Mary Burke
>
> I am looking for Stata code for multinomial probit estimation. I
> have seen
> inquiries concerning the same topic, with no replies. Has anything been
> developed that anyone know of? I greatly appreciate any clues anyone can
> provide.
You're right. This is a frequently asked question, and
from various fragmentary or non-existent replies, the
best assumption is that no such Stata program exists.
What follows is my understanding, written because no one will
seemingly write this all down (although I think I remember
part of this being mentioned on Statalist a while back), not because I
am an expert on the matter.
The most obvious reason why MNP doesn't exist in Stata is technical,
centring on calculation of the needed multivariate normal probabilities.
Greene's text on econometrics provides one sobering account,
which is probably more familiar to you than it is to me.
In essence, official Stata doesn't, at least at present, provide the most
crucial machinery, and it would be a very brave (or possibly very foolish)
Stata user-programmer who tried to program it from scratch
in interpreted code. It really isn't the kind of thing to write
in Stata's own language. The machinery needed must be
written in C for speed and with a very, very careful eye on getting
the numerical and the statistical sides correct, and that means that
if this is done it must be done by Stata Corp.
Don't ask, however, whether Stata Corp will provide this. It is
surely on their "to think about" list, but anything as big as this
would be announced at most one week before it was released.
Nick
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
