An Adaptive R-Estimate
Abstract
In 1962 Hajek proposed a test for location which was uniformly asymptotically fully efficient over a large class of distributions. Van Eeden subsequently derived the asymptotic theory for the corresponding R-estimate. Many authors have expressed reservations with regard to the small sample performance to be expected from this approach. In contrast to the methods of Hajek and Van Eeden, the procedure proposed here uses the entire data set to estimate the score function of the locally most powerful rank test. Using nearest neighbor density estimation methods, an estimate of the score function for the asymptotically most powerful grouped rank test is constructed. This function is itself a step function approximation to the score function for the locally most powerful rank test. Large sample distribution and optimality results are obtained for both the adaptive rank test and the corresponding R- estimate of location. Small sample monte-carlo results are provided.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 15, 1980
- Accession Number
- ADA096768
Entities
People
- Michael L. Cohen
Organizations
- Stanford University