Hi Anat,

> My question refers to event-related design with a trial duration of 4
> seconds, and a TR of 2 secs.
>
> Given that the onsets files include only the first TR of each trial, the
> second TR is never listed, like all the fixation trials. It implies that
> the following two hypothetical trial sequences:
> a. trialA (2secs) -> fixation (2secs) -> trialB
> b. trialA (4secs) -> trialB
>
> are treated by the model in the same way.

This depends on how you have set up the regressors. I'm assuming that in
each case you have specified all the trialA onsets as one regressor, and all
the trialB onsets as another regressor, using an event-related design with
the hrf basis set. The regressors for A in the first situation (short trial)
will then be the same as the regressors for A in the second situation (long
trial). This is because I have assumed (presumably incorrectly) that the
neural activity evoked by A is well modelled by a delta function centred on
trialA onset, and so the expected hemodynamic response will simply be these
onsets (which are identical in each situation) convolved with the hrf basis.

> treating them as identical would mean that, once a hemodynamic response
> begins, it has one possible pre-determined course.  Regardless of how long
> the actual trial lasts before the next trial/rest, the model would treat
> trial B identically.

You mention comparing trialAs of different duration below, so I'm not sure
if this is a misprint and you meant to say 'treat trial A identically'. But
the model I've assumed will produce the same model for trialA and trialB in
both situations anyway, as the onsets are identical in each situation.

Note that this model is not a particularly good one in the first place, even
for the short (2s) trialA duration alone. The expected neural activity
evoked by trialA presumably lasts throughout the 2s trial, and so is not
well represented by a delta function at the onset of the trial convolved
with an hrf basis. As trial length becomes longer and longer this will
become more and more of a problem (imagine the silliness of modelling a 30s
trial with a single 'event' at the very beginning!).

> Is there a way to differentiate these two scenarios in SPM?

Yes. You need to ensure that the regressor for trialA reflects the length of
that trial. This can be done either by specifying trialA as a 'mini-boxcar'
with length reflecting trial length, or as a train of 'events' suitably
close together (e.g. 100ms) for the duration of the trial. In each case the
goal is the same; to create a regressor that approximates the shape &
duration of the expected activity (e.g. a 2s or 4s boxcar, convolved with
the hrf. Of course you may have a more specific hypothesis for the shape of
the expected neural activity as a function of time during trialA, and this
would alter the basis set you use accordingly).

> To the extent that I have a reason to believe that a real cognitive effort
> does continue during the whole duration of the trial (i.e. that a 4secs
> trial does behave differently from a 2secs trial) - am I correct to assume
> that I would have to incorporate this belief to the model via one of the
> user-specified options?

Yes, and you should be able to do this with the options available in the
standard interface.

Best wishes,

Geraint
http://www.klab.caltech.edu/~geraint