[Limdep Nlogit List] Crosstab command for logit

William Greene wgreene at stern.nyu.edu
Sat Feb 18 04:39:34 EST 2006

Dear Branca:  As the manual explains, that crosstab is computed as the sum
of p(i) * d(i)' where p(i) is the vector of predicted probabilities and d(i) is a vector
of zeros save for a one placed in the position of the individual's choice.  If you 
have choice specific constant terms in the model, then by construction, for an MNL
model, this method will match the totals on the diagonals of the matrix.  It is 
implied by the first order conditions for computation of the coefficients.  If you
replaced p(i) with some vector or "predictions" which contained some kind of
predicted outcome - thereby a one in the position of the prediction and zeros
elsewhere, then computed the table in the same way, you would get the outcome 
you appear to be expecting.  But, this requires you to define the d^(i).  Whatever
method you use, it will be arbitrary.  NLOGIT does not show any preference on
this score - there is no natural prediction.  The ;LIST parameter will produce a 
listing in which the "prediction" is the cell with the largest probability.  You could
use this.  This is something that you, yourself, can program, however, it is not
done automatically within the software.
Bill Greene

Professor William Greene
Department of Economics
Stern School of Business
New York University
44 West 4th St., Rm. 7-78
New York, NY   10012
Ph. 212.998.0876
Fax. 212.995.4218
URL. http://www.stern.nyu.edu/~wgreene
Email. wgreene at stern.nyu.edu

----- Original Message -----
From: Branka Valcic <ffbv at uaf.edu>
Date: Friday, February 17, 2006 2:45 am
Subject: [Limdep Nlogit List] Crosstab command for logit

> Greetings,
> I am trying to get a cross-tabulation of actual versus predicted 
> outcomes for each alternative in a multinomial logit. It seems to 
> me 
> that ;Crosstab command in NLOGIT should do the trick. The 
> resulting 
> table, however, shows for each alternative total row equal total 
> column, 
> which, if I am correctly interpreting the table means that for 
> each 
> alternative the model predicts 100% of the actual outcomes! It is 
> very 
> hard for me to believe this so I must wonder if I am 
> misinterpreting the 
> table. Would someone please enlighten me on this issue?
> Best regards,
> Branka Valcic
> ------------------------
> Branka Valcic
> Assistant Professor
> Economics Program
> School of Management
> University of Alaska Fairbanks
> P.O. Box 756080
> Fairbanks, Alaska 99775-6080
> Phone: (907) 474 2754
> Fax: (907) 474 5219
> URL: www.faculty.uaf.edu/ffbv
> -------------------------
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au

More information about the Limdep mailing list