Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: RE: Problem with ivhettest after ivreg2


From   "Schaffer, Mark E" <[email protected]>
To   <[email protected]>
Subject   st: RE: Problem with ivhettest after ivreg2
Date   Fri, 1 Feb 2008 18:49:14 -0000

Ivan, 

> -----Original Message-----
> From: [email protected] 
> [mailto:[email protected]] On Behalf Of Ivan Etzo
> Sent: 31 January 2008 19:49
> To: [email protected]
> Subject: st: Problem with ivhettest after ivreg2
> 
> Dear all,
> 
> in order to test the presence of heteroskedasticity in  my 
> panel I estimated the model with explicit dummies using 
> -ivreg2- and then I runned -ivhettest -. Now the problem is 
> that I get different results depending on whether I specify 
> or not a potentially endogenous regressor. That is, if the 
> model I test is the following
> 
> ivreg2 depvar x1 x2 x3 r2-r20 (x4=l.x4 + other instrum) 
> 
> where r2-r20 are the explicit dummies (regions in my case), 
> then after running -ivhettest - I get  P-value = 0.9997 which 
> means that there is no heteroskedasticity
> 
> but 
> 
> If I test the following alternative model
> 
> ivreg2 depvar x1 x2 x3 r2-r20 x4
> 
> then after running -ivhettest - again I get  P-value = 0.0096 
> which rejects the null. 
> 
> It sounds very weird for me that there is no 
> heteroskedasticity (it is a panel of regions...), moreover I 
> also checked whether the suspected regressor is endogenous 
> including the option endog (x4), and the p-value doesn't 
> reject the null that x4 is exogenous.

Can you tell us more about the specific tests that are reported by
-ivhettest-?  In particular, are the tests comparable?

-ivhettest- defaults to reporting a test based on the levels of the IVs.
In the x4 exogenous case, this would be all the regressors; in the x4
endogenous case, this would be all the exogenous regressors plus all the
instruments.  The two tests will have different degrees of freedom,
different power, etc.

You can get comparable test statistics if you use, say, the
-fitlev- or -fitsq- options.  This will report tests based on the fitted
level, or the fitted level and the square, of the dependent variable
depvar.  It will have the same degrees of freedom in the x4 exog and x4
endog cases.  Intuitively, the test will look in roughly the same
"directions" for signs of heteroskedasticity, and the only differences
that result can be traced to the differences in the fitted value of the
dependent variable that the two estimations generate.

Cheers,
Mark (-ivhettest- author)

Prof. Mark Schaffer
Director, CERT
Department of Economics
School of Management & Languages
Heriot-Watt University, Edinburgh EH14 4AS
tel +44-131-451-3494 / fax +44-131-451-3296
email: [email protected]
web: http://www.sml.hw.ac.uk/ecomes

> Maybe I'm doing some mistake....Does anybody know why I get 
> different results? And which result is telling the truth? 
> 
> Thank you
> 
> Ivan
> *
> *   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/



© Copyright 1996–2024 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index