Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
| From | "Kloek, TG (Gijs)" <gijs.kloek@achmea.com> |
| To | "Kloek, TG (Gijs)" <gijs.kloek@achmea.com>, "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu> |
| Subject | st: RE: IV Regression and zero coefficients |
| Date | Wed, 21 Aug 2013 10:39:55 +0200 |
Dear Statalist,
With the help of Joao Santos Silva I have found the solution. The problem could be solved by dividing very large regressors by one million. As a rule one should try to ensure that the y and the regressors are all of comparable magnitudes. The problem appears to have been solved in Stata 13.
Kind regards,
Gijs Kloek
-----Oorspronkelijk bericht-----
Van: Kloek, TG (Gijs)
Verzonden: woensdag 14 augustus 2013 21:38
Aan: 'statalist@hsphsun2.harvard.edu'
Onderwerp: IV Regression and zero coefficients
Dear statalist,
I have tried to do a regression using an instrumental variable. I have used the command "ivregress 2sls". The result for the coefficient in the second stage is zero. That has surprised me. The commands "ivreg" or "gmm" lead to non zero coefficients and are more in line with my expectations. I am not the only user with this problem. There are two more similar messages in the Statlist:
Link 1: http://www.stata.com/statalist/archive/2011-07/msg00984.html
Link 2: http://www.stata.com/statalist/archive/2011-02/msg00764.html
Could anyone explain this, please?
Below you can find the command and the results:
. ivregress 2sls gcrolus (delchasf = delmsci) delsqcatindus, robust
Instrumental variables (2SLS) regression Number of obs = 11
Wald chi2(2) = .
Prob > chi2 = .
R-squared = 0.8229
Root MSE = .10261
------------------------------------------------------------------------------
| Robust
gcrolus | Coef. Std. Err. z P>|z| [95% Conf. Interval]
-------------+----------------------------------------------------------------
delchasf | 0 2.86e-23 0.00 1.000 -5.61e-23 5.61e-23
delsqcatin~s | 5.60e-11 1.87e-12 29.98 0.000 5.23e-11 5.96e-11
_cons | (omitted)
------------------------------------------------------------------------------
Instrumented: delchasf
Instruments: delsqcatindus delmsci
Kind regards,
Gijs Kloek
**************************************************************************
The information in this e-mail message is intended only for the use of the addressee. Transmission to and use by any other than the addressee is prohibited. E-mail transmission is not guaranteed to be secured or error free. Therefore no rights can be derived from the information received.
**************************************************************************
*
* 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/