4. Generalized Estimating Equation (GEE)
Last model is called generalized estimating equations (GEEs) introduced by
Liang and Zeger [9]. They are extension of generalized linear model (GLM) to
longitudinal analysis using quasi-likelihood estimation. A basic premise of GEE
approach is that one is primarily interested in the regression parameter and is not
interested in the variance-covariance matrix of the repeated measures. As such,
generalized estimating equations treat covariance structure as a nuisance and they
are not concerned about variance of each data. They have consistent and
asymptotically normal solutions by relying on the independence across subjects to
estimate constantly the variance of the regression coefficient even when the
assumed correlation structure is incorrect.
GEE has a “working” correlation R of the repeated measurements. This
working correlation matrix is of size n×n because one assumes that there is a fixed
number of time-points n that subjects are measured at. A given subject does not
have to be measured at all n time-points. Each individual’s correlation matrix Ri is
of size ni×ni with appropriate rows and columns removed if ni < n. It is generally
recommended that choice of R should be consistent with the observed correlations.
If the choice of R is incorrect, efficiency such as statistical power is reduced.
However, the loss of efficiency is lessened as the number of subjects gets large.
Some important references in the field of generalized estimating equation can be
found in [1, 3, 6, 7, 9].