At least here, when we talk about jittering, we mean by amounts
of time that are specifically not TR locked. So your case does
not seem to be jittering, at least by how we use the word.

In a fast, event-related experiment, there might be a set of
stimuli that last 300 ms, and are presented randomly on the TR
grid. The 300 ms is often considered instantaneous, since the
temporal resolution of FMRI is not very good. The remainder of
the TR would be filled with fixation, or whatever baseline task
you are using.

So if the TR is 2 seconds, even having 2 stimuli in a row would
still have them considered 2 seconds apart. This might be the
source of why you picture a minimal ISI of 2 seconds.

---

Your case seems a little different though, since you say each
stimulus occupies the entire TR. So consecutive TRs of stimulus
would not look like a sequence of events, it would look like a
block.

But you should still be able to treat them as a sequence of
events or rest periods. If you don't happen to have any real
rest between two events, that's okay.

How do you plan to analyze the data?

If you use 3dDeconvolve with lags, then you should be able to
treat your events as being instantaneous and apply the above
method, since it doesn't really matter how long they last,
provided you allow sufficient lags for the response to play out.

If you don't use lags, and do some convolution before 3dDeconvolve
(such as with waver, as is done in HowTo #3), then again you are
okay, as long as you can come up with a resonable convolution
function that deals with the 2 second stimuli.

- rick