ON LARGE DEVIATIONS AND BAHADUR EFFICIENCY OF LINEAR RANK STATISTICS.
Abstract
A new, simpler proof of the main theorem of AD-659 994 is presented. This theorem enabled one to approximate the sample size needed by a test with type I error alpha to achieve type II error beta at a fixed alternative hypothesis. This approximation involved a quantity called the exact slope, an information number which is, very roughly, something like the channel capacity of the real world-test-statistician 'channel'. Theorem 4 provides a method for numerically calculating the exact slope; such calculations are carried out for two-sample Wilcoxon, normal-scores and median tests against normal, logistic and double exponential shift alternatives. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Apr 07, 1969
- Accession Number
- AD0686701
Entities
People
- George G. Woodworth
Organizations
- Stanford University