 |
 |
Dear FSL Gurus, We conducted a moderation analysis (multiplied EV1 X EV2 = interaction term) to identify brain moderators between depression and pain using perfusion fMRI. I read the paper titled "Orthogonalizing Regressors in fMRI Models” and a couple of discussions here but still have a few questions. Subjects were scanned during noxious thermal stimulation, and we collected pain intensity and pain unpleasantness ratings afterwards. Importantly, pain intensity and unpleasantness ratings were highly correlated (naturally); r = .94. Thus, we believe that orthogonalization may be justified. In our higher level FEAT analysis, EV1 was entered as the main effect of pain (all “1’s"), EV2 = demeaned depression ratings, EV3=demeaned pain intensity, EV4 = demeaned pain unpleasantness, EV5 = interaction term between depression and pain intensity, and EV6 = the interaction term between depression and pain unpleasantness. We orthogonalized EV6 with EV5 in order to identify the unique brain moderators between depression and pain unpleasantness independent/above and beyond the depression and pain intensity brain moderators. Since pain intensity and unpleasantness are highly correlated (r=.94), we worried that not orthogonalizing would sacrifice significant shared variability between EV5 and EV6 that we wanted to capture. We are curious if our method of orthogonalizing is valid and if other regressors should be orthogonalized to each other as well (i.e, pain scores; depression scores) since they are sharing the variability with the interaction terms. NOTE: We understand that orthogonalizing is generally not recommended. For example, Dr. Jensen stated previously in this forum that, “The only situation I have come across where it makes some sense to orthogonalise is when you have a higher level design that includes lots of highly correlated EVs, but where you can identify some form of causal relationship between them.” Having said that, when we run the intensity and unpleasantness moderation analyses separately, we see significant overlap between the brain moderators (i.e., intensity and unpleasantness). However, when we entered intensity and unpleasantness in the same moderation analysis, the data are very interesting and quite comprehensible. We are very interested in determining brain moderators supporting psychological disposition and pain intensity as well as the unique contribution of pain unpleasantness. Any advice is greatly appreciated-- thank you so much! Adrienne