Re: st: 2SLS with first stage duration model

[Jorge Eduardo Pérez Pérez <perez.jorge@ur.edu.co>]

This looks similar to the "forbidden regression"

http://people.duke.edu/~ay32/metrics/Regression.pdf

Also look at Angrist & Pischke, Mostly Harmless Econometrics, p.p. 190.

Since your first stage is nonlinear and the second stage is linear. So
probably the method you are suggesting is not a good idea. A probably
better idea , in line with this literature, is to use SP_hat as an
instrument for SP in the second stage. But I am not sure if this works
in your setting.





_______________________
Jorge Eduardo Pérez Pérez


On Fri, Apr 5, 2013 at 2:26 PM, Karen Ruckman <ruckman@sfu.ca> wrote:
> hello statalisters,
>
> i am trying to run a 2 stage endogenous model where the first stage is a duration model with a log-log distribution and the main stage is OLS.  it appears that neither -gllamm- or the program aML can support a log-log (or even log-normal) distribution.  i would like to now explore trying to create a 2SLS regression on my own.
>
> if my endogenous system looks like this:
> SP = bX + cZ + error1  (the duration model with log-log distribution where Z are the instruments for SP)
> AG = dSP + eX + error2 (the OLS model)
>
> would the following be the proper method to account for the endogeneity?
> 1. regress SP on X and Z using the appropriate survival analysis, get the estimated SP (SP_hat)
> 2. regress AG on SP_hat and X using OLS.
>
> karen.
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/statalist-faq/
> *   http://www.ats.ucla.edu/stat/stata/
>
>


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/