Multinomial Logit with Instrumental Variables

[Tricia Chia]

Hi,

Was just wondering if anyone knew of an existing programme which would allow me to do a multinomial logit with instrumental variables?

Your help will be very much appreciated!

Thanks.

[Anat Tchetchik]

Hi Tricia, I am not aware of such possibility. I wonder why would one want to use instrumental variables in MNL. As far as I know MNL is designed at estimating choice probabilities. The question is what source of endogeneity do you expect to have in choice models? but again.. I may be wrong.

[FernandoRios]

Hi Tricia
In a recent paper, I got ask the same question. As far as I know, and the referees suggested, one option could be a Residual inclusion approach or Control function approach.
Hope this gives you some starting point to answer the question
Fernando

[Phil Bromiley]

You might check cmp - it does many things and might do this. Alternatively, you might be able to do something in GSEM.

[David Roodman]

cmp can do instrumented multinomial probits. A simple example with a fake instrument is:

An example with a fake instrument is:

webuse sysdsn3
gen instrument = age + runiform()
cmp setup
cmp (insure = age male nonwhite site2 site3, iia) (age = instrument male nonwhite site2 site3) , nolr ind($cmp_mprobit $cmp_cont)