Dear Dr Giannelli,
I must admit that, in the cluster level statistics, I am not
absolutely sure what the correction means (I stick to the voxel level
I assume that you must be correcting for the expected number of
clusters. In your case, your threshold is so high (p < 0.05
uncorrected) that your expected number of clusters comes out at 0.05.
This probably explains your odd results. Normally, you might have
expected to look at the statistics of, say, 30 clusters (on the basis
of the null hypothesis) and therefore you would correct your p value
upwards (in the direction of less significance) to correct for about
30 independent comparisons (perhaps a bit less). In your case,
though, you have in fact corrected for the fact that you have made
'fewer than one independent comparisons' and therefore corrected your
significance downwards (in the direction of more significance). If
this is the case, then I suspect that you should ignore the corrected
statistics, as the fact is that you must have made at least one
If this is the explanation, then it should only be a problem when you
set your spm threshold very high, and under these circumstances every
cluster is significant even at the voxel level, so you don't really
need the cluster-level statistics. My guess is that if the expected
number of clusters is less than one, then even without an a priori
hypothesis, you can use uncorrected cluster-level statistics, and you
should ignore the spurious 'corrected' values that spm has produced.
However, this is just guesswork, and someone with more experience
will need to answer this one. Sorry.
>Content-Type: TEXT/plain; charset=us-ascii
>Sorry, but in previous message I am not clear.
>In previous Z map I have in cluster-level:
>and Puncorrected is rightly less than Pcorrected (0.068 < 0.078).
>Nevertheless I have some Z maps where, in the cluster-level,
>Pcorrected is less than
>Puncorrected (Pcorrected < Puncorrected): I don't understand.
>I put an example in attachment.
>Thank you for your help again.
>Content-Type: APPLICATION/postscript; name="file1.ps"; x-unix-mode=0660
>Attachment converted: Ravel:file1.ps (TEXT/mlpr) (000634A5)
from: Dr Richard Perry,
Clinical Lecturer, Wellcome Department of Cognitive Neurology,
Institute of Neurology, Darwin Building, University College London,
Gower Street, London WC1E 6BT.
Tel: 0207 679 2187; e mail: [log in to unmask]