With a categorical nominal (i.e. not ordered) endog var that takes on
3 (or n) values, you will have to convert it into two (or n-1)
indicator variables (e.g. by by typing -tab xvar, gen(d)- and -drop
d1-). Now you have 2 endog variables, d2 and d3, and you will need at
least two excluded instruments for your estimating eq to be
identified. Assuming you satisfy that minimal rank condition, you
should use -ivreg2- for a start (see -findit ivreg2-) so you can get
some first-stage diagnostics, even if the LPM first stage strikes you
as a bit funny. If n were 2, you could use -treatreg- instead. But
-ivreg2- is always a good place to start.
On 3/16/06, Bart Vanneste <[email protected]> wrote:
> Thanks very much for the useful reactions!
>
> I should have been more precise. I have a continuous
> dependent variable and a discrete endogenous regressor
> (with 3 categories):
> y_cont = x_discrete + e
>
> Could I also extend Rivers and Vuong (1988) when the x
> is categorical (i.e. not ordered), comparable to a
> multinomial probit?
*
* 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/
