Dear Shary,
1)- By changing the extent
threshold (PET/SPECT/VBM) from zero to, for instance 100, the set level
p-value is affected because the expected number of clusters and the
actual number of clusters are also affected. Am I correct?
Yes, the probability of getting a cluster deceases with its
size.
2)- Change of extent threshold
does not affect corrected cluster level p-value in my experience.
Although it does change expected number of clusters shown at the bottom
of the SPM output. So, this does not aid multiple comparison
corrections as I test it on my data.
That is right. The extent threshold simply filters out small
clusters.
Only the set-level p-value is affected by this.
3)- This is
strange: when I increase the extent threshold I get smaller number of
expected clusters which even becomes less than 1 (changes from 2.44 to
0.37). Then, why the corrected cluster level p is not smaller than
uncorrected in this instance based on one of your (or Will's) previous
e-mails? Is it because the uncorrected height threshold is not that
high (<0.001) as it happens when I use FWE (<0.000000000) where I
get smaller corrected cluster level P compared to uncorrected p?
Still, the expected number of clusters is less than one and should affect
the corrected cluster level p-value. Should it
not?
No. The corrected cluster level p is the chance probability of
getting a cluster of the clusters size (or larger) in the search
volume.
This does not depend on the extent threshold. It depends on the
expected
number of clusters of any size and the uncorrected cluster p-value.
Only the probability of getting the number of large clusters (or more)
depends on the extent threshold. This threshold determines how
large
the cluster has to be.
I hope this helps - Karl