How good are empirical predictions of natural mortality