On Tue, 27 Apr 2004 10:22:20 +0200, Christoph Int-Veen <[EMAIL PROTECTED]> wrote:
> Dear newsgroup, > > i want to design a hypothetical clinical trial with two main questions > and a double randomization design, which means that each patient will be > randomized two times for treatment and there is no interaction between > treatment A and B. (so no 2x2 factorial design) I don't understand why you say you have no 2x2 factorial. Is it: Patients are randomized to A or B; later, to C or D. Unless you are saying that potential interaction is totally uninteresting, I would figure on looking at the interaction. Especially because: the *starting* score for C or D may depend on A or B, if one is better than the other; > As i have to main questions to be tested on the same data i must adjust > my alpha because of multiplicity issues. Well, there can be discussions about that. 'same data' is hardly a useful guideline. I would argue that it fails in both directions. > My question is: The sample size of the trial will be determined by the > overall alpha and beta, or by the individual alpha and beta (e.g. the > two tests (main questions))????? The sample size depends on the 'nominal alpha' and beta. If your nominal alpha were 0.025 because you cut 5% in half owing to having two tests, then you would have to look for the N demanded by the 0.025 test. -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
