I have the following questions about the way drifts are removed in 3dDeconvolve with the -polort option.

(1) Is the process of removing drifts done for each continuous run separately? (I'm asking because the data for all runs is concatenated, but with the -concat option 3dDeconvolve seems to know of the runs.)

(2) Is the drift removal done by adding the polynomials to the regression? Or is the data first filtered using the polynomials and only then entered into the regression (without the polynomials)?

From
[afni.nimh.nih.gov]

"For each block of contiguous data, the time range from first to last is scaled to the interval [-1,1]. The standard Legendre polynomials Pn(x) are then entered as baseline regressors, for n=0,1,..."

That says to me that (1) drifts are removed from each run separately, and (2) the polynomials are entered into the "big" regression as regressors of no interest (meaning that filtering ahead of regression is _not_ performed).