Hi Jessica,

Those TR counts are for the BOLD responses - for how
many time points is each non-zero. Since rest comes
after task in each run, the computation is more simple
for the task terms (no BOLD response goes off the end
of a run).

The 60s Base task, when modeled with BLOCK(60,1) say,
has a 75s response curve, which probably comes out to
37 non-zero time points. Having 3 such blocks means
111 non-zero time points in total. The 2 30s tasks
have ~45s ideal BOLD responses, which are probably
non-zero for 22 time points, and 3 of those is 66.

Rest is the anomaly, as it goes to the end of a run.
The first 2 in a run do not go off, giving 74=37+37
time points. And then another 29 non-zero time
points at the end of the run gives 103.

However, you should probably not be modeling rest
at all. This is very close to a multi-collinear
system, and I suspect 3dDeconvolve is showing some
warnings. Is it?

Also, since these blocks are so long, I expect
the baseline model might be a confounding effect.

- rick