论文部分内容阅读
《医学研究中的偏性》一文(本刊1986年第1期)发表后,有人就多次检验的显著性水准的算法提出了以下疑问:如果说每次检验的显著性水准(犯第Ⅰ类错误的概率的上限(定为α=0.05,则作10次检验总的显著性水准可高达10α=0.5;那么,作20余次检验时,犯第Ⅰ类错误的概率岂不要超出概率值的上限1?这个问题提得好,特此公开答复如下:
After the publication of “The Bias in Medical Research” (the first issue of this issue in 1986), some people asked the following questions about the algorithm of significant levels of multiple tests: If we say the significance level of each test (I The upper limit of the probability of class errors (defined as α=0.05, the total significance level for 10 tests can be as high as 10α=0.5; then, for more than 20 tests, the probability of committing a Class I error 岂 does not exceed the probability value. The upper limit of 1? This question is well-proposed. Hereby I publicly reply as follows: