Message-ID: <Pine.LNX.4.43.0903220029170.766@hymn32.u.washington.edu>
Date: 2009-03-22T07:29:17Z
From: Thomas Lumley
Subject: Fisher test problem
In-Reply-To: <1747306602.21617101237691301627.JavaMail.javamailuser@localhost>
On Sat, 21 Mar 2009 markleeds at verizon.net wrote:
> by definition, the one tailed p-value has to be <= 0.5 so there is still
> something wrong with your OpenEpi calc.
I think that's a little strong.
Firstly, a one-tailed test in a pre-determined direction can have any p-value.
Secondly, even the smaller one-sided tail probability can be >0.5. The problem is discreteness. It may be that 0.62 is the smaller of p(T>=t) and p(T<=t), and this turns out to be the case. The other tail probability is 0.83.
OpenEpi should still be thresholding the p-value at 1, but that's just cosmetic.
-thomas
Thomas Lumley Assoc. Professor, Biostatistics
tlumley at u.washington.edu University of Washington, Seattle