Analyses of associations with censored data by generalized Mantel and Breslow tests and generalized Kendall correlation coefficients


  • Dr. M. Schemper

    Corresponding author
    1. Arbeitsgruppe für Biometrie und Dokumentation der I. Chirurgischen Universitätsklinik in Wien
    • I. Chirurgische Universitätsklinik Alserstraße 4, 1090 Wien Austria
    Search for more papers by this author


Two nonparametric procedures are proposed for the problem of testing association between two continous variables, when one is subject to arbitrary censoring. The main advantages of the two tests—generalizations of MANTEL'S (1066) and BRESLOW'S (1970) k-sample tests for censored data—are: accurate control over the size of the test, no restrictive assumptions concerning censoring mechanisms, theoretical properties similar to those already explored for Mantel's and Breslow's tests, close relationship to descriptive measures of association, namely generalized Kendall correlation coefficients, to which part of the paper is devoted. Finally stratified analyses are discussed and calculated examples presented.