I’m running an amplitude modulated deconvolution with 3dDeconvolve.
I’m wondering if I need to scale the modulators before I input them in the model. I believe AFNI automatically mean-centers them, but I’m unsure if the magnitude of the modulators’ ranges need normalizing. My stimuli are audiovisual (and naturalistic). If one modulator is a frequency with a range in the thousands, and one modulator is a value derived from a Likert-type ranking of 1 - 5, will the difference in those magnitudes impact interpretation of which modulator contributes the most variance?
The scaling will directly (inversely) affect the betas. It will not affect the single subject statistics, but it could affect group-level statistics, depending on how consistently the values are scaled across subjects.
Note that if the scaling starts off consistently across subjects, a constant scaling will not affect group results (just the magnitude of the group betas). In such a case, it will not really matter.
If there are multiple modulators, then these comments apply one modulator at a time (the magnitudes do not affect each other, unless you start to introduce numerical truncation differences due to the instability of the regression matrix). If one is 1000 times as large as the other, it might be better to scale them more consistently.
If you ever consider a contrast between modulators (or between modulators and unmodulated betas), suddenly the scaling becomes enormously important. It is preferable to avoid such contrasts.
rick
The
National Institute of Mental Health (NIMH) is part of the National Institutes of
Health (NIH), a component of the U.S. Department of Health and Human
Services.