Meta-analysis using effect size distributions of only statistically significant studies.

Area of Science:

• Psychology
• Biostatistics

Background:

• Publication bias can distort meta-analysis results, particularly in psychology studies with small sample sizes.
• Overestimation of effect size is a common issue in traditional meta-analysis due to publication bias.

Purpose of the Study:

• Introduce p-uniform, a novel meta-analysis method designed to address publication bias.
• Enable simultaneous testing for publication bias, effect size estimation, and null-hypothesis testing.

Main Methods:

• The p-uniform method is presented as a straightforward technique requiring no additional data on missing studies.
• It avoids complex assumptions or pre-application choices, simplifying its use.

Main Results:

• Simulations indicate p-uniform surpasses trim-and-fill and test of excess significance (TES) when publication bias is present and effect sizes are homogenous or slightly heterogeneous.
• The method was applied to a meta-analysis on infant habituation and cognitive ability (IQ).

Conclusions:

• P-uniform is a robust technique for assessing publication bias and estimating population effects in fixed-effect meta-analyses.
• It serves as a valuable tool for sensitivity analysis in drawing inferences about publication bias.