Joseph Coveney wrote:
>Although resampling methods are very useful in other contexts, I've never
had much success with bootstrapping or jackknifing to overcome the inflation
of Type I error rates with mixed models applied to small samples.<
Wasn't suggesting it for the mixed model, just that in a classic clustering scenario like is here, using clustered resampling or jackknifing is an alternative to the parametric assumptions on summary measures. In a small sample mixed model case, I suspect that going to a fully Bayesian model is the only way to go. Even then things may not work so well.
I look forward to hearing the answer you and David come up with.
JV
<<winmail.dat>>
