Difference of means test with censored data

Abstract

For the two-sample problem with censored data, a difference of means test is formulated. We use simulation to compare this test to the popular log-rank test under various distributions. It is seen to perfom better than the logrank test for some nonproportional hazard alternatives, while performing roughly equivalent for some proportional hazard alternatives where the log–rank test is known to be asymptotic most powerful.

Keywords:

• two sample test
• censored data
• mean lifetime