|
Abstract:
|
We propose a deletion approach and a closely related weighting approach to handle missing outcomes in generalized estimating equation (GEE) analysis. In deletion approach, we generate artificial observation indicators independent of the outcome, given the covariates, and draw inferences conditioning on the artificial observation indicators. This method provides valid estimate under MAR. The proposed weighting method has more stable weights than those of the inverse probability weighting method of Robins et al. We also apply the results of Robins, Rotnitzky, and Zhao (1994) to improve the efficiency of the proposed weighting estimator. The proposed weighting and improved weighting estimator show an advantage in efficiency compared to the corresponding inverse probability weighting method when the proportions of missing data are high.
|
