A generalized Kruskal-Wallis test for comparing K samples subject to unequal patterns of censorship

摘要: SUMMARY A generalization of the Kruskal-Wallis test, which extends Gehan's Wilcoxon's is proposed for testing equality K continuous distribution functions when observations are subject to arbitrary right censorship. The censoring variables allowed differ different populations. An alternative statistic use distributions may be assumed equal. These statistics have asymptotic chi-squared under their respective null hypotheses, whether regarded as random or fixed numbers. Asymptotic power and efficiency calculations made numerical examples provided. comparing two populations has been by Gehan (1965a) Mantel (1967), well (1965b), considered a further case arbitrarily restricted observation, left Both these authors base on permutation statistic, conditional observed pattern combined sample. However, this model inapplicable there differences in For instance, medical follow-up studies, where procedure so far found its widest application, would happen if had study lengths time. This paper censored comparison probability relevant here large sample framework models: Model I, corresponding unconditional censorship; II, considers times Since vary with population, also extended unequal I theoretical distributions; II they empirical. Besides providing hypothesis against general alternatives, shows how single degrees freedom partitioned discriminating specific hypotheses. Several investigators (Efron, 1967) pointed out that test not most efficient certain parametric alternatives modifications increase power. below demonstrate criticisms apply equally here. Hopefully some suggest can likewise eventually generalized