I have some questions concerning the method I've been using to estimate stimulus-correlated motion in our data, and wondered if anyone has found a sound approach.

For each subject, I have been correlating the task reference vector with each of the 6 motion orts generated by 3dvolreg, across the time series of the task. However, I am concerned that this is not really helping me determine whether or not motion occurs in a cyclic fashion with the task. While it might let me know if there are different levels of motion during each condition, I don't think this will inforn as to whether subjects move at the start or end of each block.

I was wondering if 3dDeconvolve might be a more appropriate tool for estimating stimulus correlated motion, using the task reference vector as the stimulus input, and the orts as the data input. One idea might be to create a new additional stimulus input that models hypothesized stimulus-correlated motion. Then, resulting fit coefficients could be compared between patient groups to see if there is a significant difference in the correspondence between motion adjustments and the task.

Additionally, if this does seem appropriate, should the absolute values of rotation and displacement adjustments be used, so, e.g., left doesn't cancel out right?

Thank you,
Susan