st: re: clustering residuals

[Kit Baum <BAUM@BC.EDU>]

Anita wrote

When do you require residuals to be clustered in
regressions? How do you go about doing it in Stata?

For example I run pooled OLS for biateral country-pair
regression. Should I suspect residuals for i and j may
be correlated..Or is it routine to assume such
correlation in pooled estimations?


The notion of cluster-robust standard errors is to allow each cluster  
to have its own error variance, and within each cluster, the  
assumption of independence of errors is relaxed. It would be very  
sensibly applied to bilateral country-pair data if pooled OLS is  
encountered.

However, with this sort of data (with which I have worked  
extensively) you are making an assumption, in pooled OLS, that there  
is no unobserved heterogeneity. If there are common elements to a  
country-pair that might influence their errors over time, it is also  
sensible that there might be a 'special relationship' between country- 
pairs that might cause their relationship to differ from other  
country-pairs. In other words fixed effects by country-pair might be  
very sensible. If you create a country-pair indicator variable and - 
xtreg, fe-, and the F-test at the foot of the model rejects its null,  
it is saying that the pooled OLS model is delivering inconsistent  
point estimates. This problem cannot be rectified by adjusting the  
standard errors.


Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.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/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/