R help - Feb 2007 - multinomial logistic regression with equality constraints?

I'm interested in doing multinomial logistic regression with equality 
constraints on some of the parameter values.  For example, with 
categorical outcomes Y_1 (baseline), Y_2, and Y_3, and covariates X_1 
and X_2, I might want to impose the equality constraint that

   \beta_{2,1} = \beta_{3,2}

that is, that the effect of X_1 on the logit of Y_2 is the same as the 
effect of X_2 on the logit of Y_3.

Is there an existing facility or package in R for doing this?  Would 
multinomRob fit the bill?

Many thanks,

Roger


-- 

Roger Levy                      Email: rlevy at ucsd.edu
Assistant Professor             Phone: 858-534-7219
Department of Linguistics       Fax:   858-534-4789
UC San Diego                    Web:   http://ling.ucsd.edu/~rlevy

Hi Roger,

Yes, multinomRob can handle equality constraints of this type---see
the 'equality' option.  But the function assumes that the outcomes are
multinomial counts and it estimates overdispersed multinomial logistic
models via MLE, a robust redescending-M estimator, and LQD which is
another high breakdown point estimator.  It would be a simple matter
to edit the 'multinomMLE' function to work without counts and to do
straight MNL instead, but right now it estimates an overdispersed MNL
model.

Cheers,
Jas.

======================================Jasjeet S. Sekhon                     
                                      
Associate Professor             
Travers Department of Political Science
Survey Research Center          
UC Berkeley                     

http://sekhon.berkeley.edu/
V: 510-642-9974  F: 617-507-5524
======================================


Roger Levy writes:
 > I'm interested in doing multinomial logistic regression with equality 
 > constraints on some of the parameter values.  For example, with 
 > categorical outcomes Y_1 (baseline), Y_2, and Y_3, and covariates X_1 
 > and X_2, I might want to impose the equality constraint that
 > 
 >    \beta_{2,1} = \beta_{3,2}
 > 
 > that is, that the effect of X_1 on the logit of Y_2 is the same as the 
 > effect of X_2 on the logit of Y_3.
 > 
 > Is there an existing facility or package in R for doing this?  Would 
 > multinomRob fit the bill?
 > 
 > Many thanks,
 > 
 > Roger
 > 
 > 
 > -- 
 > 
 > Roger Levy                      Email: rlevy at ucsd.edu
 > Assistant Professor             Phone: 858-534-7219
 > Department of Linguistics       Fax:   858-534-4789
 > UC San Diego                    Web:   http://ling.ucsd.edu/~rlevy
 > 
 >