Robust variance estimation in small meta-analysis with the standardized mean difference

Area of Science:

• Biostatistics
• Statistical Modeling
• Meta-Analysis

Background:

• Conventional random-effects models in meta-analysis often use large sample approximations.
• These models may produce confidence intervals that are too narrow for small sample sizes, affecting accuracy.
• Accuracy is influenced by sample size configuration, heterogeneity, and the number of studies.

Approach:

• Introduce two novel variance estimators designed for improved small sample properties.
• Investigate the effectiveness of various degrees of freedom adjustments for confidence interval computation.
• Utilize simulation studies to rigorously evaluate the performance of the proposed methods.

Key Points:

• Demonstrates inadequacy of conventional random-effects models in small sample meta-analysis.
• Proposes alternative variance estimators with superior small sample performance.
• Evaluates degrees of freedom adjustments for enhanced confidence interval accuracy.

Conclusions:

• The developed variance estimators and degree of freedom adjustments offer improved accuracy for small sample meta-analyses.
• These advancements address limitations of conventional methods, leading to more reliable summary effect estimates.
• Simulation results support the effectiveness of the proposed techniques in small sample scenarios.