Marginal Models: Generalized Estimating Equations (GEE)

Marginal models have a three-part specification in terms of a regression model for the mean response, supplemented by assumptions concerning the variance of the response at each occasion and the pairwise within-subject association among the responses. The generalized estimating equations (GEE) approach provides a convenient alternative to maximum likelihood estimation. This chapter describes the GEE approach for estimating the parameters of marginal models. It briefly reviews some useful residual diagnostics for assessing the fit of marginal models. The chapter presents three case studies that illustrate the application of marginal models to longitudinal data. It considers estimation and aspects of interpretation of timevarying covariates in marginal models. The GEE approach is based on the concept of "estimating equations" and provides a very general and unified approach for analyzing correlated responses that can be discrete or continuous.