Date: Tue, 6 May 2008 11:21:52 -0700
Reply-To: Amy Smith <asmith197401@YAHOO.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Amy Smith <asmith197401@YAHOO.COM>
Subject: Re: Is it true that a zero-inflated Gamma model is not possible
in either GENMOD or NLMIXED?
Content-Type: text/plain; charset=iso-8859-1
I guess that I don't see why it is possible to do a zero-inflated Poisson as a two part model that it partly binomial and partly Poisson, but it is not okay to do a two part model that is partly binomial for whether or not you have costs and partly a GLM with a gamma distribution for those people that have costs. What makes the zero-inflated Poisson okay when the zero-inflated gamma is then totally wrong?
----- Original Message ----
From: Sigurd Hermansen <HERMANS1@WESTAT.com>
To: Amy Smith <asmith197401@yahoo.com>; SAS-L@listserv.uga.edu
Sent: Tuesday, May 6, 2008 2:11:47 PM
Subject: RE: Is it true that a zero-inflated Gamma model is not possible in either GENMOD or NLMIXED?
Amy:
I can understand your feelings of distress, but I'd say that much of your frustration comes from trying to handle a difficult statistical consulting assignment as SAS PROC GENMOD or NLMIXED syntax problem. Since the idea of joining gamma and bin distribution versions of the same model doesn't make sense, a consultant would have to go back to the basics and work forward from there.
Contributors to the list with substantial experience in fitting generalized linear and mixed models (not me) consult at rates much higher than $15. a gig. We appreciate their contributions of exceptional value, and we would prefer to keep this forum pro bono. I believe that I speak almost everyone in asking that we not make SAS-L an outpost of E-Bay.
I don't, as I mentioned earlier, see zero-inflation or transformation of cost distribution as the critical elements in assessing the impact of interventions on costs. Until you have a clearer idea of what you are trying to estimate, how you specify a model won't matter.
S
-----Original Message-----
From: owner-sas-l@listserv.uga.edu [mailto:owner-sas-l@listserv.uga.edu] On Behalf Of Amy Smith
Sent: Tuesday, May 06, 2008 1:13 PM
To: SAS-L@listserv.uga.edu
Subject: Re: Is it true that a zero-inflated Gamma model is not possible in either GENMOD or NLMIXED?
I'm now going to offer a $15 gift card for Amazon.com to the first person that can give me functional NLMixed or GENMOD code that allows me to do the two part model that I am trying to do here. I'm looking for one p-value here. Yes, I'm officially that frustrated with NLMixed. My gift card offer expires at 5PM Eastern time on May 7, 2008. A
----- Forwarded Message ----
From: Amy Smith <asmith197401@yahoo.com>
To: SAS-L@LISTSERV.UGA.EDU
Sent: Monday, May 5, 2008 5:52:43 PM
Subject: Is it true that a zero-inflated Gamma model is not possible in either GENMOD or NLMIXED?
Has anyone found a procedure in SAS for creating a zero-inflated Gamma model? My response variable is cost. I have a class variable called group that has three levels Intervention1, Intervention2, and Control.
The following two models succeed without any problem.
PROC GENMOD;
CLASS group;
MODEL cost = group / LINK=log DIST = bin;
PROC GENMOD;
CLASS group;
MODEL cost = group / LINK=log DIST = gamma;
I cannot figure out how to get GENMOD to join these to give a zero-inflated gamma model. Or I thought maybe NLMixed could handle this model but I cannot get the syntax to work right and I'm now questioning if it is even possible. Has anyone ever written SAS code that successfully runs on a zero-inflated gamma? I would really love to see the syntax on a simple example like I have here.
Thanks for your help!
________________________________
Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now.
____________________________________________________________________________________
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
____________________________________________________________________________________
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile. Try it now. http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
| 
