Hi Aaron,
There are certainly instances where that would seem reasonable. It would be strange to include 3 or 4 time points early in a run when the next 25 are censored, for example. But coming up with a concrete and reasonable method for exactly when to do such a thing might be difficult, especially since it would depend so heavily on the experiment design. Still, doing such a thing seems justifiable.
We suggest performing the entire linear regression (including just projection) in a single model, which includes band passing and censoring, as well as any other desirable or undesirable components in the model. That is how afni_proc.py does it, so you can use that to see an example if you have not already done so.
Right, it would not be reasonable to band pass kill-mode censored data (unless the band pass terms were similarly truncated - again that is most straightforward to do in one model).
- rick