[Limdep Nlogit List] MNL: Quantities greater than one

feinf at umich.edu feinf at umich.edu
Tue Jan 16 14:22:53 EST 2007


Hi Raimund,

I do think I was misunderstanding your modeling scenario. If you are
talking about /sequential/ choices, then an MNL setting is entirely
approprite, if you want to modify the utility structure of the second
choice conditional on the first. There is a huge literature in
marketing, for example, on such models, usually containing the terms
"variety-seeking" or "inertia" or both.  Probably a good place to
start would be the work of PB Seetharaman and PK Chintagunta; they
have a paper on formally combining such models.  Seetharaman has in
fact written a number of papers on such "first-order" choice models,
as have many others in marketing (one particularly sophisticated
example is the so-called "lightning bolt" model of Roy, Chintagunta
and Haldar). [The general techniques are usually subsumed under
"state dependence" and "habit persistence".]

The main idea in these models is nearly always the same: changing the
formal specification of subsequent choices based on prior ones,
usually only the last one (Kahn, Kalwani and Morrison provide one of
the few counterexamples, but their model is very primitive by today's
standards, being individual-level, so that it's pointless to try to
estimate parameters from real data, like from a panel).

In any case, I suspect a Google Scholar search of these papers would
bring up a lot of the current research in the area. Nowadays, almost
everyone uses Hierarchical Bayes models to allow for individual level
parameters, but shrinkage to an overall mean. LIMDEP doesn't support
these as far as I know, but does allow for some forms of
heterogeneity, if that's critical to your application.  Prof. Greene,
of course, knows dramatically more than I do about such matters.  Good
luck with your application,

Fred

=====

Prof. Fred Feinberg
Ross School of Business
University of Michigan
/feinf at umich.edu/

  =====Quoting Raimund Bau <Raimund.Bau at rwth-aachen.de>:

> Dear Prof Greene, Prof Feinberg,
>
> Thank you very much for your valuable answers to my question.
>
> Maybe I will have to change my model and my entire approach. I
thought
> that if a consumer chose the same product on two different
occasions
> the overall fit of the MNL model would improve if the relative
utility
> of that product was assigned a higher value compared to the
purchase of
> one product of the same brand only.
> Now as far as I understand, you are saying that if the consumer
> exhibits the same behavior on a single occasion Nlogit/MNL would
treat
> it as if the consumer bought just one item of that specific brand
> instead of treating it as if the consumer chose the brand twice.
Thus
> the same consumer behavior would be treated differently if my first
> assumption is right.
>
> Just to make sure we are discussing the same subject: I am not
thinking
> of situations where a consumer chooses two different brands from
the
> same category on a single occasion, which obviously would show a
lack
> of consistency in terms of utility maximization within the bounds
of
> such models. Furthermore, I am not really interested in quantities
but
> rather brand choice and price sensitivity depending on advertising
> expenditures. Essentially my work is similar to Keane et al in 2004
who
> used a mixed MNL as well.
>
> Thank you once again!
>
> Raimund Bau
> RWTH Aachen
>
>
> _______________________________________________
> Limdep site list
> Limdep at limdep.itls.usyd.edu.au
> http://limdep.itls.usyd.edu.au
>
>
>





More information about the Limdep mailing list