st: Re: STATA heteroscedasticity test

[Christopher F Baum]

<>
I am not aware of any test specifically for xtreg, re, but you could  
certainly use robvar to test for groupwise heteroskedasticity:

webuse grunfeld
xtreg invest mvalue kstock
predict double eps, e
robvar eps, by(company)

Please see http://ideas.repec.org/a/tsj/stataj/v6y2006i4p590-592.html   
for details.

With regard to collinearity, as the regressors do not change when  
using fe vs. re vs. OLS, you could use any collinearity diagnostics  
from pooled OLS to consider this. The one thing they would not pick up  
is potential collinearity between the Xs and the dummy variables  
implicit in the within transformation. High pairwise correlations  
imply collinearity, but not vice versa. I would recommand considering  
variance inflation factors (estat vif) for this purpose.

It is a good idea to address questions like this to Statalist, where I  
and many others knowledgeable in the subject may answer.

Kit Baum   |   Boston College Economics and 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



On Mar 16, 2009, at 08:16 , Svetlana Jefimova wrote:

> Dear Mr F Baum,
> We are Svetlana Jefimova and Irina Beinarovica and we are writing to  
> You from Stockholm School of Economics in Riga (Latvia). We are  
> writing a bachelor thesis on the topic of FDI into Turkey. We are  
> researching the determinants affecting FDI inflow using gravity  
> model. Consequently, our approach is very similar to the work of  
> Ludo Cuyvers, Joseph Plasmans, Reth Soeng & Daniël Van den Bulcke  
> "Determinants of Foreign Direct Investment in Cambodia: Country-
> Specific Factor Differentials", whom You advised on some STATA  
> commands. Therefore, we would like to ask for Your advice on several  
> issues on STATA commands for random effects model.
> 1) Is there any test for heteroscedasticity in the random effects  
> model? Wald statistics seems applicable just for fixed effects or  
> can be applicable as well?
> 2) Is there any test for multicollinearity in random effects model?  
> Or just "corr" is enough?
>
> We would really appreciate Your help, as we do not know whom else to  
> approach!
> Best reagrds,
> Irina and Svetlana

*
*   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/

[Christopher F Baum]

Well, -help xtreg- shows that you can use a robust or cluster-robust  
VCE with the RE estimator. Given that -robvar- identifies groupwise  
heteroskedasticity, it would be most sensible to use a cluster-robust  
VCE with the panel variable as the clustering variable.


Kit Baum   |   Boston College Economics and 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



On Mar 16, 2009, at 13:07 , Svetlana Jefimova wrote:

> Thank You very much again, and I feel terrible to disturb You with  
> one more question (hopefully last one), but what could we do to amke  
> the model homoscedastic? Or can we still use results obtained as  
> unbiased?
>
> 2009/3/16 Kit Baum <[hidden email]>
> Here is the Stata Tip, which should explain the interpretation of  
> these stats...
>
>
>
>
> Kit Baum   |   Boston College Economics and 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
>
>
>
> On Mar 16, 2009, at 12:40 , Svetlana Jefimova wrote:
>
> Dear Mr Baum,
>
> Thank You for a promt reply, which we found very useful!
> However, we would like to ask again for a favour, as we have no  
> available funds to aquire Your work, we would like to ask if You  
> could send it to us(Stata Tip 38), or just help to interpret the  
> results we aquire from the robvar test :
>
>  W0  = 2.0786285   df(47, 165)     Pr > F = .00038932
>
> W50 = .88411755   df(47, 165)     Pr > F = .68287619
>
> W10 = 2.0786285   df(47, 165)     Pr > F = .00038932
>
> Does it mean we have heteroscedasticity or not?
>
> Again, sorry for disturbunce!
>
> Irina and Svetlana
>
> p.s. We wrote to statalist, however, we have not received any reply  
> yet!
>
>
> 2009/3/16 Kit Baum <[hidden email]>
> <>
> I am not aware of any test specifically for xtreg, re, but you could  
> certainly use robvar to test for groupwise heteroskedasticity:
>
> webuse grunfeld
> xtreg invest mvalue kstock
> predict double eps, e
> robvar eps, by(company)
>
> Please see http://ideas.repec.org/a/tsj/stataj/ 
> v6y2006i4p590-592.html  for details.
>
> With regard to collinearity, as the regressors do not change when  
> using fe vs. re vs. OLS, you could use any collinearity diagnostics  
> from pooled OLS to consider this. The one thing they would not pick  
> up is potential collinearity between the Xs and the dummy variables  
> implicit in the within transformation. High pairwise correlations  
> imply collinearity, but not vice versa. I would recommand  
> considering variance inflation factors (estat vif) for this purpose.
>
> It is a good idea to address questions like this to Statalist, where  
> I and many others knowledgeable in the subject may answer.
>
> Kit Baum   |   Boston College Economics and 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
>
>
>
>
> On Mar 16, 2009, at 08:16 , Svetlana Jefimova wrote:
>
> Dear Mr F Baum,
> We are Svetlana Jefimova and Irina Beinarovica and we are writing to  
> You from Stockholm School of Economics in Riga (Latvia). We are  
> writing a bachelor thesis on the topic of FDI into Turkey. We are  
> researching the determinants affecting FDI inflow using gravity  
> model. Consequently, our approach is very similar to the work of  
> Ludo Cuyvers, Joseph Plasmans, Reth Soeng & Daniël Van den Bulcke  
> "Determinants of Foreign Direct Investment in Cambodia: Country-
> Specific Factor Differentials", whom You advised on some STATA  
> commands. Therefore, we would like to ask for Your advice on several  
> issues on STATA commands for random effects model.
> 1) Is there any test for heteroscedasticity in the random effects  
> model? Wald statistics seems applicable just for fixed effects or  
> can be applicable as well?
> 2) Is there any test for multicollinearity in random effects model?  
> Or just "corr" is enough?
>
> We would really appreciate Your help, as we do not know whom else to  
> approach!
> Best reagrds,
> Irina and Svetlana
>
>
>
>
>

*
*   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/