 |
 |
Hi SPMers, I am currently interested in investigating the neural correlates for error and force modulation as well as their interaction in the context of a motor paradigm. In particular, I am having some doubts about how to model the interaction term between error and force. My functional paradigm consists of subjects viewing a curve consisting of ramps and plateaus that moves across the screen. Subjects control a bar in the middle of the screen with a squeezable device in their hand. With increasing force, subjects can move the bar up the screen; upon letting go, the bar falls back to the bottom of the screen. As the ramps and plateaus move across the screen, the subjects are to move the bar so that it always stays on the curve. The visual feedback allows the subjects to adjust their movements so as to stay on the curve as accurately as possible. Error can be calculated with each event (ramp, plateau) by looking at the root mean square (RMS) deviation from the curve and subject performance. The final height of the ramps and plateaus are at either 10% or 5% of the maximum grip force measured previous to the MRI session. With this paradigm, we would like to investigate: 1 the neural correlates of force (10% vs 5%), 2 the neural correlates of error, 3 and the interaction of force and error. We are debating between a parametric modulation of conditions, ramp and plateau, by force (10 for 10% max force events, and 5 for 5% max force events), by error, and by the interaction which is the product of the force and error modulation vector. The interaction modulator was always placed last in the list of modulators so as to be orthogonalized by the first two since we wanted to observe pure effects of the interaction. This model however did not yield significant results for the interaction term, yet we were unsure if the multiple orthogonalizations were resulting in little room for variance explanation by this term. Alternatively, we also thought of a multiple regression method to model the interaction term. We first dichotomized error into small and big errors (i.e. smaller or greater than the median error for a run). Big and small errors were then further dichotomized based on whether they occurred during a big (10% max force) or small (5% max force) force event. These errors were then multiplied by 5 or 10 accordingly. We then had four regressors: (1) big errors big forces - EF (2) big errors, small forces - Ef (3) small errors, big forces - eF (4) small errors small forces – ef. Each value was treated as a block event with the duration of the force event and convolved with the HRF. We hypothesized that we could model the interaction terms as such: +EF + Ef - eF - ef (where big errors arise irrespective of force) or the inverse. I would appreciate anyone’s feedback on this model. Finally, if there is a better way of modeling this in SPM, I would be very grateful for any insight you may have.