On Cox's Criterion for Discriminating between Alternative Ancillary Statistics