Dear SPMers,
I am currently analyzing data from an event-related fMRI study with a 2 x 2
x 2 experimental design (thus, 8 conditions overall) using SPM5. I used the
“factorial design” facility for 2nd-level design specification, and
everything worked perfectly. Now, my problem is that the analysis of
behavioural data during scanning showed differences among conditions with
regard to performance (percentage of correct response) and reaction times.
Thus, I am wondering whether I can just don’t care, or whether I should
rather introduce these effects as nuisance variables, in order to discount
their possible effect on differences in activation among conditions. In this
case, however, I can not figure out how to do that, and in which stage of
the analysis.
I tried to do this at the single-subject level, as this would allow me to
consider values of performance and reaction times for every scanning session
of a given subject. However, I do not see how this can be done. In
particular, I can not use the multiple regressors option, as this would
require a value for each scan, which I do not have (I just have a summary-
value for each condition within each session).
As an alternative, I could introduce these nuisance variables at the 2nd
level, though using this approach only group-, and not inter-subject-,
behavioural differences among conditions will be considered. In this case, I
suppose that I should use the “covariates” option in the “specify 2nd level”
menu. That is, I should introduce a vector with 8 values, each
corresponding to one of the 8 cells which specify conditions in the
factorial design. Do you think this is correct?
At this stage, I can also specify an interaction between the covariate and a
chosen experimental factor: however, I have no reason to hypothesize that a
given covariate (say, percentage of correct responses) should interact with
a specific factor. Therefore, I am inclined to specify “none” here. For the
same reason, I am inclined to specify “overall mean” as a centering option.
Is this correct?
One last question: do you think it is fine to introduce two different
nuisance variables (performance and reaction times) or it would be better to
choose just one?
Any help would be greatly appreciated!
Best wishes,
nic
************************************************
Nicola Canessa, Ph.D.
International School for Advanced Studies, SISSA
Via Stock 2/2, Trieste
Italy
************************************************
|