Over the years I have gradually become more consistent in calling p-values from 0.02 to 0.08 borderline. There’s no reason for those cutoffs other than personal experience, also known as “making the same mistakes with increasing confidence [npi] over an impressive number of years“. Type 1 errors (which never really happen, but you know what I mean) have uniformly distributed p, so that’s where you’re going to find them. Just thought I’d say that.
Published by Robert