ABSTRACT
For Tukey's MCA method discussed in Section 5.1.1, if the confidence intervals for all m — p,j cover their respective true values and have width no more than S = 2J*, that is, the event
A = {fii — HJ £ fa — fij ± |<7*|<Jv/2/n for all i ^ j and |g*|<7^/2/n < S*} occurs, then 1. for all Hi — Hj > J, the correct and useful significant directional difference
assertion /^ > p,j will be made; 2. for all p>i — p,j « 0, the correct and useful practical equivalence assertion
—8<jjii — t*,j < 6 will be made. Theorem C.I.I Given an error rate a, if the sample size n satisfies
where again $ is the standard normal distribution function, 7 is the density of a f a , and u = T/n/2(6*/a)/\q*\, then P(A) > 1 - /9. Proof.
