Regressione per funzioni di rischio con covariate tempo-dipendenti: una proposta basata sulla massima verosimiglianza locale

Abstract

The aim of this paper is to propose a flexible method to explore the possible relationship between the hazard rate function associated to a duration time T* and a time-dependent covariate X(t). This method is based on a local likelihood approach that does not require an explicit specification of the functional form of this relationship. In order to automatically select the bandwidth, the variable span smoother (Friedman, 1984), also called supersmoother, is adapted to the context of duration data analysis.