Eric wondered about p-values of t-stats when using the cluster option.
reg price mpg, clu(rep78)
will show that there are four degrees of freedom in that regression, so
mpg | -226.3607 82.63217 -2.74 0.052 -455.7843 3.063024
the t-stat of 2.74 has an 0.052 pvalue. But if you look up the t
distribution with 4 df, that's right..
The cluster estimator of the covariance matrix has d.f. equal to
#clusters - 1. That's why it will fail (or should!) if you have more
regressors than clusters.