Date:         Mon, 14 Jul 2003 14:07:44 -0700
Reply-To:     cassell.david@EPAMAIL.EPA.GOV
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "David L. Cassell" <cassell.david@EPAMAIL.EPA.GOV>
Subject:      Re: Proc Genmod
Comments: To: John Conway <jpconway2@COMCAST.NET>
Content-type: text/plain; charset=us-ascii

John Conway <jpconway2@COMCAST.NET> wrote [once again]:
> This is a multi-part message in MIME format.

Please stop doing that.  This venue is a mailing list, a UseNet group,
and a digest service.  Sending MIME (and sending more than one post
of exactly the same message) actually causes problems for large numbers
of participants.  Thanks in advance.

> Does Proc Genmod have the facility for either forward or backward
selection
of predictors?

No.  PROC GENMOD is not a simple linear model, so the traditional
selection methods and rules of thumb would not be appropriate.  If they
are
ever appropriate.

Furthermore, I salute SAS for *not* having any forward/backward/
stepwise/all-regressors/throw-everything-into-the-pot-and-stir
methodologies
in PROC GENMOD.  They are too badly abused in PROC REG as things are.
I have written _ad_nauseam_ on the subject in this venue, but all too
often
these tools are misused and abused in hideous ways, causing the user to
end
up with a nice-looking formula.. which predicts nothing and is chockful
of
misleading numbers.  Add in an arbitrary number of potential link
functions
and response probability distributions, and a mathematical search for an
'ideal' equation becomes an exercise in improperly attributing meaning
to
the most serious outliers and most egregious leverage points.  Please,
please,
*please* do not go this route.

If you are searching for a way of modeling with many collinear
variables,
then perhaps you should look at PROC PLS instead.  That might be more
helpful
than listening to me gripe endlessly.  :-)

HTH,
David
--
David Cassell, CSC
Cassell.David@epa.gov
Senior computing specialist
mathematical statistician