[Limdep Nlogit List] Proportional choice data: IIA Hausman test and nested logit applicable?

[Oliver Ehrlich]

Dear all,

another two questions for the NLOGIT experts concerning the handling of
proportional choice data (= choices are not given in a 0-1-logic, but in
0.x proportions summing up to 1 across all alternatives; example: a
consumer is asked which channels he has used during a particular purchase
process to inform about the product - answer: 30% store visits, 40%
internet, 30% advertisements):

Q1: Is there a way to perform the IIA Hausman test with proportional
choice data? Key issue here is that when the MNL model is calculated with
a sub-set of choice alternatives (as required by the test), some choices
don't sum up to 1 anymore, making model estimation impossible... In the
"normal" 0-1-choice case this problem does not occur, as all observations
where the excluded alternative was the "1" choice are excluded from
estimation completely.

Q2: Additionally, does it make any sense to calculate a nested logit model
with proportional choice data? The issue here is that when the choice
proportions are located in different nests, no explicit marginal choice (=
choice of the preferred nest) exists - instead, the survey respondent has
chosen more than one nest in the given example. NLOGIT does calculate the
model in the described case - but is it ok to interpret the results the
same way as in the normal 0-1-choice case?

Thank you in advance for your thoughts on these issues,
best regards

Oliver