Identification of local sparsity and variable selection for varying coefficient additive hazards models

Varying coefficient models have numerous applications in a wide scope of scientific areas. Existing methods in varying coefficient models have mainly focused on estimation and variable selection. Besides selecting relevant predictors and estimating their effects, identifying the subregions in which varying coefficients are zero is important to deeply understand the local sparse feature of the functional effects of significant predictors. In this article, we propose a novel method to simultaneously conduct variable selection and identify the local sparsity of significant predictors in the context of varying coefficient additive hazards models. This method combines kernel estimation procedure and the idea of group penalty. The asymptotic properties of the resulting estimators are established. Simulation studies demonstrate that the proposed method can effectively select important predictors and simultaneously identify the null regions of varying coefficients. An application to a nursing home data set is presented.