Dept. of Biology, University of California, Santa Cruz, CA 95064, USA.
Trends Ecol Evol. 1994 Jun;9(6):235-7. doi: 10.1016/0169-5347(94)90258-5.
Whenever experiments make a priori predictions about the direction of change in some parameter, one-tailed test statistics offer a potentially large gain in power over the corresponding two-tailed test. This gain is rarely used in ecology and evolution because of (1) the belief that one-tailed procedures are unavailable for most statistical tests and (2) an inherent dilemma in one-tailed tests: how do we handle large parameter changes in the unanticipated direction? The first problem is a misconception, whereas the second is easily resolved by recognizing that one- and two-tailed tests are simply extremes in a continuum of testing options.
只要实验对某个参数变化的方向作出先验预测，单尾检验统计量相对于相应的双尾检验就可能大大提高功效。由于以下两个原因，这种功效在生态学和进化论中很少得到利用：（1）认为单尾程序大多数统计检验都无法使用；（2）单尾检验固有的两难处境：对于意外的方向的大参数变化，我们应如何处理？第一个问题是一种误解，而第二个问题通过认识到单尾和双尾检验仅仅是检验选择的连续体的两个极端就可轻易解决。
