Dear Sander,
> I am currently doing an event-related fMRI study using a memory task. I
> would like to do an RFX analysis on hits - baseline. However, the
> problem is that the individual subjects have considerably different hit
> scores, so that different numbers of events are taking into the first
> level analysis, giving different degrees of freedom (ranging from 90 -
> 130).
> I understand that this can be problematic in that RFX analysis assumes
> equal degrees of freedom. Is there any way to correct for this?
In fact the second-level analysis, used in SPM, assumes that the design
matrices are identical for each subject. To do your analysis properly
you would have to fit a hierachical model with different variance
estimators for each subject. However, remember it is the variance of
the contrast that is the critical variance here and the d.f. for each
contrast will be the same (assuming the same scan number for each
subject). The only thing you have to assume is that the variable
number of 'hits' in each subject does not induce substantial
differences in the variance of the contrast estimators. I think that
as long as you qualify the results, and make it clear what you have
assumed, you should be OK.
We are currently working on this issue but it is too early to quantify
how robust the two-stage approach to RFX analyses is at this stage.
I hope this helps - Karl
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|