Re: st: FW: appropriateness of cluster option with xtreg, re

[Alfonso Sanchez-Penalver <oneiros_spain@yahoo.com>]

Hi Eva,

Clustering the VCE can lower the estimates of the standard errors. You can think of it in a similar way that FGLS provides more efficient estimates of the errors than OLS in the presence of heteroskedasticity and/or autocorrelation. As Federico mentioned, the purpose of clustering the errors by panel is precisely this, to avoid the bias in the estimation of the standard errors when you don't cluster. The fact that they are decreasing seems to indicate that without clustering the standard errors were biased upwards, which goes back to the similarity between OLS and FGLS estimates of the standard errors I mentioned before.

I don't see a problem with your data that ought to stop you from clustering, either.

Best,

Alfonso Sánchez-Peñalver

> On Oct 14, 2013, at 5:03 PM, Eva Wölbert <evawoelbert@hotmail.com> wrote:
> 
> Hi Federico,
> 
> thanks a lot for taking the time to answer my question, that makes sense to me. 
> 
> As specifiying cluster(panelvar) in my case reduces the standard errors, I would like to make sure that I am not doing anything wrong. I have an experimental dataset with 2200 observations, that is 44 observations for each of 50 subjects.
> 
> My main predictors are dummies, one varies within subject, and one between subjects.
> 
> Is my dataset large enough for clustered standard errors? An what should I conclude about the data from the fact that standard errors go down when clustering?
> 
> Thanks, 
> Eva
> 
> 
> 
> 
>> Subject: Re: st: FW: appropriateness of cluster option with xtreg, re
>> From: f.belotti@gmail.com
>> Date: Mon, 14 Oct 2013 10:19:05 +0200
>> To: statalist@hsphsun2.harvard.edu
>> 
>> Dear Eva,
>> 
>> in my opinion the answer is: you should.
>> 
>> The fact that you are including in your model unit-specific intercepts (which are assumed to be "mean independent" from the exogenous explanatory variables in the case of -xtreg, re-) does not rule out the possibility that regression residuals exhibit heteroskedasticity and/or serial correlation, that is their variance is not constant between panel units and/or is correlated over time. These issues lead to inconsistent covariance matrix estimates and, therefore, to faulty statistical inference. The option -cluster(panelvar)- allows to avoid this risk.
>> 
>> You should take a look at 
>> 
>> Moulton, Brent R., 1986. "Random group effects and the precision of regression estimates," Journal of Econometrics, Elsevier, vol. 32(3), pages 385-397, August.
>> 
>> http://www.stata.com/meeting/13uk/nichols_crse.pdf
>> 
>> Hope this help.
>> Federico
>> 
>> 
>>> On Oct 13, 2013, at 8:07 PM, Eva Wölbert wrote:
>>> 
>>> Dear statalist,
>>> 
>>> I would like to know whether I can / should specify cluster(panelvar) when
>>> using xtreg, re. 
>>> 
>>> This seems to account for the panel structure of the data "twice". Moreover, I am confused because
>>> some of my standard errors are going down with the cluster option. 
>>> 
>>> Thanks in advance
>>> 
>>> Eva 
>>> *
>>> * 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/
>> 
>> -- 
>> Federico Belotti, PhD
>> Research Fellow
>> Centre for Economics and International Studies
>> University of Rome Tor Vergata
>> tel/fax: +39 06 7259 5627
>> e-mail: federico.belotti@uniroma2.it
>> web: http://www.econometrics.it
>> 
>> 
>> *
>> * 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/                         
> *
> *   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/

*
*   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/