[Limdep Nlogit List] Nested Logit Model

Mon Mar 1 00:53:00 EST 2010

M. Sabir.
Whether or not b51 and b52 are fixed, this model is not
identified. The model specifies a basic MNL in the Yes
branch and a degenerate branch for No. For the 4 choices
C,B,T,O, you must fix at zero one of the constants, one
of the coefficients on X1 and one of the coefficients
on X2.
Regards,
Bill Greene

----- Original Message -----
From: "M. (Muhammad) Sabir" <msabir at feweb.vu.nl>
To: "Limdep and Nlogit Mailing List" <limdep at limdep.itls.usyd.edu.au>
Sent: Wednesday, February 24, 2010 8:38:51 AM GMT -05:00 Colombia
Subject: Re: [Limdep Nlogit List] Nested Logit Model

Dear Prof. Greene,

 I got your kind comments on a model sometime ago. Things work fine.
However, recently, I got a question with this methodology which create
confusion for me. 

Question is, "if we fix B51 and B52 in below model, 

NLogit
;Lhs= choice, cset, altij
;Choices = C, B, T, O, N
;crosstabs
; tree = Yes(C,B,T, O),  No(N)
;start=logit
;ivset:(No)=[1.0]
;Model:
 U(C) = bc + b11*X1+b12*X2 /
 U(B) = bB + b21*X1+b22*X2 /
 U(T) = bT + b31*X1+b32*X2 /
 U(O) = bO + b41*X1+b42*X2 /
 U(N) =      b51*X1+b52*X2   
  ; utility = u1 $   
(Where X1 and X2 are individual specific and do not varies across the
choices.) 

How it could be possible to identify the coefficients of all variables
in "yes" nest (C, B, T, O), given that their utility function are only
differ from one another by ASC and by IV values, but IV values
(Inclusive values) are same for each alternative in same nest, which
means there is no variation in this nest. Then how we can identify the
coefficient of this nest ?" 

Am I missing something very basic ?  Would someone here would be kind
enough to express this a little more. I will be very thankful for it. 

Many thanks.

Kind regards,
Muhammad Sabir
PhD Candidate 
VU University
Amsterdam, The Netherlands 

-----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: woensdag 22 juli 2009 17:54
To: Limdep and Nlogit Mailing List
Subject: Re: [Limdep Nlogit List] Nested Logit Model

Agreed. B51 and B52 are not identified, and must be fixed at zero.
/B. Greene

----- Original Message -----
From: "Thomas C. Eagle" <teagle at tceagle.com>
To: "Limdep and Nlogit Mailing List" <limdep at limdep.itls.usyd.edu.au>
Sent: Wednesday, July 22, 2009 10:42:11 AM GMT -05:00 US/Canada Eastern
Subject: Re: [Limdep Nlogit List] Nested Logit Model

Muhammed,

You cannot have X values that are constant across all alternatives in a
set
and estimate parameters for every alternative.  One alternative must be
set
to have a zero utility in such a rare situation.

Tom Eagle

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Sabir, M.
(Muhammad)
Sent: Wednesday, July 22, 2009 6:55 AM
To: limdep at limdep.itls.usyd.edu.au
Subject: [Limdep Nlogit List] Nested Logit Model

Dear all,

 I am trying to estimate the nested logit model with follwing details; 

NLogit

;Lhs= choice, cset, altij

;Choices = C, B, T, O, N

;crosstabs

; tree = Yes(C,B,T, O),  No(N)

;start=logit

;ivset:(No)=[1.0]

;Model:

 U(C) = bc + b11*X1+b12*X2 /

 U(B) = bB + b21*X1+b22*X2 /

 U(T) = bT + b31*X1+b32*X2 /

 U(O) = bO + b41*X1+b42*X2  /

 U(N) =      b51*X1+b52*X2   

  ; utility = u1 $

Where X1 and X2 are individual specific and do not varies across the
choices. 

when I run the above model I get the following errors;

+---------------------------------------------+

| Discrete choice and multinomial logit models|

+---------------------------------------------+

Hessian is not positive definite at start values.

  Error   803: Hessian is not positive definite at start values.

B0 is too far from solution for Newton method.

Switching to BFGS as a better solution method.

Line search does not improve fn. Exit iterations. Status=3

Check derivatives (with ;OUTPUT=3). This may be a solution

if several iterations have been computed, not if only one.

  Error   806: (The log likelihood is flat at the current estimates.)

After these error messages, it provides the output for MNL however there
are many coefficients missing (Coefficients are there but their standard
deviation are not provided. Only written fixed parameters). 

And then the following error messages comes too. 

Initial iterations cannot improve function.Status=3

  Error   805: Initial iterations cannot improve function.Status=3

Function=  .46408498810D+05, at entry,  .46408498810D+05 at exit

  Error  1025: Failed to fit model. See earlier diagnostic.

I have no idea/guess what is going wrong ? 

Can anyone suggest that what could be the possible reason for collapse
of the model ? and any possible direction to correct for that ?

Thanks in advance. 

Kind regards,

Muhammad 

_______________________________________________
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
_______________________________________________
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