Comment of the Day: Cosma Shalizi: "Young doesn't give a table of sample sizes, but he does give a table of average maximum 'leverages', which is something like 0.25 for the two most influential observations. This matters, because regression asymptomatics are basically about the limit where leverage goes to zero (rather than sample size to infinity as such). If two observations (or clusters) are driving 1/4 of the estimate, asymptotics are irrelevant. (Young does a lot with the bootstrap to show that the pre-asymptotics aren't great either.)...
#commentoftheday
