I' estimating a 2-level nested logit. I have no choice-specific variables -
all are time-invariant individual and household characteristics, and all
should (potentially) be included as explanatory vbles in both levels. For
the bottom level it's fine. For the upper level, I then include both the
log denominator as an additional regressor as well as the same regressors as
in the bottom level. I get identification due to the fact that the log
denominator is a non-linear function of the regressors. However, I'm not
sure if this is a correct thing to do, as I've never seen it been done like
this before (normally there are choice specific vbles). Also, not sure
about the interpretation to be placed on the regressors at the upper level?
Would be grateful for any thoughts,
Emla.
_________________________________________________________________
Find a cheaper internet access deal - choose one to suit you.
http://www.msn.co.uk/internetaccess
