<[log in to unmask]>
<[log in to unmask]>
<[log in to unmask]>
From: Thomas Nichols <[log in to unmask]>
Date: Thu, 23 Feb 2006 8:53:06 -0500
Content-type: text/plain; charset=ISO-8859-1
Content-transfer-encoding: Quoted-Printable
Shary & Graci=E1n,
Maybe I can amplify some issues below...
> My understanding is that we cannot infer anything from the clusters (p-va=
lues and spatial location).
When there is substaintial spatial heterogeneity in the smoothness of the r=
esiduals, Shary is correct, SPM's stationary cluster size test can't be tru=
sted. (see other recent posts about this).
> I also have a question regarding a priori hypothesis and inference based=
on uncorrected voxel level p-value. Can we infer anything based on voxel=
level p-value while we have a priori hypothesis for a cluster? =20
Uncorrected voxel-level p-values have almost no infrential value: They are=
only of use if you have a single-voxel a priori hypothesis.
Regarding, =22a priori hypothesis for a cluster=22, that's a slippery thin=
g. You need to unambiguously define a cluster *before* you see your data. =
I can imagine specifying =22cluster that contains voxel x,y,z=22, or =22ne=
arest cluster to x,y,z=22, and then you need to mechanistically use that ru=
le with the SPM results interface.
That said, again, no, uncorrected voxel-level p-values are only appropriate=
if you know which single voxel to look at.
> So, what can I base my inference on when I have a priori hypothesis for a=
region?
If you have a region of interest (defined independently of your data) you c=
an use SVC to get voxel-wise corrected inferences.
Hope this helps=21
-Tom
|