Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: st: re: how to explain the results of ivendog after -ivreg2-?


From   "Schaffer, Mark E" <[email protected]>
To   <[email protected]>
Subject   RE: st: re: how to explain the results of ivendog after -ivreg2-?
Date   Tue, 4 Aug 2009 18:24:35 +0100

If I can chime in with a couple of thoughts...

> -----Original Message-----
> From: [email protected] 
> [mailto:[email protected]] On Behalf Of 
> [email protected]
> Sent: 04 August 2009 18:12
> To: statalist
> Subject: Re: st: re: how to explain the results of ivendog 
> after -ivreg2-?
> 
> Dear Kit, thank you for your help and gave me a visual example!
> 
> I know your meanings.
> However, I still need to decide which one or both are 
> endogeous, which is a previous step of regression later.
> 
> Only deciding the endogeneity first can I decide the equation 
> ultimately.
> 
> Could you please give me any suggestion ?

The Durbin-Wu-Hausman version of the test reported by -ivendog- is not
robust to heteroskedasticity or clustering.  You should perhaps use
-ivreg2- with the -endog- option.  It's quite possible your test
statistics will decrease and your p-values will go up.

A more substantive point is this: do you have any sort of strong priors
about which of your two regressors is more likely to be endogenous?  If
there are good reasons to think that, e.g., x1 is very likely to be
endogenous, but x2 might or might not be, then you have good reasons to
use the specification with x2 as exogenous and x1 as endogenous, and
then test for endogeneity of x2 and exogeneity of x1 vs. this benchmark
specification.

Hope this helps.

--Mark


> Thank you very much!
> 
> Best regards,
> 
> Rose.
> 
> 
> 
> 
> ----- Original Message -----
> From: Kit Baum <[email protected]>
> To: [email protected]
> Subject: st: re: how to explain the results of ivendog after -ivreg2-?
> Date: 2009-8-4 22:33:36
> 
> <>
> 
> Rose said
> 
> Tests of endogeneity of: x1 x2
> H0: Regressors are exogenous
> Wu-Hausman F test: 5.32266 F(2,1224) P-value = 0.00499 
> Durbin-Wu-Hausman chi-sq test: 10.70872 Chi-sq(2) P-value = 0.00473
> 
> Tests of endogeneity of: x1
> 
> H0: Regressor is exogenous
> Wu-Hausman F test: 1.86877 F(1,1225) P-value = 0.17187 
> Durbin-Wu-Hausman chi-sq test: 1.89181 Chi-sq(1) P-value = 0.16900
> 
> Tests of endogeneity of: x2
> H0: Regressor is exogenous
> Wu-Hausman F test: 0.59057 F(1,1225) P-value = 0.44235 
> Durbin-Wu-Hausman chi-sq test: 0.59848 Chi-sq(1) P-value = 0.43916
> 
> 
> 
> and wants to know how these results could be obtained after 
> estimating the model
> 
> ivreg2 y z1--z8 (x1 x2= z9--z12)
> 
> Apparently if you ask "is x1 correlated with error" 
> (equivalent to using endog(x1) option on ivreg2) the evidence 
> against the null that it is orthogonal is weak. Likewise for 
> x2. If you ask "can we run the whole thing as OLS" there is 
> sufficient evidence to reject. The result of a joint test 
> cannot be predicted from those of its components; you can reg 
> y x1 x2 x3 with insignificant t's on x2 and x3 (particularly 
> if their correlation is high) but test x2 x3 will strongly 
> reject. Likewise here.
> 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/
> 
> *
> *   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/
> 


-- 
Heriot-Watt University is a Scottish charity
registered under charity number SC000278.


*
*   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/



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