> that you can't say anything about the rest of the voxels in the cluster
When relying on peak-level correction then you can at least interpret the local maxima to be significant (assuming the threshold was set correctly). This can be an advantage over cluster-level correction. In practice it turns out people relying on peak FWE correction report one (or a few) peak coordinates and the peak statistics plus k instead of every single voxel, which brings us to the debate about voxelists, topologists, closet topologists, see https://www.jiscmail.ac.uk/cgi-bin/webadmin?A2=spm;76e830ef.1005 .
> If you specify the mask as the first step of viewing the results, and then just threshold at p(FWE)=0.05, you should get volume-corrected
> results in one step
The threshold would still correspond to .05 FWE on whole-brain. The mask basically hides voxels/clusters falling outside the small volume, but it does not affect the statistics in the way one would expect for SVC. In case parts of a cluster fall outside of the mask its peak statistics are affected, but this is based on the smaller k within the mask with regard to the whole brain properties (and not with regard to SV properties).
There's a general limitation as SVC will always rely on the smoothness estimate from the whole brain. Accordingly, results obtained via SVC are not identical to those when restricting the whole-brain analysis to the smaller volume already during model specification.
Best
Helmut
|