An improved uniformly more powerful exact Fisher-Hayter pairwise comparisons procedure.

Pairwise comparison is a very common multiple comparison problem. It is known that Fisher's LSD test does not control the familywise error rate (FWER) when there are more than three groups to be compared. Improved testing strategies include the Tukey-Kramer (TK) test that eliminates the F-test step and the two-step Fisher-Hayter (FH) test which requires a significant F-test. We propose a modified FH-test that is uniformly more powerful than the original version and relies on exact size α test under the balanced model. We provide simulations to show that the new procedure is preferred to the FH-test and the TK-test.