Hi,
This is a followup question of the previous post.
Say I have 1 EV in my first-level model which contains 10 repetitions of stimuli. Subjects were requested to give a rating regarding the level of the stimulus after each stimulus and thus 10 scores were collected. I want to know brain regions which are NOT ONLY activated to this stimulus BUT ALSO correlated with the scores. I did three analyses:
(1). In the first-level model, I selected 3-column format for EV setup and used txt files containing onset time, onset duration, and intensity (all 1s). I calculated the average of the 10 scores for each subject and then analysed another second-level regression model, which contained one EV modelling mean group activation (all 1s) and another EV modelling behavior scores (demeaned average scores for each subject)]. Two contrasts were then set up, and I did contrast masking by masking contrast 2 with contrast 1.
(2). Same as (1) but rather than doing contrast masking, I use "fslmaths thresh_zstat2 -mas thresh_zstat1" to mask contrast 2 with contrast 1.
(3). I reset the first-level model: In the intensity column of the 3-column format for EV setup, the corresponding 10 scores (but not demeaned) were used (rather than all 1s). Then I performed a second-level analysis, setting just one group mean contrast (“1”).
My Question is, are these procedures correct [should we use demeaned data in (3)?]?
And does the three approaches all give me what I want? What is the difference?
Actually the level of the stimulus varied from low to high, and within-subject variability for the rating did exist. But the average rating scores of the 10 stimuli are quite similar across all subjects, and if we used them as independent variable, just like (1) and (2), it seems difficult to identify any correlated voxel. That's why we did (3).
Thanks in advance.
Mark
|