A caution in the use of multiple criteria for selecting working correlation structure in generalized estimating equations. Issue 3 (4th March 2023)

• Record Type: Journal Article
• Authors: Sultana, Abida; Lipi, Nasrin; Jaman, Ajmery
• Is Part Of: Communications in statistics. Volume 52:Issue 3(2023)
• Page Start: 980
• Page End: 992
• Publication Date: 2023-03-04
• Subjects: Correlation structure selection -- Determinant-based criterion -- Gaussian pseudo-likelihood -- Generalized estimating equations -- Multiple criteria -- Trace of empirical covariance matrix; Mathematical statistics -- Periodicals; Mathematical statistics -- Data processing -- Periodicals; Digital computer simulation -- Periodicals; 519.5
• DOI: 10.1080/03610918.2021.1871924
• Languages: English
• ISSNs: 0361-0918

Abstract

Abstract: The generalized estimating equations (GEE) procedure is a common approach for regression analyses of correlated responses, which is common for data obtained in longitudinal studies. Selecting a working correlation structure has been a momentous topic in longitudinal data analysis using GEE. Although consistent and asymptotically normal estimates can be obtained with misspecified working correlation structure, an appropriate selection of the correlation structure improves the efficiency of the regression estimators. Numerous criteria have been proposed to select the best working correlation structure in GEE. Besides single criterion, some authors suggested different approaches of using multiple criteria. In this paper, a comparison of available criteria is done along with different combinations of selected multiple criteria. We investigated their performance using both simulated and real data. Simulation results indicate that data analysts should be cautious in selecting multiple criteria to be combined. Practical recommendations are also given for choosing a working correlation structure in GEE.