Duong,
My opinion is that FMRI data are not suited for causal modeling through GCA and SEM considering the amount of noise and coarse temporal resolution.
> It seems to me we can only do this with resting state data. Is this right?
At least there is one less concern with the resting-state data than task-related data about the mixture of task and resting state.
> when cutting out a 'task' block, should we keep the whole block or should we remove the first 3-4 TRs of hemodynamic lag?
Instead it probably makes sense to include a few more TRs after the stimulus ends due to hemodynamic recovery phase.
Gang