Hi,
I have a conceptual question regarding the scaling of parametric regressors. I have two groups of subjects who are rating a stimulus on a 4-point scale. I plan on comparing how the increase in activation associated with the ratings is different between groups. It's unlikely that all subjects will express all four ratings, and one group is likely to have higher ratings on average.
I'm wondering if the height of each event in my timing file (i.e., the third column in a 3-column file) needs to be scaled to the same absolute range for all subjects, irrespective of whether the subject used 4 ratings (e.g., [-1 -0.333 0.333 1]), 3 ratings (e.g., [-1 0 1]), or 2 ratings (e.g., [-1 1]).
That approach seems to ignore the absolute magnitude of the rating, so I'm wondering if the relative scaling of the event heights is more important: 4 ratings (e.g., [-1 -0.333 0.333 1]), 3 ratings (e.g., [-0.667 0 0.667]), or 2 ratings (e.g., [-0.333 0.333]).
Hopefully that's clear. My concern would be biasing my group-level test since the groups are likely to differ on their ratings and potentially use different ends of the scale.
Thanks!
David
|