Hi Laura, Phil, Colm,
These points should not make or break a person's paper.
It is good to keep the voxels size fairly close to the original size
when transforming to standard space. Going down to 1x1x1 is not
generally advised (when initial voxels were much larger). This is
probably what lead Doug to insist on getting cluster volumes in orig
space and applying them to group space. It is cheating to heavily
oversample the data (in terms of cluster requirements).
We suggest staying closer to the original size (afni_proc.py will
truncate the minimum dimension to 3 significant bits, by default,
though that is erring on the conservative side). Whether 2x2x2 is
reasonable depends on the original size. For 3x3x3, it does not
seem too bad. But for 4x4x4, it seems like too much.
Note that the blur estimates help smooth this over... :)
- rick