Detecting and correcting for publication bias in meta-analysis - A truncated normal distribution approach

Area of Science:

• Statistics
• Biostatistics
• Meta-analysis

Background:

• Publication bias poses a significant threat to the validity of meta-analysis.
• Existing methods for addressing publication bias primarily rely on funnel plot techniques or selection models.

Purpose of the Study:

• To propose novel parametric solutions for detecting and correcting publication bias by framing it as a truncated distribution problem.
• To develop methodologies for estimating the overall effect size and the extent of publication bias, considering small effect size and large p-value scenarios.

Main Methods:

• Formulated publication bias as a truncated distribution problem, proposing new parametric solutions.
• Developed estimators for overall mean and truncation proportion under fixed- and random-effects models using maximum likelihood estimation and method of moments.
• Compared the proposed methods with the non-parametric Trim and Fill method via extensive simulation studies.

Main Results:

• The proposed parametric methods, based on truncated normal distribution, demonstrated consistent performance in detecting and correcting publication bias.
• These methods effectively estimated the underlying overall effect size and the severity of publication bias under various conditions.
• Performance was evaluated through extensive simulations, comparing favorably against the Trim and Fill method.

Conclusions:

• Parametric methods utilizing truncated distribution offer a robust approach to address publication bias in meta-analysis.
• The developed methodologies provide reliable estimation of effect sizes and bias severity, enhancing the validity of meta-analytic conclusions.
• These findings suggest a promising alternative to existing non-parametric approaches for handling publication bias.