Selection of Working Correlation Structure and Best Model in GEE Analyses of Longitudinal Data

• Published in: Communications in statistics. Simulation and computation Vol. 36; no. 5; pp. 987 - 996
• Main Authors: Cui, James, Qian, Guoqi
• Format: Journal Article
• Language: English
• Published: Colchester Taylor & Francis Group 30.08.2007; Taylor & Francis
• Subjects: AIC; Distribution theory; Exact sciences and technology; GEE; Longitudinal study; Mathematics; Multivariate analysis; Numerical analysis; Numerical analysis. Scientific computation; Numerical methods in probability and statistics; Parametric inference; Primary 62J12; Probability and statistics; QIC; Sciences and techniques of general use; Secondary 62P10; Statistics; Biometrics; Rank statistic; Correlation; Statistical distribution; Information criterion; QIC. Primary 62J12; Multivariate analysis; Computing; Epidemiology; Parametric method; Medical science; Generalized equation; Distribution function; Repeated measurement; Correlation function; Selection method; Approximation theory; Secondary 62P10; AIC; Likelihood function; Data analysis; Statistical analysis; Model selection; GEE; Longitudinal study; Statistical association; Akaike information criterion; Statistical estimation; Statistical method; Selection problem; Correlation analysis; Quasi likelihood; Numerical simulation; Estimating equation; Software; Maximum likelihood; Application; Computing method
• DOI: 10.1080/03610910701539617
• ISSN: 0361-0918

The Generalized Estimating Equations (GEE) method is one of the most commonly used statistical methods for the analysis of longitudinal data in epidemiological studies. A working correlation structure for the repeated measures of the outcome variable of a subject needs to be specified by this method. However, statistical criteria for selecting the best correlation structure and the best subset of explanatory variables in GEE are only available recently because the GEE method is developed on the basis of quasi-likelihood theory. Maximum likelihood based model selection methods, such as the widely used Akaike Information Criterion (AIC), are not applicable to GEE directly. Pan ( 2001 ) proposed a selection method called QIC which can be used to select the best correlation structure and the best subset of explanatory variables. Based on the QIC method, we developed a computing program to calculate the QIC value for a range of different distributions, link functions and correlation structures. This program was written in Stata software. In this article, we introduce this program and demonstrate how to use it to select the most parsimonious model in GEE analyses of longitudinal data through several representative examples.