Hi,
Well, almost. The input to fslmeants (stage2_ic0000 in this case) is #subjects by #voxels. each column describes how each individual subject's time series relates to the original network specific time course.
The eigenvariate is a summary statistic (across all voxels) that explains the most variance in this collection of association vectors. It's different from the mean, in that the latter describes the average association only. As an example, imagine a 2 voxel brain and 4 subjects. Now imagine that stage2_ic0000 looks like
+1 -1
+1 -1
-1 +1
-1 +1
That is, in different parts of the brain the associations are inverted. AN example would be a 2 group study where 2 groups happen to implement 2 different mechanism associated with the same function. In one group a certain part might get upregulated, in another group of subjects a different part gets upregulated...
The mean would simply be (0 0 0 0)', fitting the mean into the matrix, removing it and then calculating how much this has removed variability in the data (i,e. checkin how much the mean vector explains in terms of variance) will tell you that no variance has been removed. (In the F-stats/ANOVA sense) the mean is not a good description of what is going on in that data. The eigenvariate is (+1 +1 -1 -1)' (or it's inverse (-1 -1 +1 +1)', that's important later). When fitting this into the matrix you get a perfect fit, i.e. info in that matrix is accounted for (as all variance is accounted for).
If there is an associaion with reaction time then you'll need to post-hoc figure out the directionality of that association.
hth
Christian
On 25 Jan 2013, at 08:08, Christiane Jockwitz <[log in to unmask]> wrote:
> Dear FSL Experts,
>
> I did a dual regression analysis and generated eigenvariates using
> fslmeants –i dr_stage2_ic0000 –eig
>
> I already searched for the exact meaning of eigenvariates, but I would like to make sure that I understood it in the correct way.
>
>
> As I understood the meaning of these eigenvariates, they show me the variance of the component that is explained by every single subject. So, in other words, the eigenvariates tell me how the subject contribute to that specific component. Is that right?
>
>
> And secondly I would like to correlate these eigenvariates to another variable, let’s say, for example, reaction time. My results then would show a positive correlation between reaction time and the eigenvariates. Would that then mean that subjects that explain more variance of that specific component also show a higher reaction time?
>
> Thank you in advance.
>
> Greetings
> Christiane Jockwitz
|