Best way to look for effect of within-subject continuous variable?

AFNI Message Board
1

Hi AFNI experts,

I’m looking for the best way to look for a relationship between a certain continuous behavioral metric and beta image outputs of a GLM.

Each subject has 3 betas representing different parts of a task (difficulties), and each beta has an associated behavioral score. I expect these behavioral scores to differ a little between the three, but for now am only interested in the general effect of this score on activation. I will probably eventually want to know the effect of these 3 conditions as well, so I’m hesitant to average them.

I don’t necessarily expect that this relationship is linear, though it’s probably at least monotonic if not.

Is 3dMVM the best way to do this? I was going to just use a Spearman rank correlation, but am concerned that this may be wrong without accounting for having multiple measures per subject?

Thank you,
James

2

I expect these behavioral scores to differ a little between the three

You may consider centering the covariate per each difficulty so that it would not ruin the interpretation of cross-difficulty effects.

Is 3dMVM the best way to do this?

Use 3dLME, and check out example 2 in the help:

3dLME -help | less