Date:         Wed, 18 Apr 2001 09:47:19 -0500
Reply-To:     Jaclyn Whitehorn <jaclyn@BAMA.UA.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Jaclyn Whitehorn <jaclyn@BAMA.UA.EDU>
Subject:      Re: Multicollinearity in Logistic Regression
In-Reply-To:  <200104181425.JAA13423@bama.ua.edu>
Content-Type: text/plain; charset="us-ascii"; format=flowed

At 10:25 AM 4/18/01 -0400, Kevin Brunson <kbrunson@ND.EDU> wrote (in part):
>Also, if your goal is a good predictive model then multicollinearity can be
>ignored.

Not to be too picky, but that's only partially true.  You have to make sure
to avoid extrapolation if you have strong multicollinearity.  This is based
not just on the values of each explanatory variable, but on their
combination.  So if you have multiple X variables, it can be very difficult
to determine whether you are trying to extrapolate or not.

Just my $0.02...


---------------------------------------------------
Jaclyn Whitehorn * User Service Consultant
Seebeck Computer Center * The University of Alabama
(205)348-8720