A simulation experiment was conducted to investigate the behavior of the largest (and sec- ond, third and fourth largest) cluster Cook’s distance statistics for marginal mean and marginal association parameters, separately, in the context of an application of the orthogonalized residu- als estimating equations formulation of alternating logistic regressions. The behavior of extreme Cook’s distance statistics for the association parameters was responsive to random contamina- tion, whereby binary responses were transposed with equal probability, and without regard to cluster membership. Specifically, for moderate (20) to large (50) cluster sizes, the distri- bution ”shifted to the right” with increasing levels of contamination; however, shifting was not observed for small cluster sizes (5). For other situations - association parameters under a cluster-concentrated contaminated model, and mean parameters, generally - the behavior of the extreme Cook’s distance did not behave in the anticipated manner.
Biostatistics | Longitudinal Data Analysis and Time Series | Numerical Analysis and Computation | Statistical Models
By, Kunthel; Preisser, John; and Qaqish, Bahjat, "A Simulation Experiment to Investigate the Distributional Behavior of Extreme Cook's Distance for GEE to Models with Contaminated Binary Responses" (July 2008). The University of North Carolina at Chapel Hill Department of Biostatistics Technical Report Series. Working Paper 9.