Regression with Censored Data: The Synthetic Data and Least Squares Approach

Mai Zhou

Abstract:

Regression with censored data is an important problem. Koul, Susarla and Van Ryzin (1981) proposed a new way of inference in such models with randomly right censored data and showed that it is consistent and asymptotically normally distributed. This paper reviews the many extensions and modifications of this method that appeared since. To further advance the method, we propose in this paper to study an extension of the residuals in such models with censored data. This have potential to be useful in model diagnostic analysis. Examples are given to show the usefulness of the newly defined residuals.

Keywords: Accelerated failure time models, Regression models, Censored data, Residuals

Here is the full postscript text for this technical report. It is 152192 bytes long.