So the 3dmerge blur of 5 mm was on top of whatever came out of
the scanner. The actual blur in your data is ~ 8.26, 8.74, 7.72,
which is what should be given to AlphaSim. You can average those
values across your subjects.
There have been a few related posts on measuring the blur during
the pre-processing stage. A Message Board search on the terms
"afni_proc blur" pulls up a few posts, if you are interested.
---
Regarding what threshold to convert to a pval as another input
to AlphaSim, note that a small p-value (so higher t) will mean
smaller clusters are required. I don't know what p your t=3.0
corresponds to, but 1608ul is not surprising.
It may be difficult in your case since perhaps your experiment
is subtle. So clustering may not work well for the reasons you
have stated (your stats may not survive either a high enough t
or a large enough cluster).
I don't know what those in your position state about the subtlty
of their experiment, but it is certainly relevant. Maybe you
cannot use a corrected p of 0.05.
---
Another option might be using q-values. I don't know whether
they would work better with your data, but maybe. The False
Discovery Rate method is often thought to be more conservative
than the Monte Carlo simulations method, but you can look at it.
- rick