0
$\begingroup$

As is stated in the title, I have several questions concern the combination of nested logit and panel data.

1, Can we run nested logit with the individual fixed effect? this is useful when one has data that individual makes multiple decisions over time and the choice structure is nicely nested.

2, If yes, could you provide a real case (a paper or a tutorial would be the best)?

3, If no, is there any reason, especially theoretical ones, that prohibits us from doing so?

Any effort would be deeply appreciated.

$\endgroup$
0
$\begingroup$

Probably someone can provide a more complete answer than this, but I think the short answer is, in principle, you can do it, but it is tricky. Multinomial logit is a nonlinear model, so things like differencing out the fixed effect (as is common in linear models) aren't going to work here. Instead, you will probably need the number of time periods to be large (i.e. go to infinity) so that you can consistently estimate the fixed effects. Even then, things still will be tricky.

$\endgroup$
  • $\begingroup$ Thanks for your reply, you are making a good point that makes things more clear to me. In the case of conditional logit, the individual fixed effect indeed is 'integrated' out rather than 'differencing' out in linear regression. When it comes to nested logit, I tried to add individual fixed effect in the equation of one level in Stata, no error reported. But I totally agree with you about the theoretical consistency of such estimation. Hope someone can work this out, in theory. $\endgroup$ – Jason Goal Apr 17 at 17:26

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.