Date: Wed, 10 Nov 2004 10:23:21 -0800
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: "David L. Cassell" <cassell.david@EPAMAIL.EPA.GOV>
Subject: Re: Logistic Regression Question
Content-type: text/plain; charset=US-ASCII
"DePuy, Venita" <depuy001@DCRI.DUKE.EDU> sagely replied:
> My understanding is that Catmod is a) painful, b) geared toward
> logit regression, and c) can be accomplished in Genmod.
[A] CATMOD is older, and not as flexible. It's so old that my wife
used it in her doctoral dissertation. (Why does that sound vaguely as
if Rodney Dangerfield should be saying it, followed by a rimshot? :-)
[B] It's still fairly useful. You can do logits, generalized logits,
cumulative logits, means, proportions, blah blah blah. BUT you can't
really control the estimation method. You get MLE with logits and
generalized logits, and weighted least squares (but NOT the sample
as in this thread!) elsewhere. GENMOD and other, newer, procs are more
[C] Pretty much.
> Why don't you try proc surveyreg? Although I don't know that it will
It won't. But you're dead on. The survey design effects have to be
for. PROC SURVEYLOGISTIC (SAS 9.1) is made for this problem.
> At any rate, you can do logistic regression in Genmod with a logit
> I believe it has more options than Proc Logistic does.
But doing all the mucking around to adjust for survey design effects in
variance-covariance matrix is a headache, and that still doesn't take
account all the other problems implicit in trying to wedge the survey
paradigm into a model-based paradigm. Like the lack of proper domain
> Don't do survey stuff myself, so perhaps someone else can speak to
> of things.
Like anyone around here can get me to shut up. :-)
David Cassell, CSC
Senior computing specialist