Date:         Wed, 16 May 2001 12:58:59 -0700
Reply-To:     John Hendrickx <john_hendrickx@YAHOO.COM>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         John Hendrickx <john_hendrickx@YAHOO.COM>
Subject:      Re: Multinomial Logit/Probit
In-Reply-To:  <3B025D34.26920.CA9E67@localhost>
Content-Type: text/plain; charset=us-ascii

--- Inke_Böddeker <boeddeke@POST.MED.UNI-MARBURG.DE> wrote:
> I have a question regarding multinomial logit/probit regression
> models:
> In my data set, I have one dependent variable in three distinct,
> mutually exclusive categories y1, y2 and y3. One of the categories
> (y2) is fixed as the reference against the others. I would
> therefore
> like to predict the probability of y1 versus y2 and y3 versus y2.
> In contrast to the standard solutions usually given, I would like
> to
> use different covariate vectors for both of these predictions, that
> is,
> X1, X2, X3, and X4 for y1 versus y2, but X1, X2, X5, and X6 for y3
> versus y2.
> Does anyone know of an implementation (in any program) for this
> problem? Or has anyone seen an example of this hybrid model
> published or mentioned anywhere?

You could do this by specifying the multinomial logit model as a
conditional logit model and estimating it with COXREG. This gives the
flexibility to specify a "response function" for each independent
variable in your model. Details on how to implement this are at
http://baserv.uci.kun.nl/~johnh/mcl/

Good luck,
John Hendrickx

__________________________________________________
Do You Yahoo!?
Yahoo! Auctions - buy the things you want at great prices
http://auctions.yahoo.com/