Maximum likelihood estimation and a specification test for non-normal distributional assumption for the accelerated failure time models (Q789140)

This paper has derived a specification test for any specified distributional assumption for the truncated and censored regression models. The test is a scoring test. Any random variate with specified non normal distribution can be transformed into a normal random variate. The transformed normality is tested against the general Pearson family of distributions. Simple and intuitive interpretations in terms of moments can be provided for the test statistics. The derived test is useful for testing distributional assumptions in the accelerated failure time regression models for the analysis of duration data with time invariant explanatory variables.