Assessing the censored linear regression model using martingale approximation

Abstract

This paper presents a simulation-based procedure to assess the adequacy of the censored linear regression model. The procedure is based on a stochastic process whose null distribution can be approximated by a mean zero Gaussian process in terms of a martingale. A comparison of the stochastic process and its approximations leads to the construction of a lack-of-fit test for the linear regression model with censored failure time data. The test is asymptotically consistent against a general departure from the model. Simulations are performed to investigate the size and power properties of the test. For illustration, the procedure is applied to two real-world applications.