Parametric Modulation - Questions about variability

AFNI version info (afni -ver):
Precompiled binary linux_openmp_64: Jan 29 2024 (Version AFNI_24.0.03 'Caracalla')

Hello! We have run an experiment where individuals behave aggressively (on a scale of 2 (no aggression) to 6 (highest aggression) to mean and nice peers (24 trials each, we'll call mean and nice "condition").

We have used this aggressive behavior as a parametric modulator in individual level analyses. We do this by having a timing file with onset of each trial (regardless of condition, so 48 trials) * aggression value for that trial. Then, in group level analyses, we are looking at mean condition x PM - nice condition x PM as our contrast of interest.

BUT what we noticed was that for some participants, they ALWAYS responded with no aggression (a value of 2) on nice condition trials. Therefore, although there is variability in our parametric modulator across all 48 trials, there would be no variability in the parametric modulator within that one condition.

That brings us to two questions:

  1. Is the individual level activation for a participant with no variability in the PM equivalent to just the HRF for that individual?

  2. Would a participant who for example always responded with an aggression value of 2, be modulated the same as an individual who always responded with an aggression value of 6? Or does the PM analysis consider scale as well, such that, because 6 is more than 2, the modulation would be different?

  3. Is it problematic to have a contrast between the mean and nice conditions if the nice condition has less variability in the modulating variable than the mean condition?

Please let me know if you need more information! We really really appreciate any guidance you can provide on this! Thank you!!!

Is your research centered around comparing the slope effect of aggressiveness between the two conditions? If a participant did not exhibit any variation in aggressiveness within a given condition, the slope effect cannot be determined. In other words, individual-level analysis would not be able to estimate the slope effect.

Is aggression rated on a discrete scale from 2 to 6? To simplify matters, consider discarding the modulation approach and, instead, adopting a straightforward method by modeling each aggression level separately.

Gang Chen