论文部分内容阅读
临床试验设计举例 多组比较Glantz曾从《Circulation》分析了142篇论文,其中39％不用统计,34％用正确的t检验或ANOVA等方法.而27％是应用了不正确的t检验法(图3)。用于两组比较的t检验以5％水平作为显著性界限,这意味阳性结论犯错误的概率不超过0.05.如果有3组比较,t检验需3次,若每次t检验都以5％水平作界限,则得出阳性结论犯错误的概率不是0.05而是0.15;若有4组比较,t检验需6次,则阳性结论犯错误的概率为0.30。这就大大提高了有假阳性
Examples of Clinical Trial Design Comparison of Multiple Groups Glantz analyzed 142 articles from Circulation, 39% of them without statistics, 34% with correct t-test or ANOVA, and 27% with incorrect t-test image 3). The t-test for comparison of the two groups, with a 5% level as the significant cutoff, implies that the probability of making a false negative for a positive conclusion does not exceed 0.05. If there are 3 comparisons, the t-test takes 3 times, and if each t-test takes 5% Level, the probability of making a positive conclusion error is not 0.05 but 0.15; if there are 4 groups of comparison, t-test needs 6 times, then the positive conclusion error probability is 0.30. This greatly increases the false positives