# Re: st: About the consequence of doing a two step estimation manually

 From "Leda Inga" To statalist@hsphsun2.harvard.edu Subject Re: st: About the consequence of doing a two step estimation manually Date Mon, 9 Jun 2008 18:53:27 -0500

```Thank you very much. I definitely donīt want to get standard errors
that are too small, the problem is that I don't know how to it right.

My model is like this:

Y*= aZ + bC +bC^2 + e

Y*>0,  Y=1

Y*<0, Y=0

Z: exogenous regressors

As I mentioned before C is correlated with e. I have instruments (W)
but donīt know how to do a correct estimation.  ivprobit doesn`t allow
for a square term.

I asked before in statalist if this would be right, but got no answer:

ivprobit Y Z (C Csquared=W Chatsquared)

Chatsquare: square of predicted values of regressio of C on Z and W.

2008/6/9 David Greenberg <dg4@nyu.edu>:
> If you do the estimation m,anually by substituting the predicted value of a predictor into an equation for the second-stage estimation, the software will treat that predicted value as a true observed value, ignoring the fact that there is some uncertainty surrounding the prediction in the first stage. Consequently the second-stage estimation is going to produce estimates with standard errors that are too small. A question that one might ask is why you want to do this manually. David Greenberg, Sociology Department, New York University
>
> ----- Original Message -----
> From: Leda Inga <ledainga@gmail.com>
> Date: Monday, June 9, 2008 7:19 pm
> Subject: st: About the consequence of doing a two step estimation manually
> To: statalist@hsphsun2.harvard.edu
>
>
>> Dear statalisters:
>>
>> I've tried to found a way to do a correct two step estimation but
>> haven't found an answer. I have an explanatory variable correlated
>> with the error term and that variable enters the model also with a
>> square term. I know that if the estimation is done manually the betas
>> are still consistent and that the problem is the standard errors are
>> not correctly calculated. What I would like to know is if the standard
>> errors are subestimated or overestimaded? The pvalue of one of the
>> regressors of my model is very high and would like to know if the true
>> pvalue is higher or lower.
>>
>> *
>> *   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/
> *
> *   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/
>

*
*   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/
```