Hi Pinson,

Keep in mind that SPMG1 is a fixed-shape model, meaning that
in using it you are ASSUMING it is the shape of the BOLD HRF
(similar to BLOCK or GAM). Any iresp curves generated from
that are fairly uninformative, because they will all look
like the SPMG1 curve, and will only vary in their magnitudes
(the magnitude is the beta weight, or Coef value).

TENT(0,16,9) can look like anything. If it ends up looking
just like SPMG1 everywhere (significant), then you can rest
assured that using SPMG1 is sufficient for modeling the data.
But of course, it will probably not look just like it all
over.

The TENT functions allow you to examine the actual shape of
the responses (assuming you have modeled things well).


Regarding the negative betas in the left Caudate, perhaps
there is some aspect of your stimuli that is not being
modeled well, and the Caudate is a part of the brain which
responds to that stimulus aspect.

Alternatively, perhaps there are too many stimuli of similar
types presented too frequently. If the Caudate responds to
multiple stimulus classes, maybe the responses cannot be well
separated by the regression.

Of course, this is assuming the curves are significant.

Importantly, note that "too many stimuli" to model with TENTs
is far less than too many to model with a fixed-shape response
curve. It is quite possible that SPMG1 is still appropriate
and models the data well, while using TENTs does not work so
well because there are too many stimuli. The TENTs will always
fit the data better, so if the fit time series for these look
similar (or maybe more simply, if the stdev(errts) are similar),
then maybe you can still rest easy using SPMG1.


Regarding how to get the activation maps, it does not really
make sense to generate the stats using SPMG1 and then to get
curves using TENT. The statistics are based on the assumption
that the responses look like SPMG1 curves.

Maybe it would be okay were you to refrain from doing anything
with those TENT curves outside of simply showing some (i.e.
not doing any tests or report any statistics using them). But
if they look like SPMG1, readers might wonder why you bothered
to show them. If they do not look like SPMG1, readers might
think you should not have used SPMG1 as the basis of your
regression model.

I like the notion of using TENTs for personal use (not to be
reported), to verify that the applied fixed-shape HRV was
appropriate.

- rick