All-positive parametric modulation in GLM

Hi all,

Yet another hopefully not too stupid question: I’m trying to apply GLM with parametric modulation. Some of my modulations include both positive and negative values - with them I don’t have a problem - but for those that are only positive (0-maxVal), I’m afraid I’m getting something that is an artifact of their un-modulated (all ones) version.

The options I thought to handle this are all problematic in their ways:

  • Ideally, I would contrast with the non modulated version of the same events. Problem: since the modulation is positive, they are highly correlated and I’m not sure how it will be expressed in the results.

  • I can centralize the modulation values around zero. However, I’m not sure this is the right thing to do since my expectation is the zero modulation means baseline (is that true?) and consequently, looking for below-baseline values will be wrong.

  • I can run the two GLM (modulated and all-ones) separately and compare values post-hoc. But then what would be the better way of doing so?

Any ideas are warmly welcomed!

Thank you,

I would do centering in order to distinguish the main effect from the modulated effect.
(suggestion 2). The idea is when you use the modulation, you want to study the effect of the difference.
HTH,
Bertrand

1 Like