|
Abstract:
|
Under coarsening at random, doubly robust estimators exist for the parameters of semiparametric models for continuous time multistate longitudinal data processes in the presence of dependent censoring explainable by high dimensional time-dependent covariate processes. In this talk, we describe new multiply (i.e., triply, quadruply, etc.) robust estimators based on higher order influence functions. These estimators are higher order U statistics. We posit a number of models for the joint distribution of the full data and for the censoring mechanism such that if any one of the many models is correct, we obtain a consistent asymptotically normal estimator of the parameter of interest. The more models one uses the higher the required order of the U-statistic.
|
