Date: Mon, 25 Jun 2007 22:22:19 -0700
Reply-To: David L Cassell <davidlcassell@MSN.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: David L Cassell <davidlcassell@MSN.COM>
Subject: Re: Collinearity interpretation
Content-Type: text/plain; format=flowed
>I agree with Peter.
> The SVD method used by the COLLIN option is more
>accepted than VIF. VIF will not identify collinearities involving the
>intercept term. More information is provided by COLLIN to identify which
>variables are implicated in the collinear relationship and the option
>provides more insight on which variables to remove if a collinearity
>Some modellers run the diagnostic after the final model is built, but I
>like to run at the start to get a clean base of variables to examine.
>Do people include interaction temrs and polynomial terms in their
Well, they should. But I assume that many do not. In fact, many people
seem to go out of their way to build as many highly-correlated variables
as is humanly possible, then trust icky things like stepwise regression to
drag them out of the morass they just created.
David L. Cassell
3115 NW Norwood Pl.
Corvallis OR 97330
Need a break? Find your escape route with Live Search Maps.