I think I have a simplier reply for Sam...
On Thu, 19 Apr 2001 Sam Reyes wrote
> I am in the process of analyzing some data on sensory effects of a drug,
> but am having some problems understanding just what the cluster level
> statistics represent. Specifically, we have one site which reaches the
> 0.004 corrected at the cluster level but only 0.08 corrected at the
> voxel level. Since it didn't rise to the level of significance at the
> voxel level how do I report the location of the activation.
I would simply say "The region is significant based on spatial extent."
> How much localizing information can I claim to have based on cluster
> level statistics.
Except in some odd situations, a significant cluster size implies that
one or more voxels within the cluster have a signal (the null
hypothesis is to be rejected). You have localized your signal to
within the cluster.
What are odd situations? Here's one example: Consider a U-shape
region of truly active voxels. Due to partial volume effects, the
observed cluster is found in the center of the U, completely outside
any of the active voxels.
...and a clarifiaction of Thomas's message...
On Mon, 7 May 2001, Thomas Kamer wrote:
> you could try to break up the whole cluster by playing with the
> threshold. [...] the significance is reported by SPM.
This isn't exactly right: SPM doesn't account for a fishing
expedition over all possible thresholds. The theory is based on a
single, a priori threshold.
Hope this helps.
-Tom
-- Thomas Nichols -------------------- Department of Biostatistics
http://www.sph.umich.edu/~nichols University of Michigan
[log in to unmask] 1420 Washington Heights
-------------------------------------- Ann Arbor, MI 48109-2029
|