Hi Jen,

How many TRs do you have per run, and now many runs are there?

Each run is probably long (since you use -polort 4), but still,
you have 22 stimulus regressors with 33 lags each, besides your
motion parameters. That is asking a lot from your data. With
4 times as many regressors as before (the jittering), you lose
a lot of statistical power.

Note that you should not need to use -stim_nptr on your motion
parameters. Those are going to exist on the TR grid, and should
not need to be broken up.

It seems to me that you would be better off applying the actual
stimulus times using -stim_times, as opposed to breaking up the
TRs with -stim_nptr. If you apply the stimulus times, then you
could use TENT(0,secs,9) functions for those options (where secs
would be the number of seconds in 8 TRs).

I am guessing that you would use min/max lags of 0/8 if you were
not using '-stim_nptr 4', so instead of 9 regressors on the TR
from lags, you can get those 9 regressors from the coefficients
of the TENT functions (again, on the TR). This should come out
to basically the same thing, except that your stimuli need not
be TR-locked.

- rick