[Limdep Nlogit List] Model structure for nlogit

William Greene wgreene at stern.nyu.edu
Fri Jun 27 20:30:46 EST 2008

Dear Caroline.  You will find the nested logit model fit by NLOGIT
documented in great detail in the program manual.  There are different
forms of the nested logit model.  The one you have fit is the RU1
form of the random utility model.  The RU1 form of the model is shown on
page N16-19 of the manual. You can make it more explicit by adding ;RU1
to the command. (Also correct the typographical error in the command,
you should have a comma before "hinc.")  This will produce the same
numerical results, but a bit more detail in the output.
   It is impossible to discern from the Stata command what model is 
being fit. Unfortunately, we are receiving a fair number of inquiries
like this in which it is unclear what Stata is doing, and very difficult
to determine from the commands - there isn't much information in the
Stata command you have provided (and you have not provided the output). What 
is clear is that the model you are fitting with Stata is different from the 
one that you fit with NLOGIT. Perhaps the Stata manual contains some 
documentation of what form of random utility model (if any) is being estimated. It
might well be the placement of the alternative specific constant terms (ASCs). 
Note that NLOGIT has used three ASC's, one each for Air, Train 
and Bus, and omitted the constant term for Train. I can't see in your Stata 
command how you would have done that. There don't appear to be constant 
terms in there at all.  Also, the NLOGIT output will contain an interaction 
between the three constants and income, "Hinc." Again, I do not see that in 
the command you show below, though maybe it is in there somewhere.  I might
guess from the Stata command that Hinc is being used in the branch level model, 
but if so, that would be incorrect. Hinc enters the twig level model. Note
as well that the constant terms and the interactions with HINC are placed
in the twig (lowest) level of the tree, and there are no covariates specified
for the branch level choice in your model.
Bill Greene

----- Original Message -----
From: "Caroline B. Theoharides" <cbtheo at gmail.com>
To: limdep at limdep.itls.usyd.edu.au
Sent: Thursday, June 26, 2008 5:48:39 PM (GMT-0500) America/Bogota
Subject: [Limdep Nlogit List] Model structure for nlogit


I apologize in advance for the simplicity of my question.  I'm a very
new user of LimDep/NLogit.  I'm currently trying to run a two level
nested logit model in NLogit.  However, my results differ dramatically
from the results I previously attained using -nlogit- in Stata 10 (rum
estimation).  In order to try to figure this out, I brought the clogit
project data from NLogit into Stata.  However, I still can't seem to
rectify the issue.

I'm running the following code in NLogit:
NLogit ;Lhs = mode
     ;Choices = air, train, bus, car
     ;Tree = travel[fly(air),ground(train,bus,car)]
     ;show tree
     ;Rhs = invc,invt, gc
     ;Rh2 =one hinc$

I'm running this in Stata:
nlogit mode invt invc gc|| fly_ground: hinc || type:, case(id)

NB: id is simply a variable I created to identify the groups (from
aasc, basc, tasc, and casc).  fly_ground is the categorical variable
that identifies the level 1 decision for fly or ground transportation.

The tree structure and observation counts match.  My hunch is that the
problem is with the way I'm defining constants.  I've tried a variety
of constant specifications in both NLogit and Stata without success.

Thank you for your assistance,
Limdep site list
Limdep at limdep.itls.usyd.edu.au

Professor William Greene
Department of Economics
Stern School of Business
New York University
44 West 4th St., Rm. 7-78
New York, NY   10012

More information about the Limdep mailing list