Date:         Mon, 4 Jun 2001 14:27:28 -0400
Reply-To:     ahutson@BIOSTAT.UFL.EDU
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Alan Hutson <ahutson@BIOSTAT.UFL.EDU>
Organization: University of Florida
Subject:      Re: constraining logistic slopes
Comments: To: Paul von Hippel -- Ohio State <pvh@CCRMA.STANFORD.EDU>
Content-Type: text/plain; charset=us-ascii

Paul

 If you  don't mind doing a bit of programming you can
in general carry out constrained maximum likelihood
via PROC NLP without too much trouble.
Logistic regression is just a specific case.

Best
Alan

Paul von Hippel -- Ohio State wrote:

> To state my question more generally, I would like to constrain the
> parameters of a logistic regression model. As far as I can tell,
> the only parameter constraint available in SAS is suppression of
> the intercept. If other types of constraint are available, I'd be
> grateful for a reference to pertinent documentation.
>
> Many thanks,
> Paul von Hippel
>
> P.S. Below I clarify two details from my original posting.
>
> On 4 Jun 2001, Seymour Douglas wrote:
>
> >Am I reading you right, you want to constrain the coefficient
> >to be zero if i isgreather than or equal  2.
>
> Let me clarify. When I say
>         b_i >= 0 for all i >= 2
> I mean that the coefficient is *non-negative* if i is at least 2.
>
> >Is I a variable value?
>
> i is a subscript on the coefficient and variable names.
> In my original statement of the model, i=0,1,2,...,k, as follows.
>
>         log odds (Y) = b_0 + b_1*X_1 + b_2*X_2 + ... + b_k*X_k