I thought about this more and realized the approach to take is probably to average all the subject func data first using 3dcalc and then extract a single dataset, rather than comparing tlrc coordinates from individual datasets. I will try this and see how it works out.