Kaplan-Meier V- and U-statistics (Q2180082)

The authors study bivariate U-statistics and von Mises statistics for censored data, that often appears in survival analysis. These statistics can be expressed as functionals of the empirical distribution function, which is replaced by the Kaplan-Meier-estimator to obtain a consistent estimator. Under moment conditions also involving the distribution of the censoring time, the authors show a central limit theorem for statistics with nondegenerate kernel functions, and they show the convergence to a linear combination of Chi-square distributed random variables for degenerated kernel functions. The theory is then applied to the Cramér-von Mises-statistic for testing hypothesis on the distribution function. This statistic might be or might not be degenerate under the hypothesis, depending on the censoring. Some simulation results are given for this application.