Hi,
Yes, that's right - confirmed. The GM fit wrt the time domain is to be
understood just like in the case of stadard GLm analysis on time
courses. Though keep in mind that because the sign in an ICA
decomposition is not uniquely identifiable (if you flip the spatial
maps and either the time course or the subject mode vector you'll get
the same outer product) you shuld (along with any positive contrast)
also look at the negaive contrast (i.e. look for significant de-
activations).
hth
Christian
On 4 Mar 2008, at 18:59, John Herrington wrote:
> Hi all,
>
> I'm looking for confirmation on how to interpret significant
> contrasts (COPEs) of ICA components from Melodic. I ran tensorial
> ICA on 18 participants and isolated a component that appeared
> positive (red) on output. I also got output for significant task-
> related effects (COPEs from single-run Feat jobs) as part of the
> analysis. This parameter reflects a contrasts of high-arousal and
> low-arousal conditions in my experiment (in the Feat GUI, high
> arousal was coded as +1 and low arousal was coded as -1). The
> output on the web page for this contrast was as follows:
>
> COPE(3): z = 10.65, p < 0.00000
>
> Given that the z-value is positive, is it correct to say that this
> particular ICA component was stronger (more significant) for the
> high arousal (coded in Feat as +1) than the low-arousal (coded in
> Feat as -1) conditions?
>
> Thanks - John
|