SIZE AND POWER OF SOME TESTS UNDER EXPERIMENTAL RANDOMIZATION.
IOWA STATE UNIV OF SCIENCE AND TECHNOLOGY AMES
Pagination or Media Count:
This report deals with an investigation of the behavior of competitive tests for the paired design under experiment randomization. The tests considered are the Fisher randomization test R, the Wilcoxon paired test W, the Sign test S, and the normal law F test F. The paired design with N pairs results in the drawing at random of one of 2 to the Nth power possible plans and the performance of the tests in this population of repetitions is considered. Theoretical and Monte Carlo results are presented. The general conclusion is that the R and W tests are definitely superior to the other two tests. The R test is slightly superior to the W test in having a greater number of possible significance levels, but at sizes of test possible with both, the powers of the two tests are very close. The relationship of the results to asymptotic theory is discussed. Author
- Statistics and Probability
- Test Facilities, Equipment and Methods