Dear Shary and SPMers,
At 07:47 PM 7/14/2005 +0100, Shahryar Rafi-Tari wrote:
>Dear Karl/John, Attached please find two SPM2 outputs. One is showing the
>cluster level p- values (corrected/uncorrected) for a two sample analysis
>(RegularOutput.jpg) of SPECT images. The second one shows output of SVC
>(15 mm) at the same coordinates for the same comparison (decreases in the
>test group)(SVCOutput.jpg). Why in the SVC output the corrected cluster
>level p is smaller (more significant) than the uncorrected cluster level
>p- value. It is usually the opposite. Thanks, Shary
The "corrected" p-values for clusters are corrected for multiple
comparisons among all the clusters. SPM calculates the expected number of
clusters based on the random field theory, and uses this number to correct
for multiple comparisons. In case of SVC, since the search volume is so
small, this expected number of clusters could become less than 1. If this
happens, then the multiple comparison correction doesn't make much sense
and may produce spurious p-values. To verify this, I suggest comparing the
expected number of clusters (at the bottom of your SPM printout) between
your two analyses. My guess is that the expected number of clusters in your
SVC analysis is a lot smaller than 1.
If I were doing an SVC analysis, then I would just look at voxel p-values
since the multiple comparison problem is less severe compared to a
whole-brain analysis, and I wouldn't worry about cluster p-values.
Hope this helps!
-Satoru
Satoru Hayasaka ==============================================
Post-Doctoral Fellow, MR Unit, UCSF / VA Medical Center
Email: shayasak_at_itsa_dot_ucsf_dot_edu Phone:(415) 221-4810 x4237
Homepage: http://www.umich.edu/~hayasaka
==============================================================
|