On the validity of lesion-behaviour mapping methods