Re: st: Estimate endogenous fractional response variable

[Christopher Baum <kit.baum@bc.edu>]

<>
On Dec 8, 2011, at 2:33 AM, Andreas wrote:

> I would like to estimate a 2SLS model where the endogenous variable is a fractional response variable with values between 0 and 1.
> However, the ivreg command only allows for simple linear regressions in both the first stage and the reduced form. Is it possible to conduct the IV estimation manually by transforming the dependent variable into a log-odds ratio as proposed by Papke and Wooldridge (1996)? So instead of 
> 
> ivreg r lmval capint (drank = ffloat age)
> 
> I would apply the following two-step procedure:
> 
> regress drank_transf ffloat age lmval capint   // using drank_transf = log(drank/(1-drank))
> predict drank_fitted, replace
> regress r drank_fitted lmval capint
> 
> Now I am not sure if this manual approach consistently replaces the ivreg procedure in the usual linear setting. If so, I hope it produces unbiased estimates for the fractional response variable in this special case as well. But how can I specify a test for exogeneity of the excluded instruments ffloat and age without making use of the standardized ivreg procedure?

Don't try to do 2SLS "by hand"--you will get the wrong residuals, the wrong sigma^2, the wrong r^2, etc.  Use Baum-Schaffer-Stillman's -ivreg2- from SSC on the transformed endogenous variable. Our 2003 and 2007 SJ papers (freely downloadable) explain all of the testing options that you can apply in estimation.

Kit

Kit Baum   |   Boston College Economics & DIW Berlin   |   http://ideas.repec.org/e/pba1.html
                             An Introduction to Stata Programming  |   http://www.stata-press.com/books/isp.html
  An Introduction to Modern Econometrics Using Stata  |   http://www.stata-press.com/books/imeus.html


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