micro@
Books on multiple testing in linear models.
Scheffe controls error rate for ALL possible linear contrasts.
So it is conservative, but general.
I don't think it matters for post-hoc or ad-hoc, if you use Scheffe's.
But it matters for other methods.