Dear Haris,
There is no principled difference between fMRI and MEG. If you are
using the statistical test to test the hypothesis about presence of
an effect you should use corrected p-values. You can easily convince
yourself of that by testing for something completely random and seeing
that with uncorrected p-values you'll always get some blobs.
The only reason to use uncorrected p-values would be if you've tested
for your hypothesis on some other way (for instance at the sensor
level) and now only want to localize the effect. Then applying the
uncorrected threshold would be one way to find the more consistent
features across repetitions. But there is nothing principled about
that particular threshold and you can also threshold the t-values any
way you want.
The only valid way to increase the power of your inference is to do
small volume correction i.e. not to test the areas in the brain where
you don't expect any effects a-priori.
Best,
Vladimir
On Thu, Jan 28, 2010 at 12:28 PM, Charalampos Styliadis
<[log in to unmask]> wrote:
> Hallo i have done an MEG SAM group analysis and used SnPM in order to get the
> statistic significance. I have read some replies about the issue regarding
> corrected and uncorrected p values and i would like an updated view about this
> issue as far as MEG is concerned. I am asking because most replies were about
> FMRI. I would like to know under which circumstances I can use the uncorrected
> results. Although most suggestions for FMRI are preventing the use of
> uncorrected I have come across some recent papers on MEG that use these
> values.
>
>
> Thanks
>
> Haris
>
|