can a statistic be both sufficient and ancillary