Date: Thu, 17 Aug 2006 06:40:38 -0400
Reply-To: Nash Smith <nash2sasl@YAHOO.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Nash Smith <nash2sasl@YAHOO.COM>
Subject: Re: A Question on iterating SAS Codes
Content-Type: text/plain; charset=ISO-8859-1
I looked over a few of your postings on why we should **not** be
considering stepwise selection and I think I understand your reasoning
even though I am not an expert in this area -thanks for pointing that out.
Also, in an attempt to simplify the question I guess I ended up with
providing inadequate information. I have a situation wherein I am asked to
build models to predict propensity to visit a store for a national retail
chain with 300+ stores. For each of the models, a customer would have had
to made a purchase at that store(say Store A) in the past year or another
store in the vicinity(say a 5 mile radius)during the same period to
qualify for selection for modeling Store A's likely customers. Given the
time constraints I think it would be a nightmare trying to build each
model individually and decided to post this question to the group to see
if one of us have an idea on how to iterate this process.Once again,
thanks much for your feedback and the links to past postings on this
On Wed, 16 Aug 2006 23:21:02 -0700, David L Cassell
>>I have a need for building numerous regression models (in 100s) from a
>>single dataset. The underlying data for each of the models will be
>>selected based on a qualifying criteria for that model. I would like to
>>have a process by which I can use macros to iterate the procedure
>>I am planning to spend additional time to refine the results from such a
>>procedure that may use stepwise selection to select the initial models.
>>Since I have very little knowledge on macros, I am trying to get some
>>to achieve my objective because it would be a time-consuming process if I
>>were to run each of the models separately. Any help would be a huge
>>benefit and would save loads of time and system resources.
> Building hundreds of regression models from a data set is a Bad Thing.
> Using stepwise selection is a Very Bad Thing.
> If you don't know why these two are really bad things to do, then
>check out the STAT-L FAQ, or look up some of my discussions (read:
>screaming spittle-flying rants) on the ills of stepwise regression in the
>archives of SAS-L.
> Perhaps, if you write back to SAS-L, and explain what your are
>trying to do, and what the larger picture is that is driving this process,
>then someone here ought to be able to give you some more constructive
> For fun, try this:
>You'll only have 50 or 60 hits where I discuss the problems you are
>bringing upon yourself.
> Once you have a better model-building strategy, you may not
>need a macro solution.
>David L. Cassell
>3115 NW Norwood Pl.
>Corvallis OR 97330
>Donít just search. Find. Check out the new MSN Search!