st: Estimating Nested Logit Model for a Choice Experiment

[Tim Bolt <timbolt2@gmail.com>]

Dear Stata List,

I am trying to help in estimating a model for a choice experiment.
The non-nested version was estimated fine with "xtlogit" as a random
effects model with multiple responses per participant.

However, the decision structure is likely nested in the presented
options, and hence we are keen to estimate this as a nested logit
model.

Using "nlogit" after "nlogitgen" for creating the nesting variable
does seem the appropriate commands if it were a single choice among
options, as the example discussed in the thread which started with
http://www.stata.com/statalist/archive/2007-08/msg00561.html
demonstrates.  However, the example data in this thread has only one
positive response ("Choice") for each participant ("Id").

As typical for discrete choice experiments, our data is built around
each respondent answering a series of choices so this is not adequate
for us.

The nested model estimated with "nlogit" is troubled by the multiple
positive responses per respondent.
Unsurprising, the output is:
>>  note: 350 cases (4176 obs) dropped due to no positive outcome or
>> multiple positive outcomes per case.

Can Stata handle this kind of model?  Any suggestions?

Thank you!
Tim
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/