[Limdep Nlogit List] welfare calculation in discrete choice experiments

Brett Smith bsmith at biz.uwa.edu.au
Thu Mar 5 12:54:08 EST 2009


I Agree with Mikolaj, but would like to add a little more. 

If you are to use alternate specific marginal utility of money, then this is
based on the relative prices of the alternatives (or more correctly on the
residual income available for other purchases). A way to include this in
your model is to use a non-linear (monotonic) function of (y-pj) : i.e.
f(y-pj), where f'>0 and f''<0. However, your welfare calculations get
somewhat more complicated (Kalstrom's exact method). 

If you do not wish to go down that path and you have rejected parameter
equality for the LC-MNL another way forward is to include  correlations
between alternatives (Hensher and Greene 2000 wrote an article which
included a discussion on the relationship between NL and generic/specific
parameters). It is possible that what seems to be alternate specific may be
different scales for the alternatives. The logsum formula still applies
(however it also has nesting). 

I am not too sure how to include a NL and LCM in LimDep...perhaps it is
possible to add LCM sub-command to your nested logit? However, failing this
you may wish to identify characteristics of the class members from your
LC-MNL and use a cov-het model with the variables that distinguish
membership as covariates in the inclusive values. It is not the same as the
LC-MNL but it may possibly keep some distinction between the classes. You
can use non-nested tests for preferred model. 

Good Luck,


-----Original Message-----
From: limdep-bounces at limdep.itls.usyd.edu.au
[mailto:limdep-bounces at limdep.itls.usyd.edu.au] On Behalf Of Mikolaj
Sent: Wednesday, 4 March 2009 11:16 PM
To: Limdep and Nlogit Mailing List
Subject: Re: [Limdep Nlogit List] welfare calculation in discrete choice

Caroline Johanna Biehl wrote:
> Dear all, 
> I have a question concerning welfare calculations (compensating variation
CV, Hanemann (1982)) based on a choice experiment. I did a latent class
model in nlogit and found 2 distinct classes. I have alternative specific
price effects in my model. 
> CV= 1/alpha (ln (sum exp (V0))-ln (sum exp(V1)) )
> Since alpha is supossed to equal the price effect when assuming no income
effects, i have now the problem that i have several alphas (for each
alternative). Does anyone know how to deal with that?

Dear Caroline,

In virtually all CE applications I have seen the price coefficient was 
not alternative specific. What is your interpretation of alternative 
specific price coefficient? Does it make sense for the marginal utility 
of money to vary with alternative?

Best regards,

    Mikolaj Czajkowski

    Warsaw Ecological Economics Center
    University of Warsaw

Limdep site list
Limdep at limdep.itls.usyd.edu.au

More information about the Limdep mailing list