Hello Sebastian,
I think what you are expecting is not what SPM is doing in that way for the
following reasons:
- The size of the clusters just emerge from the initial voxel threshold you
applied when you first
apply the contrast.
- The cluster SIZES are not undergoing some type of correction depending on
the total test volume - I think
that this maybe the core of the confusion. They are just the count of the
voxels that assemble to a cluster at the voxel treshold.
After you apply s.v.c. you'll find changed FDR-corrected p-values and
changed FWE-corrected p-values -
two different methods that correct for testing many voxels (with the same
hypothesis).
You will also find corrected/adjusted cluster p-values (uncorrected and
FWE-corrected) - they also change
as there is a different chance to get an assembly of X voxels just by
chance if your search volume is different.
So: your original cluster (and its size) is interpreted in a new way after
s. v. c. but does not change its size.
Cluster level and VBM:
The violation of non-normality is not that dramatic - and, yes, in a strict
thinking needs to be demonstrated first.
However, see Salmond et al. on this issue: usually smoothing (even in
unbalanced designs) does the job in VBM here.
The assessment of the cluster size using the random field theory for
FWE-corrected cluster p-values can be critical, yes,
as the smoothness is (or may be) non-stationary in VBM - and SnPM (Hayasaka
et al.) could give you more reliable p-values here.
Why are you not using s. v. c. and the voxel level?
There is an example from a large VBM study (SPM2) where this method was
used, and published well, maybe of interest for you:
Pezawas L, Meyer-Lindenberg A, Drabant EM, et al (2005): 5-HTTLPR
polymorphism impacts human cingulate-amygdala interactions: a genetic
susceptibility mechanism for depression. Nat Neurosci 8:828-34
hope this helps,
and other comments are wellcome,
br,
Philipp
Max-Planck-Institute of Psychiatry
NMR Research Group
Kraepelinsr. 2-10
80804 Munich
Mail: [log in to unmask]
Phone: 0049-89-30622-413
|