Under what conditions might a biased but efficient sample be perferable to an unbiased but inefficent one?