Generalized Estimating Equations

Abstract Generalized linear models cannot be routinely applied to clustered data (e.g., repeated measures on an individual) owing to the correlation among responses within a cluster. This has prompted the development of extensions of these models to the clustered data setting. Generalized estimating equations (GEEs) are most widely used for estimation and inferences about regression parameters in a semiparametric extension of generalized linear models to cluster‐correlated data known as marginal models . In a marginal model, the main focus is on a regression model relating the mean response to a set of covariates; the within‐cluster covariance among the responses is regarded as a nuisance feature of the data. Without requiring assumptions about the joint distribution of the responses within a cluster, the GEE approach yields consistent estimators of the regression parameters even when the within‐cluster covariance has not been correctly specified. This property of the GEE method accounts for its widespread use in the analysis of longitudinal or cluster‐correlated data, especially when the response variable of interest is discrete.