On the efficient allocation of resources for hypothesis evaluation: a statistical approach