Date: Tue, 5 Dec 2006 22:00:58 -0800
Reply-To: David L Cassell <davidlcassell@MSN.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: David L Cassell <davidlcassell@MSN.COM>
Subject: Re: Selection of variables
Content-Type: text/plain; format=flowed
sue.middleton@ADELAIDE.EDU.AU wrote back:
me> >Catherine.It depends on your data, and your meta-data, and your data
me> > and your data scope, and the size of your data, and 57.341 other
me> > things as well. I personally do not recommend PROC GLM for anything
me> > any more, and PROC LOGISTIC may not be suitable for your data
me> > problem either. If you would write back to SAS-L and explain more,
me> > someone here ought to be able to give you better direction.
>why dont you recommend proc glm? How can you fit, say, the log-binomial
>if not with proc glm?
Are you sure you're thinking of PROC GLM?
You cannot fit a log-binomial with PROC GLM. PROC GLM only does
basic Ordinary Least Squares stuff. Linear models.
For a log-binomial, I would be looking at PROC NLMIXED. If you mean
a binary logit model, then PROC LOGISTIC and several other procs
would be my choices.
PROC GLM does not do the regression diagnostics of PROC REG.
It does not do the robust analyses of PROC ROBUSTREG.
It does not handle repeated measures as well as PROC MIXED.
It does not handle missing values as well as PROC MIXED.
It requires a really stupid data structure for repeated measures, anyway.
It does not handle random effects.
It does not handle data transforms anywhere near as well as PROC TRANSREG.
It does not handle survey sample data.
It does not handle measurement error problems like PROC CALIS can.
The places where it does well are places where PROC MIXED does
just as well. So I don't recommend it anymore.
David L. Cassell
3115 NW Norwood Pl.
Corvallis OR 97330
Get free, personalized commercial-free online radio with MSN Radio powered
by Pandora http://radio.msn.com/?icid=T002MSN03A07001