Date: Fri, 20 Apr 2007 16:07:28 -0400
Reply-To: Peter Flom <firstname.lastname@example.org>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject: Odd logistic regression results (CI and p value)
Content-Type: text/plain; charset=UTF-8
When I first replied, it bounced because (SUBSCRIBE SAS L) sounded like a command!
Anyway, the person with that user name wrote about LOGISTIC, here is my reply
>Using the SURVEYLOGISTIC procedure, I have sig Odd Ratios but the Wald
>confidence intervals include '1'. Has anyone seen this before? And how
>would you interpret it?
I've not seen it with SURVEYLOGISTIC, but I've seen it with LOGISTIC. For
some reason (maybe someone will know the details of why) SAS defaults to using different
methods for the p values and the CIs in LOGISTIC. I don't have the details
to hand, but I think it uses the likelihood ratio for the p-values and the Wald
for the CIs.
Whenever I've seen this, the two results have been very 'close', e.g.
the CI ALMOST includes one, and the p is 0.051 or something like that
In short, unless the discrepancy is severe, I don't think it's anything
to worry about