Hi Lydia,
The TENTzero() basis functions are identical to those of TENT,
except that the first and last basis function are not included (so
the endpoints are *assumed* to be zero, and are not estimated).
So TENTzero(0,14,8) is identical to TENT(2,12,6). I suggest you
stick with TENT(2,12,6) because it is more clear what you are
doing. It is not terribly appropriate to plot the zero endpoints in the
case of TENTzero, because they were not estimated as such.
The only magic here is that you are using fewer basis functions,
making it more likely that the system of equations has a single
least squares solution.
You got into trouble when going from GAM to TENT(0,14,8) because
that multiplied the number of basis functions by 8. But it became an
even bigger issue because the stimuli are TR-locked (so regressors
are 0/1 binary, rather than fractions). When switching to TENT(2,12,6)
or even TENT(2,10,5), as suggested by Gang, the number of basis
functions was then reduced by 25 to 37.5%, and since it was apparently
just the endpoint regressors which were identical, the model worked (or
was at least solvable). It might have still failed.
I expect that the results are noisy, even though the analysis completed,
because there are so many regressors.
- rick