Hi,
I’m using FSL to analyze an event-related, within-subject fMRI task with three conditions (A, B, and C). On each trial, participants saw an image and gave a subjective rating. Images from all three conditions were randomly intermixed.
At the first level, I built a GLM that includes:
- Three parametric regressors: subjective rating in condition A, B, and C
- Three task-related regressors modeling general activation for each condition
Each regressor is defined using a standard 3-column EV file. For the parametric regressors, the third column contains demeaned subjective ratings. For the task-related regressors, the third column is filled with 1s (i.e., modeling onset events without modulation). The first and second columns specify stimulus onset times and trial durations, respectively.
I then set up a contrast [1 1 1] across the three subjective rating regressors to identify brain regions whose activity is associated with ratings across conditions. My understanding is that this contrast tests whether the combined contribution of the three condition-specific rating effects is significantly associated with BOLD activity.
My question is:
How is this approach (contrast [1 1 1] over three condition-specific subjective rating regressors) different — statistically and interpretationally — from building a separate GLM that uses a single parametric regressor representing subjective ratings across all trials, regardless of condition?
I’m especially interested in:
- Whether the [1 1 1] contrast can be interpreted similarly to the across-condition regressor
- Whether there are differences in statistical efficiency or assumptions
- In which cases one approach is preferred over the other
Any insight would be much appreciated!
Shuning