> One rationale backing the current approach might be that we first need to regress out the (possibly existing) signals
> imposed by the driving input in order to generate data composed of pure noise.

I'm not so familiar with the implementation, but I suspect that is likely the rationale. I wonder how to flip the sign with the observed data (rather than the residuals) when one or more covariates (e.g., age) are involved. Is your simulation process similar to what's described in Dr. Cox's following paper?

[www.ncbi.nlm.nih.gov]

My guess is that the approach adopted in 3dttest++ is intended to address the issue of spatial neighborhood relatedness, which may cause some deviations in the case of a single voxel as you found out.

Gang



Edited 1 time(s). Last edit at 01/26/2023 08:34AM by Gang.