>Dear SPM-team,
>
>recent discussions have revealed problems around the p- and Z-values
>before and after correction. My question: Does a Cluster containing 364
>voxels at a Z-level 3,88 with (un-)corrected k & Z of 0.002 / 0.0 show a
>significant activation in an episodic memory activation study;
>cluster-level: 0.033; voxel-level: 0.153. What is the correct way to
>address the significance level of our data?
As far as I see there is no problem whatsoever as long as you do not
violate the restrictions mentioned by Karl in previous emails, especially
regarding d.f.
In your specific example there is no contradiction between the cluster- and
voxel-level inference; the cluster level may reach significance before the
voxel-level does so, because of the extent of the cluster. In other cases
the voxel-level inference may be significant but not the cluster-level
inference if the peak is high and confined.
In principle it is preferrable to determine beforehand which inference
level you will be using.
Anecdotally, it may also be worth noting that I carried out one of the
tests proposed by Karl, i.e. the test for false positive clusters using a
contrast that should not yield any results. No false positives were found.
Rik
|