Hi,
> The 3D brain maps
> I've been using only consider a total of 2394 voxels over the whole
> cortex,
...
> So, on this basis, is it reasonable to use p<0.01
> rather than p<0.001 to 'suggest' an important area of activation? And can
> FWE be used at p<0.5? Or does this not follow?
The FWE in SPM adapts to something close to the number of comparisons
in your image. If you ask for an FWE of 0.5, you would expect a false
positive in every second repeat of your experiment, assuming all the
data is noise. So, dropping the FWE criterion will drop your
specificity and made it harder for you to argue on statistical grounds
this is a real positive.
For the uncorrected criterion - as the criterion is arbitrary, in
terms of FWE, we your readers will have no good idea of how seriously
we should take your activation on statistical grounds. That was why I
was arguing that we should not be using uncorrected thresholds If you
do use a threshold like that, the argument has to be made in other
terms, such as the plausibility of activation at a particular point or
the pattern of detected activation at that threshold.
> My other question is to do with the use of 'extent thresholds' - if I tell
> SPM to only consider, say, five adjacent voxels above my P<0.001 (or
> p<0.01 or whatever) threshold, to what extent does that increase the
> reliability of the results? Is there a simple way to express how this
> effects the reliability mathematically?
It's possible to calculate that using the SPM machinery - have a look
at spm_P.m in the SPM distribution,
Best,
Matthew
|