Dear fellow SPMŽers,
I am currently analyzing the data from a Stop-signal task in which
we have multiple sessions, plus some problems with regards to the
quality of the data.
Allow me to delineate the task: The subjects are instructed to
respond to a visual cue with a button press. 75% of the trials within
a session consists of these trials. The remainder of the time they are
presented with a visual cue, followed by a stop signal prompting them
to inhibit their initial response. IŽve divided these trials into 4
regressors, respectively Go, NoGo, Hit (Successful inhibition) and
Miss (Unsuccessful inhibition), where NoGo is the onset of the initial
visual cue before the stop signal. Of course, there were subjects who
had a skewed proportion of "Hit"/"Miss" trials within any given
session, so we had to run multiple sessions to ensure that they had
~50% of each. The mode of sessions used to achieve this was 3.
However, achieving this entailed running one or even two additional
sessions for some of the subjects.
I am using [-1 0 1 0] and a [-1 0 0 1] vectors replicated that I
compare at the second level through a 2- tailed T-test. Išve gathered
that replicating vectors across sessions is equivalent to calculating
the mean difference between the conditions, which obviates the need to
use different weightings based on proportions (provided the total
proportion of hit vs miss trials is reasonable, i.e. .50 +/-5), and
also corrects for the different number of Go trials used as -1.
My first question is whether or not I am right in assuming this, or
if I should use session specific weighting.
My second question is whether or not it is valid to compare data on
the 2nd level when different subjects have different numbers of
sessions, and hence differ with regard to the number of trials used in
to render the FFX contrasts.
My third question is regarding my 2nd level analysis: Ideally I
would want to construct a first-level contrast contrasting hit and
miss trials after they both have been subtracted from Go trials, i.e.
[-1 0 1 0] - [-1 0 0 1]. As I mentioned I am currently doing this
through a two-tailed T-test on the 2nd level, with Independence
specified as No. Is this the correct approach, or is there some
simpler way to achieve this on the first level?
Lastly, we have data on both normal controls and patients with
substantial brain lesions. We have manually drawn masks of the
lesions. Would it be appropriate to include these as explicit masks in
the first level analysis, so as to remove noise from the data? In
addition, we have some instances of dropout, probably as a consequence
of magnetic disturbances from surgical clips/screws (see
attached .jpg). How should I go about handling these? Would masking
them out be appropriate, or should I leave them be? And to what degree
can I trust the data from the areas surrounding these magnetic
Any help would be greatly appreciated,
Center For The Study of Human Cognition,
University of Oslo