Date:         Thu, 5 Feb 2009 00:10:02 -0800
Reply-To:     John Hendrickx <John2.Hendrickx@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         John Hendrickx <John2.Hendrickx@GMAIL.COM>
Organization: http://groups.google.com
Subject:      Re: Adjacent Category Models
Comments: To: sas-l@uga.edu
Content-Type: text/plain; charset=ISO-8859-1

Hello Zai,

One way of doing this is as a "multinomial conditional logit" (MCL)
model. The idea is to use program for a conditional logit model to
estimate an multinomial logit model. This will produce the same
results but allows you to impose constraint on the parametrization of
the dependent variable.

More information is available at http://home.wanadoo.nl/john.hendrickx/statres/mcl/
The LR.SAS program at http://home.wanadoo.nl/john.hendrickx/statres/mcl/sas/
shows how to estimate a multinomial logistic model using PROC PHREG.
For a dependent variable with 3 categories, you'd have to make the
following modification:

array y{3};
do i=1 to 3;
    if newy <= i then y[i]=1-i/3;
    else y[i]=-i/3;
end;

(And create interactions of y: with your independent variables).

Good luck,

John Hendrickx

Statistician
Clinquest Europe BV, Oss - The Netherlands
www.clinquest.com

On 3 feb, 17:50, Zai Saki <zais...@gmail.com> wrote:
> Hi All,
> I am trying to build a model with an ordinal dv (scale 1 to 4). We are
> looking for something like the adjacent category model that is
> available in proc catmod. However our independent variables are
> ratings on a scale of 1 to 10 and we have upto 25 of these. So proc
> catmod is ruled out.
>
> One key goal is to have unique parameter estimates for each transition
> unlike a proportional odds model.
>
> I believe we can do it in proc genmod by specifying a customized link
> using "fwdlink", "invlink", "deviance", and "variance" statements as
> such a link is not available inherently in proc genmod.
>
> Any suggestions on how to fit such a model using proc genmod or any
> other sas procedure would be greatly appreciated.
>
> Thanks,
> Zai