Date:         Wed, 11 Apr 2007 14:41:02 +0200
Reply-To:     Anne-Elie carsin <aneliecarsin@HOTMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Anne-Elie carsin <aneliecarsin@HOTMAIL.COM>
Subject:      Proc Glimmix without Random _residual_ VS Glimmix Macro
Content-Type: text/plain; charset=iso-8859-1; format=flowed

Dear SAS-lister,

I would like to run a multi-level analysis on some data structured as
follow: level 1=Patient, level 2=surgeon nested in hospital (level 3).  The
outcome is potsop=0/1.
I used the Macro %Glimmix before, and it works fine.
Now, I wanted to use proc Glimmix (seems easier and give the Odds Ratio
which save me time).
Anyway, I translated the macro code to the procedure code as follow:

%glimmix(   DATA = may06,
     procopt =   method=reml covtest ,
     stmts=%str(
       class surgeon hospital hosp_ter;
       id surgeon hospital ;
       model postop=hosp_ter /  ddfm=kr ;
       random int / subject=hospital ;
       random int / subject=surgeon(hospital) ;
     ),
    error = binomial,
    link = logit
);
run;

proc glimmix DATA = may06;
     class surgeon hospital hosp_ter;
     id surgeon hospital ;
     model postop=hosp_ter / ddfm=kr oddsratio link=logit dist=binomial ;
     random int  / subject=hospital;
     random int  / subject=surgeon(hospital) ;
     nloptions tech=nrridg;
     random _residual_;
run;

I just don't get why the RANDOM _residual_; is needed in the Proc Glimmix?
Results are slighlty different when omitted and I wonder what results would
be the correct ones? Any idea?

Thanks for your help!

Anne-Elie

_________________________________________________________________
Avec Windows Live OneCare éliminez tous les virus de votre PC !
http://www.windowslive.fr/liveonecare/default.asp