XGrid Incomplete data regression
The development of methods for the estimation of parameters in regression models where some subjects may have only partially observed covariates is an important area of research. While a variety of ad-hoc approaches have been proposed (e.g. indicator methods) these have a number of limitations. While other principled approaches have been suggested, we consider extensions of the general maximum likelihood model of Ibrahim to develop computationally expensive yet flexible and practical methods for missing predictor models. In addition, we consider the robustness of alternative approaches (chained equations in multiple imputation) as an alternative approach.