[Limdep Nlogit List] Comparing two models

[Richard Hofler]

Andy,

Bill is, of course, correct. I might extend what he recommends by
referring you to Vuong's paper (cited below.) It is quite good and
includes a section on testing overlapping models, which appears to be
the case that you describe below. Note, however, that the null is that
the two models are equivalent, so don't expect that every test will
yield a clear choice of one model over the other.

Bill has also included some LIMDEP code for a Vuong test (hurdle vs.
Poisson) on p. E20-89 of the LIMDEP 8.0 guide. 

Q.H. Vuong, "Likelihood Ratio Tests for Model Selection and Non-nested
Hypotheses," Econometrica, Vol. 57, No. 2 (March 1989), 307-333. 

Richard


Richard A. Hofler
Professor 
Office: BA2 302F
Voice: (407) 823-2606



-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of William
Greene
Sent: Wednesday, January 24, 2007 7:47 AM
To: Limdep and Nlogit Mailing List
Subject: Re: [Limdep Nlogit List] Comparing two models

Andy.  You cannot use the LR test for this pair of models.  They are
nonnested, so the LR test
does not apply.  There is a huge literature on testing procedures for
nonnested models and
nonnested hypotheses.  This includes, for example, tests by Cox and by
Vuong.  There was also
a special issue of the Journal of Econometrics about 10 or 15 years ago
on the subject.
Regards,
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: Andy Sungnok Choi <Andy.Choi at anu.edu.au>
Date: Wednesday, January 24, 2007 6:29 am
Subject: [Limdep Nlogit List] Comparing two models

> Dear All,
> 
> I am still at the learning stage in Limdep. Regarding 
> loglikelihood ratio 
> (LR) tests (Hensher et. al 2005: 335), I am not clear when two 
> models being 
> compared are overlapping (i.e., not extended from one model). When 
> two 
> models have both common variables and uncommon variables with the 
> same 
> number, and when I want to compare the model significance of 
> different set 
> of uncommon variables, how can I do this?
> 
> Assuming a generic model form for simplicity,
> 
> Variables for Model 1 = A + B
> Variables for Model 2 = A + C
> 
> where A, B, and C are matrixes of variables, and the number of 
> elements for 
> B and C are the same. Then, the degree of freedom for the LR test 
> is 0. 
> Right? ... Even when the number of elements for B and C are 
> different, can 
> I use LR tests?
> 
> Point is whether I can use LR tests for this kind of model comparison.
> 
> Any sharing of experience, references, and tips will be greatly 
> appreciated.
> Thanks.
> 
> Andy S. Choi
> 
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au
> 

_______________________________________________
Limdep site list
Limdep at limdep.itls.usyd.edu.au
http://limdep.itls.usyd.edu.au