Should imperfect data be used to guide public policy?