LONG-RUN PROPORTIONAL HAZARDS MODELS OF RANDOM CENSORSHIP

In the Koziol-Green or proportional hazards model of random censorship, the survival distribution (GBAR) or the censoring variable is a power function of the survival distribution FBAR of the lifetime, with the exponent in the power being the censoring parameter. In this paper, we propose a more general (semiparametric) model G(t)BAR = F(t)theta-BAR L(F(t))BAR, where L is some slowly varying function. A specific parametric example of an L-function is found to perform well for many well-known survival data sets. In this case estimators of the parameters and the survival function FBAR are proposed and asymptotics are given. This model is also compared with a recent proposal of Pena and Rohatgi (1989).