Date: Fri, 22 Oct 2004 11:50:02 -0500
Reply-To: Robin High <robinh@UNLSERVE.UNL.EDU>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Robin High <robinh@UNLSERVE.UNL.EDU>
Subject: Re: Negative binomial Proc genmod variable selection
In-Reply-To: <s178f3d8.096@MAIL.NDRI.ORG>
Content-Type: TEXT/PLAIN; charset=US-ASCII
> I use "Proc genmod" to fit "negative binomial" regression. But there
> is no backward, forward, and stepwise in proc genmod to help me select
> the important varibles(proc reg has variable selection function). I
> have about 30 variables. Is there any method to select variables in
> genmod automatically. Thanks.
The number of messages and frequency of replies and irritation therein
regarding the application of stepwise statistical procedures has generated
some "classic" quotes that I've collected over the past few years. I admit
to not documenting who said what, so if anyone wants to take credit for
these gems, please step... er excuse me, please come forward.
Robin High
Univ. of Oregon
Stepwise regression has few good uses, and *meaning* is seldom one of
them.
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.
It is our experience and strong belief that better models and a better
understanding of one's data result from focussed data analysis, guided
by substantive theory.
Thinking about what analysis you are doing, I think it's safe to say, is
always a good idea. As Abraham Lincoln said "If I had 8 hours to chop
down a tree, I would spend 4 of them sharpening my axe".
The data analyst knows more than the computer.
contrasted with:
Stepwise allows us to NOT think about the problem.
Failure to use that knowledge produces inadequate data analysis.
Personally, I would no more let an automatic routine select my model
than I would let some best-fit procedure pack my suitcase.
