[Limdep Nlogit List] start=logit with NL

Mausam Duggal DuggalM at mmm.ca
Fri Jul 10 07:46:10 EST 2009


Hi Klandice,

I could be wrong here, but I think the only way that the two IV parameters could be inverses of each other is if each of the attribute parameters were alternative specific in your utility equations. If that is the case then the following logic might work: 

I am giving an example calculation below that I have worked out. 

IV (RU1) = Scale Parameter (Lvl 2) / Scale Parameter (Lvl 1)
      
         = SP2 / 1 (normalization at level 1)
  
         = 0.66 (hypothetical value)

IV (RU2) = Scale Parameter (Lvl 2) / Scale Parameter (Lvl 1)

         = 1 / SP1 (normalization at level 2)

 Therefore, SP1 = 1 / 0.66 = 1.5   (This would work because the IV parameters would be equal under both RU1 and RU2 because all the attribute parameters are alternative specific)

In your case, have you tried any other nest specification? You could test a degenerate model to see if there is a more likely nest than this one. This is my two pence.

Thanks.
 

Mausam Duggal, MCIP, RPP

Senior Project Coordinator

MMM GROUP

100 Commerce Valley Drive West

Thornhill, ON Canada L3T 0A1

T: 905-882-4211, x6289

F: 905-882-7277

W: www.mmm.ca

-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au [mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Kandice L. Kleiber
Sent: Thursday, July 09, 2009 3:49 PM
To: Limdep and Nlogit Mailing List
Subject: Re: [Limdep Nlogit List] start=logit with NL

Dr. Greene -

Thank you for your quick response!

When I run the program (and I adjusted the IVSET command to brackets),  
the output still does not show the starting values.  I guess I'm not  
clear as to whether or not the program should ALWAYS show starting  
values (when running a nested logit model)?  I get slightly different  
results when I put start=logit than when I do not.  I guess to be on  
the safe side, I should always include the start command?

Also when I use the command RU2, the reported IV parameter, μ, is  
greater than 1 (which makes sense because then the *real* IV parameter  
is (1/μ) since λ=1 and then (λ/μ)<1).  But when I use RU1 with  
IVSET: (wh,ot)=[1.0], I should receive a reported IV parameter, λ,  
with λ<1  (since μ=1 and assuming I change nothing else) ...  
correct?  However I keep receiving values >1 with RU1.

I guess I'm not sure what is happening, since they should just be  
inverses of one another?  Any ideas?

Thanks again -
Kandice.



On Jul 9, 2009, at 12:19 PM, William Greene wrote:

> Kandice: Whether an MNL starting value is computed is independent
> of the setting of RU1 vs. RU2. (At least it is according to the
> internal program code.) You should not have to add ;Start=Logit
> to the command. However, that will do no harm; the program should
> always compute the initial MNL model. There is no other place to
> get starting values. I can't say I can come up with a reason why
> the MNL results were not reported.  On your ;IVSET instruction,
> you should use IVSET:(...)=[1] to force the value to equal one. I.e,
> include it in square brackets. Otherwise it is just a starting value.
> The program reports mu, not 1/mu.
> /Bill Greene
>
>
>
>
> ----- Original Message -----
> From: "Kandice L. Kleiber" <kleiberk at onid.orst.edu>
> To: "Limdep and Nlogit Mailing List" <limdep at limdep.itls.usyd.edu.au>
> Sent: Thursday, July 9, 2009 2:26:31 PM GMT -05:00 US/Canada Eastern
> Subject: [Limdep Nlogit List] start=logit with NL
>
> Hi -
>
> I have a few questions regarding the *black box* in NLOGIT.  The
> nested logit model I have been running looks like this:
> tree = cs(corn,soy), wh(wheat), ot(other), and the data does NOT very
> across alternatives - only across observations.
>
> Because my nested model is a partially degenerate model, I'm using the
> RU1/RU2 commands to normalize.  When I use RU1, I add the restriction
> IVset:(wh,ot)=1.0.  However, when I run the 2 models, they "run"
> differently.  RU1 begins by running a MNL to get starting values,
> while RU2 does not, unless I call start=logit (at least not
> explicitly).  Even when I use the start=logit command on RU2, the
> value of my log likelihood function differs between RU1 and RU2 - from
> what I understand, this value should be the same (given the
> restriction with RU1).
>
> Should I always include the start=logit command?  If not, when  
> should I?
> When the command RU2 is used, is the IV parameter NLOGIT reports = (1/
> μ)?  Or does it report μ?
>
> code:
> nlogit;
> shares;
> tree = cs(corn,soy),wh(wheat),ot(other);
> lhs = lev;
> choices=corn,soy,wheat,other;
> ru2;  /*or ru1 with restriction ivset:(wh,ot)=1.0*/
> model:
> U(corn) = c_c+c_price*price+ ... /
> U(cs) = cs_cs+cs_soil*soil+.../
> U(wheat)=w_w+w_soil*soil+...$
>
> Any help is appreciated!  Thank you!
> Kandice.
>
> ---
> Kandice L. Kleiber
> Graduate Teaching Assistant
> Department of Economics
> Oregon State University
> e: kleiberk at onid.orst.edu
> w: 541.737.6628
>
> _______________________________________________
> 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




---
Kandice L. Kleiber
Graduate Teaching Assistant
Department of Economics
Oregon State University
e: kleiberk at onid.orst.edu
w: 541.737.6628

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



More information about the Limdep mailing list