A MONTE CARLO STUDY OF MULTIPLE COMPARISON TECHNIQUES.
NAVAL POSTGRADUATE SCHOOL MONTEREY CALIF
Pagination or Media Count:
A study was conducted on the multiple comparison methods presented by Scheffe, Tukey, Student-Newman-Keuls, and Duncan under the experimental situation in which all populations were normal with equal variances and all means but one were equal. The characteristics of all four test procedures were compared for the case of multiple comparisons of pairs of means. These tests were conducted both with and without the prior performance of an analysis of variance. The Tukey and Scheffe procedures were compared in tests of linear combinations of three means. Estimates were made of the power of the tests and of Type I error rates under both the null and alternate hypotheses. Scheffes method was found to be too conservative for pairwise comparisons of means, but it was to be preferred over Tukeys method for combinations of more than two means. Duncans method was the most powerful test of pairwise comparisons, but it maintained little control over one kind of Type I error. The S-N-K procedure showed a good balance between power and control of Type I errors. Author
- Statistics and Probability