PERCENTILE MODIFICATIONS OF TWO SAMPLE RANK TESTS,
Abstract
A simple method is presented for increasing the efficiency of rank tests relative to the best tests for samples from normal distributions without using complicated scoring systems such as the normal scores tests. Two numbers p and r (0 is less than p, r is less than 1) are selected and then scored, with simple weights, the date in the upper pth and lower rth percentiles of the combined sample. This proposal is motivated by the fact that the optimum rank test for normal distributions places more weight on the extreme ranks then on the central ones. The optimum values of p and r depend on the populations sampled. For most distributions, the efficiency, is near its maximum in a large neighborhood of the optimum choice of p and r. An advantage of the proposed test is that prior knowledge of any symmetry properties of the populations can be incorporated into the selection of p and r. The criterion used to compare tests is Pitman's efficiency. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 11, 1964
- Accession Number
- AD0433869
Entities
People
- Joseph L. Gastwirth
Organizations
- Stanford University