I'm new to 3dDeconvolve so apologies in advance.
It seems that when you use -nfirst>0 with -legendre then the integrals of the 1st and higher order polynomial columns of the xmatrix are not zero. This was somewhat unexpected. As a result, estimating the voxel average response is cumbersome (i.e. I think I need to integrate the columns of the xmatrix and use all fit coefficients?) On the otherhand when nfirst=0 the columns integrate close to zero.
So it seem there are two options:
(1) Chop the data before 3dDeconvolve and use -nfirst 0
(2) Customize the xmatrix and reconvolve by removing the mean from the columns
Advice? Thank you
--judd