On Thu, Feb 14, 2013 at 9:00 AM, P Andersson <[log in to unmask]> wrote:
> Thanks Donald!
> Really helpful, especially regarding the difference to SPM's output.
> I will take the opportunity to ask a couple of other questions on whether or not what I plan to do makes sense.
> I would be very grateful if someone could comment on this!
> I have results from an ICA analysis of resting-state data containing a control-group and a test-group.
> (I have used GIFT, with GICA to back-reconstruct the individuals' components)
> I now applied a FullFactorial analysis on the components, with a factor stating if a subject is from control or test group
> (the model also included some covariates for age etc).
I'm a little confused here. If you enter more than 1 ICA component in
the model, then you should not be using full factorial as your model;
rather you should be using the flexible factorial model. However, as
your questions is probably does component X differ by group, the
flexible factorial would not be appropriate. The three solutions are:
(1) GLM Flex; (2) Full factorial with only 1 component; or (3)
Two-sample t-test with only 1 component.
> Now to my question:
> Would it be ok to first (for each component) create a mask based on a corrected cluster size threshold on the main-effect, and then use this as an explicit mask in "control VS test" contrasts for the same component?
> If so, what would be the best way to do this? Do I have to rerun an analysis with the mask included as 'explicit mask'?
You should not mask your data with this contrast or the results from
the one-sample t-test. The reason is that you are saying the average
of both groups must be significant. You have the potential to miss
detection regions that are positive in one group and negative in the
other group OR only significant in one group and subthreshold in the
other group.
>
> Cheers,
> Patrik
>
|