Hi,

I’m running a second-level seed-to-voxel analysis and I’m trying to test whether the change in rsFC from scan time 1 (pre) to scan time 2 (post) is correlated with behavioral improvement over the same period. here is the setup:

• subject effect: group division, covariates, behavioral variables at each timepoint (behav_T1 and behav_T2)


• conditions: time1 (pre), time2 (post)

approach 1 :
contrast for condition is [-1 1] and contrast for subject effet is [0 0 -1 1]. This gives significant clusters.

approach 2 (manual delta):
I compute dBehav = behav_T2 − behav_T1 manually and test the dBehav regressor with a contrast like [0 0 1] (for group/covariates/dBehav), again with condition [-1 1]. This gives different (kind of weaker or non-significant) results.

I expected these two approaches to be same tests of whether ΔFC (post–pre) is associated with Δbehavior (T2–T1), but the results differ. I personally suspect it may be related to differences in the second-level GLM (like the implicit constraints when using a single delta regressor instead of using T1/T2 separately), but I’m not sure which one is correct.

Could you advise which approach is recommended (or correct) in CONN to test (post–pre) FC change correlated with behavioral improvement or should I find another way to do this analysis?

Thanks!