Hi Nic,

As a side note, I do not recommend using TENTzero, it can
be confusing to those who do not use it, which is probably
99.8% of FMRI researchers. You are already cutting out
the first 12.5 seconds, and that just adds 4.28 (= 30/7)
seconds in a subtle way.

TENTzero(12.5,42.5,8) means approximately TENT(16.8,38.2,6),
and the latter is much more clear.

Actually, I am not sure that you want to use the '8' in the
first place, since that divides up those 30 seconds into 7
pieces, which is an odd division. Since your TR is 2.5s,
I might guess that you would shoot for 5s intervals, i.e.
TENT(12.5,42.5,7) or maybe TENT(15,40,6).

Are your stimuli TR-locked?


1) Using 30% certainly seems reasonable, and is a common
limit to use, at least in the small community that I am
familiar with.

2) Having 30 TRs is often less possible with TENTs, since
in many cases it equates to the number of stimulus events.
For you, I guess it is approximately twice the number of
events (since you seem to be sampling on a 6 se And
you do not seem to be using the TENT data for the main
finding, but rather for a pre-processing scaling step.

3) In your case, 20 does not seem too bad as a lower limit.
A smaller number suggests "noisier" TENT functions, since
fewer TRs of data are used for the estimation.

Also, using the TENTs to help scale the data suggests using
a combination of the betas for the scale computation. And
that would essentially increase the number of samples used
for the scaling operator.

A peek at the paper shows use of Bave for scaling, which
is the average of all of those TENTs. That value would
be computed across something like half the sum of TENT TRs,
plus an endpoint fraction of the TRs (tents overlap), I
guess, i.e. maybe 62 plus another 16%, which is (not so)
suspiciously almost exactly the 71 you state for BLOCK.

- rick