This has been brought up on the list a few
times in the past, but let me give you a quick and simple explanation. Corrected
cluster-level p-values are corrected for the number of expected clusters. You
can see this number at the bottom of the SPM output. For certain combinations
of parameters (FWE threshold, smoothness, df, etc), however, this expected
number of clusters actually becomes smaller than 1. Then SPM tries to correct
for multiple comparisons among, say 0.25 clusters (or whatever the number of
expected clusters is), which is quite counterintuitive. You can read more
details on:
http://www.jiscmail.ac.uk/cgi-bin/wa.exe?A2=ind05&L=SPM&P=R306103&I=-3
It seems like your voxel survived the
voxel-level FWE threshold. But if you want to report cluster p-values for some reason,
then you might consider lowering your threshold (p<0.005, 0.01, etc,
uncorrected). It should produce more sensible cluster p-values while voxel
p-values for local maxima remain the same.
Good luck!
-Satoru
From: SPM (Statistical
Parametric Mapping) [mailto:[log in to unmask]] On Behalf Of Manish Dalwani
Sent: Tuesday, May 27, 2008 3:51
PM
To: [log in to unmask]
Subject: [SPM] cluster p-value
question
Hi SPM'ers,
I am using the small volume correction for certain regions and then when I
check the corresponding p values for voxel-level and cluster-level threshold, I
notice that the p-value for p-uncorrected at cluster level is > than
p-corrected (look at the example below). Can someone kindly explain why?
Example:
Right Inferior Orbito Frontal Gyrus BA47 (X,Y,Z = -20, 30, -10 {mm}):
Cluster-Level: Pcorrected = 0.132
Puncorrected=0.699
Voxel-level: PFEW-corr = 0.017
PFDR-corr = 0.014
Puncorrected = 0.002
T = 3.05
Z = 2.87
Regards,
Manish Dalwani
Sr. PRA
Dept. of Psychiatry