Combining the Advantages of One-Sided and Two-Sided Test Procedures for Comparing Several Treatment Effects
Tetsuhisa Miwa, Tony Hayter · Journal of the American Statistical Association · 1999
We consider the one-way layout for comparing k treatment effects μ i , 1 ≤ i ≤ k. Hayter earlier proposed a one-sided studentized range test (OSRT) for testing the null hypothesis H μO: μ1 = μ2 = ··· = μ k against the simple ordered alternative HA : μ1 ≤ μ2 ≤ ··· ≤ μ k with at least one strict inequality. The OSRT gives a set of simultaneous confidence intervals for all ordered pairwise differences μ j - μ i , i 0, but provides two-sided finite confidence intervals if the sample averages are all far apart.