Hi Neurostars!
I’m trying to perform design optimization on our fMRI food cue valuation task (a Becker-DeGroot-Marschak auction paradigm).
Current timing:
4s food cue visualization → up to 4s bid period (right hand moves a cursor along a visual analogue scale from $0–5, left hand confirms the bid) → 1s bid confirmation window → 3–4s jittered fixation cross.
This repeats for 45 food cues per run: 15 high-calorie salty, 15 high-calorie sweet, 15 low-calorie.
We assume valuation takes place during the visualization period, so that’s our event of interest. The bid period is entered in the GLM as a nuisance regressor. The confirmation window and fixation cross aren’t modeled as separate regressors, they’re not of interest, and modeling them separately isn’t feasible given collinearity with the other events (currently)
Our contrast of interest is usually visualization of high- vs. low-calorie cues: [+1/2 highcal-salty, +1/2 highcal-sweet, −1 lowcal].
Problems I’m running into:
-
Collinearity between the visualization and bid regressors. My suspicion is that this is because the bid period always starts the instant the food cue disappears, a fixed 4s lag with zero jitter between the two events on every trial, so the two regressors end up almost perfectly time-shifted copies of each other after HRF convolution.
-
Need to make sure the same food category isn’t repeated too many times in a row.
What I’m hoping for:
General design recommendations for addressing the above, and suggestions for good software to implement/optimize the timing.
We also think the motor/somatosensory response from the bidding movement may be collinear with the visualization-period estimates, given the fixed lag described above, we’d appreciate any thoughts on how to improve this, e.g. adding a jittered interval between the visualization and bid periods.
We’re finalizing a new protocol, so this is the right time to get the design right before data collection starts.
Thanks in advance!