Mike Arenas <[email protected]> asked some questions about using the
cluster VCE with panel data.
The cluster VCE estimator requires that there are lots of clusters and
that you cluster at the highest level of correlation. Along with a few
technical assumptions, these conditions assure that the cluster VCE will
be consistent.
In the case of the fixed-effects estimator with panel-data implemented in
-areg- and -xtreg, fe-, clustering on the panel variable fulfills these
conditions, as long as the disturbances are correlated within panels but not
across panels. (As is the case, when there is within-panel serial
correlation but the disturbances are uncorrelated across panels.)
Mike hinted that he was interested in some references, so I point out that
in the case of the fixed-effects estimator with panel-data, the cluster
VCE was shown to be consistent by Arellano (1987). Wooldridge (2002)
provides a very nice textbook introduction to this issue.
The -areg, cluster- FAQ at
http://www.stata.com/support/faqs/stat/aregclust.html) was recently updated
to reflect this point.
--David
[email protected]
References
----------
Arellano, M. (1987) ``Computing robust standard errors for within-groups
estimators," Oxford Bulletin of Economics and Statistics 49, 431-434.
Wooldridge, J. (2002) ``Econometric analysis of cross-section and panel data,"
Cambridge, Mass.:The MIT Press.
*
* 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/
