Thanks for the link.

One point is still not clear to me.

I see that the data time series (3d vol ts) is scaled before 3dDeconvolve and 3dTcat in the example program.
but it doesn't look to be detrended before concatenating across runs from different sessions.

For context-dependent correlation analysis (http://afni.nimh.nih.gov/sscc/gangc/CD-CorrAna.html),
The input data time series (3d volume ts) and seed time series (ROI 1d ts used to construct interaction regressor) seem to be differently processed.
input data time series- scaled and centered at 100 but not detrended
seed time series- detrended time series but not scaled and centered at 100 (before deconvolving and constructing regressor).

Am I missing that each run is detrended and scaled to be centered at 100? Then, the ROI ts would be centered at 100 and input 3d ts would be detrended.

Having different trends and offsets in seed and target time series make any difference to the regression analysis?

Thanks for your help.