Hi Irene,
That is not too much variance in stimulus duration time,
so using TENTs would not be so bad, and that is only for
the 'co' and 'cs' classes, the others do not even vary.
The potential trouble with expecting TENT to be consistent
with dmBLOCK is that with dmBLOCK, one is usually meaning
that the stimulus duration will vary, and therefore the
*shape* of the response will vary. But when using TENT
to model the shape, it is not expected to vary.
In this case, with the only variance being between 2, 3
and 4 s, that is probably consistent enough to get
something reasonable with TENT, too.
Getting to your initial question then, the inconsistency
is probably due to having many events that overlap too
much. So the TENTs might be at least somewhat overmodeling
the data. Not only should events be more spread out, but
they should not tend to overlap in any predictable way.
How many time points are there? How many regressors in
the TENT case? How "random" is the timing and ISI?
- rick