I think that your model and interpretation are correct (although whenever there's a nice check box that does everything automatically for you, it's hard to be 100% certain). However, might I suggest a potentially-improved model? EV1 would stay the same (heat stim: 1's = on, 0's = off). Perhaps I'd break down EV2 into two parts: pain ratings during pain stim (newEV2), and pain ratings ratings during no stim (newEV3). The advantage to this method is that you can still estimate the same correlation as before by making a contrast [0, 0.5, 0.5] that averages across newEV2 and newEV3. However, you can also test the correlation during each period separately [0, 1, 0] and [0, 0, 1], and the difference between correlations for those two periods [0, 1, -1].
You would still want to make sure that newEV2 and newEV3 are orthogonal wrt EV1, but it's easier to do it manually be sure to get it right. For each subject separately, you'd want to mean-deviate (aka demean, aka zero-mean) both newEV2 and newEV3 by subtracting the mean from each value. The TRs that correspond to the other timeperiods (i.e., pain off period for newEV2, pain on period for newEV3) should be assigned 0's.
You can check that you've done it right in two ways. First, the sum/mean of all values in newEV2 and newEV3 should be 0. Second, when you hit the "Efficiency" button in the Feat GUI, the covariance matrix for the relationship with EV1 should be black for all of the off-diagonals.
Good luck!
--Greg
____________________________________________________________________
Greg Burgess, Ph.D.
Staff Scientist, Human Connectome Project
Washington University School of Medicine
Department of Anatomy and Neurobiology
Email: [log in to unmask]
|