Abstract: In the standard accelerated lifetime model, log lifetime is, up to noise, a lin ear function of a vector of covariables. In the present paper this model is extended to admit general nonlinear functional relationships. A Weighted Least-Squares estimator for the unkown true parameter is proposed. Consistency and asymptotic normality are shown when the lifetimes are subject to censoring but the covariables are known. The accuracy of the procedure is demonstrated in a small sample simulation study under various degrees of censoring.
Key words and phrases: Nonlinear regression, random censorship, weighted least-squares.