On Fri, 18 Dec 2009 18:57:10 +0000, Ananth Narayanan
<[log in to unmask]> wrote:
>Secondly, if I wanted to show that a particular area was more activated in
>one group than another, how would I go about doing that? I assume looking
at
>the number of significant voxels, k sub E, which is 49 in this case, do it
>for each subject in every group and run statistics on it? How else would I
>do it? Thanks.
>- Ananth
I don't think that that's a good way of going about it. While I don't have a
mathematical proof in front of me, doing group-wise stats on subject-level
cluster size is too indirect. For example, it would be very dependent on
choice of threshold at the subject level; and I assume there are issues with
normality.
The conventional ways of doing it are:
(a) just take the contrasts of interest (con*.img images) "to the second (i.e.,
group) level." In the simplest case, you'd be doing an unpaired t-test
between the groups. You'd then look to see if there was activation in that
same region.
(b) do an ROI analysis, using MarsBar or some other tool. There are variants
of this, but basically at the subject level you look at things _after_ you
average the signal over the region.
|