Date:         Thu, 22 Jan 2004 16:50:27 -0600
Reply-To:     Paul R Swank <Paul.R.Swank@UTH.TMC.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Paul R Swank <Paul.R.Swank@UTH.TMC.EDU>
Subject:      Re: Appropraiteness of PROC GENMOD
In-Reply-To:  <A7D4140CA0A3D24CA55F8A4495F1B251037CACBD@argon.dcri.duke.edu>
Content-Type: text/plain; charset="us-ascii"

It sounds to me like the data is clustered, not repeated since family member
is not a factor. Sounds like proc mixed or nlmixed if the distribution of
N_episodes is Poisson or negative binomial.

Paul R. Swank, Ph.D.
Professor, Developmental Pediatrics
Medical School
UT Health Science Center at Houston


-----Original Message-----
From: SAS(r) Discussion [mailto:SAS-L@LISTSERV.UGA.EDU] On Behalf Of DePuy,
Venita
Sent: Thursday, January 22, 2004 4:37 PM
To: SAS-L@LISTSERV.UGA.EDU
Subject: Re: Appropraiteness of PROC GENMOD


Have you considered Proc GLM?
I think it's easier to comprehend than Genmod (at least to me) and it will
handle the Repeated option. Just a quick thought. -Venita

> ----------
> From:         Cacialli, Doug[SMTP:Doug_Cacialli@URMC.ROCHESTER.EDU]
> Reply To:     Cacialli, Doug
> Sent:         Thursday, January 22, 2004 5:21 PM
> To:   SAS-L@LISTSERV.UGA.EDU
> Subject:      Appropraiteness of PROC GENMOD
>
> Y'all,
>
> I'd like to open with an apology for my statistical ignorance here ...
> I'm not extremely savvy with complex analyses.  I'm needing to do some
> regression analyses on the effects of five independent variables, four
> of which are continuous and one of which is categorical, on number of
> episodes of a given disorder (which is obviously continuous).
>
> My first instinct was to run PROC LOGISTIC on account of the
> categorical independent variable, but as the dependent variable isn't
> categorical, I'm under the impression that PROC LOGISTIC would be
> inappropriate.  I've been using PROC GENMOD lately with the assistance
> of a consultant with much more training than I, and it occurred to me
> that a generalized linear model might
> work here.  I threw this together:
>
> proc genmod data = work.ANALYSIS_I;
>         class   FAMILY_ID SEX;
>         model   N_EPISODES = ONSETAGE_MDD ONSETAGE_FIRST SEX AGE_AT_T1;
>         repeated subject = FAMILY_ID/corr=cs corrw;
> run;
>
> The repeated subject = FAMILY_ID takes the relation between certain
> subjects (these data were originally collected as part of a
> longitudinal family study).
>
> My question, in short, is if this is a legit use of PROC GENMOD.  And
> if it's not, can someone suggest an appropriate tool?
>
> I've received a bit of help on a number of different issues from SAS-L
> lately.  Thank you all for your help, past and present.  I'm learning
> so much ... it's greatly appreciated.
>
> Doug out.
>
> -------------------------------------------
> Doug Cacialli - Data Manager / Data Analyst
> Depression Research Laboratory
> University of Rochester Medical Center
> 300 Crittenden Boulevard - Box PSYCH
> Rochester, New York 14642
> Phone: (585)273-3309 Fax: (585)506-0287
> -------------------------------------------
>