Hi Robin,
If you run 3dinfo on the 2 stats results, you should see the
extra degree of freedom used in the 3dREMLfit results. It is
applying the exact same regression matrix, with the addition
of ANATICOR. So yes, you have results with and without the
ANATICOR term.
However the regressions are a little different. 3dREMLfit
applies an ARMA(1,1) model to the noise, while 3dDeconvolve
uses ordinary least squares. So there are 2 differences
between those results.
Perhaps in the future the default will be to not even run
the 3dDeconvolve regression in such a case (though it would
still generate the X-matrix). It would be the equivalent to
adding -regress_3dD_stop.
- rick