Geometric standardized mean difference and its application to meta-analysis

2025-06-04

AI Summary

Conventional standardized mean difference (SMD) estimators rely on the equal-variance assumption and exhibit substantial bias under heteroscedasticity; standard arithmetic-mean variance approaches fail to yield unbiased estimates. Method: We propose a novel SMD paradigm based on the geometric mean of group variances, introducing the first Cohen-type and Hedges-type estimators specifically designed for unequal variances. We rigorously derive their asymptotic distributions, small-sample bias corrections, and confidence intervals. Contribution/Results: Monte Carlo simulations demonstrate that the Hedges-type estimator substantially reduces bias and mean squared error while improving confidence interval coverage. Empirical meta-analyses further confirm its robustness and practical utility. This framework overcomes the restrictive equal-variance assumption, providing a theoretically rigorous and computationally feasible benchmark for effect-size standardization in heteroscedastic settings.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Intelligent Robots: State Estimation
• Search and Optimization: Metareasoning and Metaheuristics

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

The standardized mean difference (SMD) is a widely used measure of effect size, particularly common in psychology, clinical trials, and meta-analysis involving continuous outcomes. Traditionally, under the equal variance assumption, the SMD is defined as the mean difference divided by a common standard deviation. This approach is prevalent in meta-analysis but can be overly restrictive in clinical practice. To accommodate unequal variances, the conventional method averages the two variances arithmetically, which does not allow for an unbiased estimation of the SMD. Inspired by this, we propose a geometric approach to averaging the variances, resulting in a novel measure for standardizing the mean difference with unequal variances. We further propose the Cohen-type and Hedges-type estimators for the new SMD, and derive their statistical properties including the confidence intervals. Simulation results show that the Hedges-type estimator performs optimally across various scenarios, demonstrating lower bias, lower mean squared error, and improved coverage probability. A real-world meta-analysis also illustrates that our new SMD and its estimators provide valuable insights to the existing literature and can be highly recommended for practical use.

Problem

• Proposes geometric SMD for unequal variances in meta-analysis
• Develops Cohen-type and Hedges-type estimators for new SMD
• Demonstrates superior performance via simulations and real-world application

Innovation

• Geometric averaging for unequal variances
• Cohen-type and Hedges-type estimators introduced
• Improved bias and coverage probability demonstrated