Application of generalized estimating equations to a study of in vitro radiation sensitivity

We describe an application of the generalized estimating equation (GEE) method (Liang and Zeger, 1986, Biometrika 73, 13-22) for regression analysis of correlated Poisson data from a split-plot design with a small number of experimental units. As an alternative to the use of an arbitrarily chosen working correlation matrix, we demonstrate the use of GEE with a reasonable model for the true covariance structure among repeated observations within individuals. We show that, under such a split-plot design with large clusters, the asymptotic relative efficiency of GEE with simple (independence or exchangeable) working correlation matrices is rather low. We conclude by summarizing issues and needs for further work concerning efficiency of the GEE parameter estimates in practice.