On 1 Dec 2003 15:26:04 -0800, [EMAIL PROTECTED] (Karl) wrote: > Dear group members! > > A trial will be stopped if a new sample size calculation shows that a > requested number of subjects is larger than N (because for more than N > subjects the trial will take too long). If the study is stopped, the > analysis will be done with a significance level of 0.05. If not, an > interim analysis will be done with an adjusted type I error (Pocock > alpha spending function with an overall type I error of 0.05).
Are you introducing us to the *middle* of a discussion? - it sounds rather silly on its own. Unless I am mis-reading, that whole business of doing "adjusted type 1 error" (whatever) is practically nonsense, because you must have already been assured that there's nothing to report. If you look at your present data and conclude, "There is no WAY that we can collect enough data to turn this into a '5% test' " ... you have jumped far beyond the need to show that there is not yet a '5% test'. - What I can comment, further, is that the oversight committees for U.S.-funded health research (NIH, etc.) have tried to take statistical power into account, at various stages in their arguments about funding. That includes, incorporating 'sequential decisions' about trials: (a) if they show great success; or (b) if they show too little success. That is not a bad idea, on the face of it. Unfortunately -- as I glean from an episode or two -- some of these discussions might include nobody influential who understands either statistics or science more generally. <The original, above> impresses me as the reflection of a bad discussion. Since the posting was made from Germany, I don't know whether the concern is with U.S. funding agencies, or European -- or where it is that the concern arises. [ snip, rest] -- Rich Ulrich, [EMAIL PROTECTED] http://www.pitt.edu/~wpilib/index.html "Taxes are the price we pay for civilization." . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
