Dear Sean,
> If both the P(n>K)uncorrected and corrected values are significant but
> the voxel P(t>u) corrected values are not significant within the cluster,
> can I still be confident in stating the cluster is indeed significant
> despite the voxel p-values (corrected).
In general you should use either the cluster level of inference, or the
voxel level of inference, but not both. Strictly speaking you should make
your decision about which level of inference to use before analysing the
data. It is perfectly acceptable to decide to use either (but not both!).
> If the cluster P-values are significant but the voxel P-values
> (corrected)are not, what exactly does this mean?
The cluster level of inference takes into account both the peak height and
the spatial extent of the cluster when assessing significance, while the
voxel level of inference assesses height only. So at the cluster level both
(very) large but weakly activated blobs and small but very strongly
activated blobs can (in principle) reach the same level of statistical
significance. At the voxel level only peak height is taken into account.
You can read more about this in Chapter 4 of the SPM course notes,
downloadable at http://www.fil.ion.ucl.ac.uk/spm/course/notes.html
Best wishes,
Geraint
___________________________________________
***NOTE NEW ADDRESS FROM 28TH FEB***
Geraint Rees
Institute of Cognitive Neuroscience
University College London
Alexandra House
17 Queen Square
London WC1N 3BG
voice +44-(0)207-679-1177
fax +44-(0)207-813-2835
email [log in to unmask]
web http://www.klab.caltech.edu/~geraint
____________________________________________
|