xtgeebcv: A command for bias-corrected sandwich variance estimation for GEE analyses of cluster randomized trials

Area of Science:

• Biostatistics
• Public Health Research Methods
• Social Science Statistics

Background:

• Cluster randomized trials (CRTs) are prevalent in public health, education, and social sciences.
• Individual-level outcome analysis in CRTs requires accounting for within-cluster similarity.
• Generalized estimating equations (GEE) are a common analysis method for such data.

Purpose of the Study:

• To address the finite-sample bias in standard errors from GEE analysis when a small number of clusters are randomized.
• To introduce a practical tool for Stata users to implement bias corrections.
• To provide guidance on selecting appropriate bias correction methods.

Main Methods:

• Review and description of popular bias-corrected standard error methods for GEE.
• Development and introduction of the Stata command 'xtgeebcv'.
• Demonstration of 'xtgeebcv' usage with practical examples.

Main Results:

• Standard GEE sandwich variance estimators are biased with few clusters, leading to inflated Type I errors.
• 'xtgeebcv' provides accessible implementation of finite-sample bias corrections for GEE standard errors in Stata.
• The command facilitates more accurate statistical inference in small-cluster CRT analyses.

Conclusions:

• Accurate standard error estimation is critical for valid inference in cluster randomized trials.
• 'xtgeebcv' enhances the utility of GEE for researchers working with small numbers of clusters.
• Further research can refine bias correction methods and the 'xtgeebcv' command.