Hi
>
> In one of my studies (which I designed to be analyzed with the GLM), I counter-balanced event order. It was a 2 x 3 design, in which the 2 factor was "task". So, I created 2 versions of the experiment, in which the "task" part was different in each version. As a result, the order of event types is different for half the participants.
>
> My initial impression is that this makes it impossible to do group ICA, as the time course of the components would be in conflict. Is there a way around this? Does event related ICA always require that every participant get the exact same experimental design in terms of order and timing of events?
No, identical event timing is (currently) important for 'Tensor-ICA' but not for 'Concat-ICA'. The latter will be OK given your data.
>
> Also, for each participant we have 4 runs of data. Does this pose an issue for group ICA? i am not clear (at this very early stage of learning) how this would be incorporated into the group ICA approach.
You can add all data into concat-ICA to get group average results. The fact that there is 4 runs per subject would then enter in any between-group design, e.g. using dual regression to determine group differences on the basis of group means.
>
> Thanks a lot for any advice. I am very excited by the potential power of ICA analysis, but I am not sure it is compatible with my current data. And I want to make sure I set up future studies to enable ICA analysis.
hth
Christian
|