In article <[EMAIL PROTECTED]>, dennis roberts <[EMAIL PROTECTED]> wrote:
>actually, p values are rather useless (i am almost prepared to say >"useless") since, it would be the RARE case when the null is REALLY exactly >true >thus, in 99.9999% of the cases ... we KNOW the null is not true so, setting >some cutoff for rejection and then actually rejecting the null ... what has >this added to our knowledge? >and, p values don't speak to the notion of the null being "approximately" true I do not see any way to do this other than a decision theoretic approach; how much weight do we give to a given action in a given state of nature? This is what an approach to consistent behavior yields; if one wants to call the weight loss time prior probability, go ahead, but this is not necessary. AFAIK, I know of one paper on this, written by me more than 30 years ago. I can add a little; if the region in which one wants to accept the null is small compared with the behavior of the likelihood function, it can be well approximated by a decision approach to testing the point null, but not the usual one. If it is very large compared to that same behavior, just estimate the parameter and act accordingly. Unfortunately, the in-between part is rather large, and the procedure to be used depends rather heavily on the precise form of certain assumptions, which the user might have great difficulty in making. -- This address is for information only. I do not claim that these views are those of the Statistics Department or of Purdue University. Herman Rubin, Deptartment of Statistics, Purdue University [EMAIL PROTECTED] Phone: (765)494-6054 FAX: (765)494-0558 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
