[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: st: Multinomial probit vs multinomial logit

From   jverkuilen <>
To   <>
Subject   RE: st: Multinomial probit vs multinomial logit
Date   Fri, 17 Apr 2009 10:31:26 -0400

Maarten's right that models such as -amsprobit- would be what you need. 

What are the products? Correlation is due to similarity of the choices not accounted for by your model. So two choices that are quite similar and a third that's diferent will cause IIA violation. Nested logit might accommodate what you want. 

Anyway, I highly suggest getting the excellent book by Kenneth Train, Discrete choice with simulation, Cambridge, 2003. (Or see his web page at UC Berkeley.)

-----Original Message-----
From: "Maarten buis" <>
Sent: 4/17/2009 5:05 AM
Subject: Re: st: Multinomial probit vs multinomial logit

--- On Fri, 17/4/09, Paley, Irina wrote:
> I am analyzing a choice between 3 types of mortgage
> products, and the iir is consistenlty rejected by all tests.
> I have seen several sources saying that multinomial probit
> can be used instead of multinomial logit in this case-- but
> all the examples of multinomial probit I see have to do with
> joint decisions (such as voting for taxes and using public
> schools) rather than chosing between, in this case,
> different types of mortgage products.
> As a practical matter, multinomial probit gives estimates
> similar to multinomial logit...

-mprobit- is very similar to -mlogit- because -mprobit- also 
enforces a variant of the IIA assumption. If you want to 
relax that assumption you will need to use -asmprobit-. The
problem with these models is identification: In addition to
the regular parameters that you try to estimate you are also
trying to estimate correlations between latent (= not directly
observed) variables. That information has to come from 
somewhere, usually in the form of some other pretty strong 
assumptions. So my feeling about these models is that they 
are just replacing one set of strong and wrong assumptions 
with another set of equally strong and wrong assumptions...

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index