Balancing Type I and Type II error concerns in fMRI through compartmentalized analysis.

We seek to balance the need to minimize false positives with the need to maximize power. We propose a compartmentalized series of analyses that a priori selects regions of voxels that have different degrees of predicted involvement. Alpha thresholds are allocated based on the strength of expected theoretical relationships. For example, confirmatory studies might allocate most of the error to the regions predicted from the literature and thus use a relatively more liberal threshold on these voxels. Simulations reveal that this technique increases power for hypothesized regions, while maintaining a constant false-positive rate and allowing exploratory analysis.