Tests for censored paired data

The performance of tests for the equality of two lifetime distributions in paired data is examined by simulation within the context of a model proposed by Owen, Sinha and Capozzoli. Previous work is extended by allowing censored data. The tests compared are likelihood ratio, Wald, score and Owen’s MM test. The Wald test is inferior to the others regarding the accuracy of nominal significance levels. The tests have similar powers. It is recommended that the likelihood ratio test be used with asymptotic χ2χ2 percentage points. All tests perform badly when censoring affects only one of the two variables, so this must be avoided in the study design.