An aligned rank test for censored data from randomized block designs

SUMMARY The aligned rank test of Hodges & Lehmann (1962) is extended to censored survival data collected in matched pairs or randomized blocks. The test statistic for matched pairs is a direct extension of the original aligned rank test, whereas the test statistic for use with block size k > 2 is constructed from the paired data statistics comparing all possible pairs of treatments. The efficiency of recovery of interblock information with censored matched pairs is examined by calculating the asymptotic efficiency of the proposed test relative to the sign test under a Weibull model with a single overall censoring time. The powers of these two tests and two other tests are compared under a more general model in a Monte Carlo study. A numerical example illustrating the technique is included.