> I have a set of five regions that I have selected from the literature that I
> would like to evaluate in my data. I got the MNI coordinates for those
> regions from the literature. I first pull up my SPM.mat file using the
> Results button. I viewed the data using p=.05 and then plugged the
> coordinates into the coordinate boxes in the SPM Results window. I then
> pressed the small volume corrections button and specified a 10mm sphere
> around that region. From here, I am only choosing regions that have the
> family wise corrected p-value less than .05.
> My main question concerns the corrected p value. When I initially pull up
> the results file, is it 100% necessary to use FWE .05 and then use small
> volume correction, or can I use some uncorrected p value (i.e. .001, .05,
> etc.) and then only select the peaks that have a FWE corrected p value less
> than .05?
Yes, you should be fine to use a different threshold for the initial
results—it's the statistics that are output from the SVC button that
are the relevant ones.
> Also, is it better to report peak-level or cluster level
> statistics in SPM? If the peak is significant but the cluster is not, is it
> worth reporting? I'm a little confused about what is acceptable to do here.
If a voxel reaches corrected significance, you're definitely justified
in reporting it. The cluster vs. voxel level has to do with the level
of inference you can make about the results. See e.g.
For a small SVC (which seems redundant, but in theory you can use any
size volume to constrain results using SVC), I don't know how
meaningful a cluster-level correction is; I think voxel-level
corrections are more often reported. Though I don't know that cluster
level is incorrect, per se. This is mentioned briefly in an old post
by Karl here:
Hope this helps!
Dr. Jonathan Peelle
Department of Neurology
University of Pennsylvania
3 West Gates
3400 Spruce Street
Philadelphia, PA 19104