On the extended lack-of-fit tests for the censored regression model

Abstract

In this paper, we improve a class of lack-of-fit tests for the censored regression model by considering the new classes of weight functions. Some processes assigning mass at all observations are used as components of data-dependent weight functions in the tests. The weight functions incorporated with these processes are reflective of the influence of the censoring because they jump at censored data points as well as non-censored data points. We also consider the weight functions involving the censoring proportion and the product-limit estimates. The tests with some new weight functions are asymptotically normal under the regression model. Simulation studies show that the lack-of-fit tests of the model perform very well for some new weight functions. The model-checking procedures are illustrated in an application.

Keywords:

• Accelerated life regression model
• Lack-of-fit tests
• Weighted log-rank estimating functions
• Survival analysis