Statistics for the Two-Sample Survival Analysis Problem Based on Product Limit Estimators of the Survival Functions.
Abstract
A class of statistics for the two-sample survival analysis problem is introduced. The SW-statistic may be written as the integral of a weighted difference in the estimated survival functions, the integral being with respect to Lebesgue measure on time. Since the integral is with respect to real-time the statistics are not generalized rank-statistics. However, with an appropriate choice of weight function they are non-parametric in the sense that weak convergence is guaranteed for any underlying configuration of survival and censoring distributions. Asymptotic distribution theory under the null is derived. Consistency under a fixed alternative is shown. Efficacy expressions under natural sequences of local alternatives are given and an expression for the most efficient weight function is developed. The asymptotic efficiencies of some specific SW-statistic under the proportional hazards alternative are examined. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Nov 01, 1986
- Accession Number
- ADA176956
Entities
People
- Margaret O'sullivan
- Thomas R. Fleming
Organizations
- University of North Carolina at Chapel Hill