Hi Guiseppe,
It's taking some time for me to absorb your points, I do not
pretend to be a statistician.
Please correct me if I'm wrong, but it seems like that means if
the data has a high variance due to a subject effect, it should
be considered less spatially correlated, and therefore require
smaller clusters to find significance. Or from the other side,
if there is a fantastic effect from the model, with lots of low
variance, should the data should be assumed to be blurry?
I believe the blur values are supposed to be a measure of the
variance of the noise in the FMRI data, not of the noise in some
particular ANOVA model.
It seems circular for it to depend on the noise in the model.
Then a bad model would require smaller clusters, while a good one
would require larger clusters.
Do you feel that I am off-base here?
- rick