Chapter 9 - Generalized estimating equations (GEEs) models

Abstract

Chapter 9 is devoted to the descriptions of generalized estimating equations (GEEs). First, I summarize the basic specifications and inferences of the original GEE model, including the construction of the working correlation matrix and the development of the quasi-likelihood information criteria. Some GEE advances are then introduced, consisting of Prentice’s GEE approach, Zhao and Prentice’s GEE method (GEE2), and the GEE models on odds ratios. Next, I compare the conditional and the marginal regression models with the argument that the application of GLMMs is a more suitable perspective than GEEs to predict marginal means in the analysis of non-normal longitudinal data. An empirical illustration is provided to display how to use GEEs in longitudinal data analysis. The analytic results from the illustration provide strong empirical evidence that modeling a complex covariance structure does not necessarily improve the quality of parameter estimates and the goodness-of-fit statistic in GEEs.