Re: st: standard error calculation of marginal effects from ivprobit

["Brian P. Poi" <bpoi@stata.com>]

On Tue, 12 Dec 2006, Alejandro Delafuente wrote:

> Many thanks Richard. This is very useful. I still have a problem as all my
> marginal effects are non-significant. I wonder if this has anything to do with
> the fact that when Stata is fitting the exogenous probit model right at the
> beggining of my ivprobit it tells me that the regressor was dropped by probit.
> Can anyone tell me why would Stata do so? How can I solve it?
> I think the reason for non-significance derives from here because a similar
> ivprobit for another cross-section gives no message like this and my marginal
> effects come out significant.
>
> Here is the message after my ivprobit:
>
> Fitting exogenous probit model
> Iteration 0:   log likelihood = -3835.5461
> Iteration 1:   log likelihood = -2887.7454
> Iteration 2:   log likelihood =  -2824.864
> Iteration 3:   log likelihood = -2823.5037
> Iteration 4:   log likelihood = -2823.5018
> regressor dropped by probit; using ivreg for initial values instead...
>
> Many thanks,
> Alejandro
Alejandro,

You might want to fit a regular -probit- model of your dependent variable 
on your endogenous and exogenous variables.  Does -probit- issue any kind 
of warning about dropping a regressor?

To get initial values for maximum likelihood estimation, -ivprobit- fits a 
regular -probit- model ignoring the endogeneity issue.  If, for whatever 
reason the coefficient vector returned by -probit- does not have k 
elements, where k is the number of regressors passed to it plus one for 
the constant, then -ivprobit- calls -ivreg- instead.

In short, if you see that message and want to determine the root cause, 
fit a plain -probit- model ignoring the endogeneity issue.  The appearance 
of that message does not, however, automatically imply that the final 
results are invalid.

Hope this helps.

   -- Brian Poi
   -- bpoi@stata.com
*
*   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/