Date: Thu, 24 Nov 2005 10:19:05 -0500
Reply-To: "Nick ." <ni14@MAIL.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "Nick ." <ni14@MAIL.COM>
Subject: Re: Campaign Marketing--Modeling Question
Content-Type: text/plain; charset="iso-8859-1"
Thanks Bora !!! I now have something to work with. I will try and track down this paper so I can learn how to do this project at work. As for your question to data reduction techniques:
(1) Don't ever mention D.C. and stepwise and all of its sisters and cousins in the same sentence or breath or universe. It will drive D.C. mad. We need him sane! For right now anyway. When I become rich and famous and knowledgeable and smart, we won't need him then. -:)
(2) D.C has mentioned PLS (Partial Least Squares) as a way, among many other ways, for data reduction. I am not certain how to do this. Can PLS deal with character variable reduction? I am pretty sure it cannot.
(3) I think clustering, as you mention, will probaly work, just pick one variable from each cluster. At least this way maybe you will avoid collinearity issues and such. But does VARCLUS deal with character variables? Say, for excample, I have a variable called INCOME coded as A=0-14K, B=15-45K, C=45-75K, D=75-99K, E=99-200K, F=200K+. I have many such variables, like AGE, GENDER, MARITAL_STATUS, GEOGRAPHIC_REGION, and on and on and on. Now what? How are we going to reduce these? With numerics, at least, we have some tools available like clustering or PRINCOMP, etc. Just my two cents worth.
Play 100s of games for FREE! http://games.mail.com/