Dear Eric
It sounds like the basic problem is that your PPI regressor captures very little extra variance in addition to the ROI timeseries and our psychological regressor, because the events (Y) that you are modelling represent a very small proportion of the total time.
Consequently your PPI regressor will look a lot like a your ROI timeseries (except multiplied by -1).
I think it would be valid for you to de-min, instead of mean-centre your psychological regressor before generating the PPI regressor and you are welcome to try. However, theory suggests it ought not to make a difference to the fit of the PPI regressor, only to the fit of the ROI timecourse (which will now have a better fit because it is not corelated with your PPI any more). This is equivalent to orthogonalising your PPI regressor with respect to your ROI timeseries.
Jill
-----
Dear FSL experts,
I am trying to conduct a PPI analysis on a task that includes trials in which participants press a button to a stimulus (X) and trials in which the button press is to be inhibited (Y). The Y trials occur on ~8% of all trials, and are either modeled according to whether they are correct or incorrect. The X trials are presented on the remaining 92% of the trials and are presented with an ISI of 1 second. In my original GLM model, the X trials are not explicitly modeled, but rather serve as an implicit baseline. For the PPI analysis, I am interested in determining those voxels that show increased connectivity with a seed region during correct inhibitions compared to the implicit X baseline.
I set up the PPI analysis according the instructions at http://www.fmrib.ox.ac.uk/Members/joreilly/how-to-run-a-ppi-analysis-in-feat, but had a question about the interaction term. I have tried setting my PSY regressor (Y correct trials) zero point to Centre, but the group results seem odd in that there are only negative correlations, counter to my predictions (and the original GLM comparison of Y correct to the implicit baseline). Would it make sense in this case to set the zero to the min of the regressor instead of the center, given that the conditions that are being compared in the PSY regressor are Y correct and the implicit baseline? Is this approach invalid?
Thanks in advance.
Eric
|