Semiparametric analysis of recurrent discrete time data with competing risks

The regression analysis of the cumulative incidence function on recurrent discrete time data with competing risks has not been widely investigated. We propose semiparametric analysis for regression modelling of the cumulative incidence function for recurrent discrete time competing risks data. The maximum likelihood inferences are developed for the estimation of the model parameters in a discrete time competing risks model, which are based on a working independence likelihood. We derive the sandwich variance estimator to account for correlations between multiple discrete times within each subject. Simulation studies show that the procedures perform well. The proposed methods are illustrated with two applications, a study of contraceptive use in Indonesia and data on unemployment in Germany.