Greetings, SPMers.
I would like feedback about the statistical and theoretical validity of an
approach I'm trying to evaluate a region's importance in performing a
task. Briefly, I'm trying to use random effects correlational analysis to
identify regions correlated with accuracy across a group of subjects, then
determine whether individuals from the group actually do show greater
activity in these regions during correct compared to incorrect trials.
The experiment used an event-related design.
The correlation effects from the random effects are weak due to factors
intrinsic to the experimental design, demonstrable only with a threshold
of p=0.05 uncorrected. There are also issues for individual analyses that
impacts sensitivity, most notably that the number of error trials is quite
variable across individuals.
I used ImCalc to create a difference image for every individual between
the brain response to correct trials ("Cor") and the brain response to all
trials ("All"). If the brain response is necessary for a correct
behavioral response, the brain response to "Cor-All" should be greater
than 0 -- but the sensitivity will be reduced since the brain responses to
correct trials is incorporated into the "All" response.
Using this approach, I find one of 4 regions identified from the random
effects analysis has greater mean activity during correct trials than
during all trials, significant at p=0.05 (uncorrected).
Two questions:
1) Since the activation identified from the group analysis is orthogonal
to the individual analyses, can the results reasonably be reported? (With
4 clusters surviving the initial random effects analysis, I think I only
need to correct for 4 comparisons if I do this as a ROI analysis, else a
small volume correction for the number of voxels in these activation
clusters.)
2) Is there a problem using the "Cor-All" difference image as an estimate
of the difference in brain response between correct and error trials,
either from a statistical or theoretical viewpoint? (There are reasons
why a direct measurement of the brain response to error trials is
impractical, which for brevity's sake I will not go into here.)
Thanks for any feedback.
Doug Burman
|