Small-sample adjustments in using the sandwich variance estimator in generalized estimating equations.

Area of Science:

• Statistics
• Biostatistics
• Econometrics

Background:

• Generalized Estimating Equations (GEE) are standard for correlated data.
• The sandwich estimator is commonly used for variance estimation in GEE.
• The Wald chi-squared test, relying on the sandwich estimator, can exhibit inflated Type I errors in small samples.

Purpose of the Study:

• To propose and evaluate novel approximate t- or F-tests for GEE that address small-sample limitations.
• To improve the accuracy of statistical inference in regression analyses with correlated response data.

Main Methods:

• Development of approximate t- or F-tests incorporating the variability of the sandwich estimator.
• Simulation studies to assess the performance of the proposed tests against existing methods.
• Comparison with alternative approaches that modify the sandwich estimator.

Main Results:

• The proposed t- or F-tests maintain Type I error rates no larger than the Wald chi-squared test.
• Simulation results confirm the satisfactory performance and improved accuracy of the new tests.
• The new approach demonstrates advantages over other modified sandwich estimator methods.

Conclusions:

• The proposed t- or F-tests provide a more reliable method for hypothesis testing in GEE, especially for small sample sizes.
• These findings have direct implications for constructing more accurate Wald-type confidence intervals and regions.