Hello Sara:

I am concerned about the 8 minute length of the "on" periods for the different
experimental conditions. In fMRI experiments, 8 minutes is a very long time.
A lot can happen in 8 minutes. There can be "scanner drift". The subject
can move slightly inside the scanner. The subject might get bored. The earth's
magnetic field might reverse polarity.

All fMRI experimental data is modeled relative to the baseline. This makes
it important to have frequent updates of the baseline estimate. If the
"on" period is lengthy, then any change in the baseline during this "on" period
can be confounded with the experimental effect that is being measured.

A related problem is that, due to the length of the "on" period, there can be
only one "on" period during the run. So, in effect, there is only one trial,
even though there are many fMRI measurements during the run.

Have you used the Deconvolution plugin to plot the fit on top of the fMRI data?
If you are using the estimated motion parameters as baseline regressors, I'm
surprised that there is any statistical significance left over for modeling
the input stimulus function.

Response to other questions:
The usual procedure is to apply the Tlrc transform first, followed by Gaussian
blurring. In this case, the "-dxyz=1" is unnecessary.

Doug Ward