Most people like to have TR-locked regressors, since that is
the timing grid that they would get used to for the study.
However in your case there are 2 reasons why it may not be so
useful. One is that your TR is short (1.2 seconds), requiring
twice as any regressors as usual over the same time span. The
other is that it seems your stimuli are independent of the TR
timing. You can choose anything that makes sense.
I suggested 2.4 seconds so that the time grid will mean something
compared to the original one. But you could just as well have
one sample every 2 seconds, say.
Basically it doesn't matter much, except that people get used to
a certain time grid. So if the grid of the regressed signal has
nothing to do with the grid of the input EPI signal, it can be
confusing. Using 2.4 seconds might be a reasonable compromise.
Sticking with 1.2 seconds might be great if you could handle the
degrees of freedom. But with such a small jitter range, I suspect
you will have a lot of overlap in the regressors and get warnings
of multi-collinearity. It may be necessary to use fewer tents.
----
As a side note, timing_tool.py can be used to further evaluate the
ISIs across runs (averages, min, max, etc). See example 4 from
'timing_tool.py -help'.
- rick