between-study variance with ten or more studies, especially when publication bias was not extreme.
• P-uniform* utilizes a more parsimonious model compared to 3PSM.

Conclusions:

• P-uniform* offers an improved and more efficient approach to handling publication bias and estimating between-study variance in meta-analyses.
• The method provides valuable tools for applied researchers, with available R code and a web application for practical implementation.