用于整体假设检验的截尾强度方法的进一步研究(英文)  

A Further Study on a Truncated Tail Strength Method for Testing an Overall Null Hypothesis in a Dataset

作　　者：张慧[1] 蒋波 

机构地区：[1]曲阜师范大学图书馆,山东省曲阜市273165 [2]阿拉巴马大学生物统计系,伯明翰美国35294

出　　处：《曲阜师范大学学报（自然科学版）》2011年第2期31-39,共9页Journal of Qufu Normal University(Natural Science)

摘　　要：在检验一组m个单个假设都是正确的整体原假设下的显著性时,Tail Strength方法和Fisher概率方法是非常有效的。该文进一步研究了一种截尾统计量.模拟研究充分显示了这种统计量具有以下有特点:(1)在整体原假设成立条件下截尾统计量的分布依赖于截点.(2)当截点小于或等于0.05或者0.005时,m大于500时,在原假设下截尾方法很好的控制了第一类错误.(3)当截点是0.05,m等于100,500,和1000时,截尾方法比Tail Strength方法更显著的效能.(4)在某些情形下,截尾方法具有比Fisher概率法更显著的效能.To overall measure significance for a set of m single hypothesis tests under the overall null hypothesis that all m single hypotheses are true,a tail strength statistic and Fisher＇s probability method are useful and it can be applied.In this paper,we further study the properties of a truncated tail strength statistic proposed by Jiang et al.,in which they consider only the values whose corresponding p-values are less than some specified cutoff for a number of single hypotheses less than and equal to 1,000 in one data set.Simulation studies show that the truncated tail of strength statistic：（1）it has a distribution dependent on the cutoff under the overall null hypothesis,（2） it controls type one error rate quite well with more than more than 500 overall null hypothesis when the cutoff is less than and equal to 0.05（or 0.005）,（3） it has a significantly higher power than the tail strength method with a cutoff of 0.05 for 100（or 500 and 1,000） overall null hypotheses,（4） it has higher power than Fisher＇s method in some cases.

关 键 词：截尾强度方法 整体假设 Fisher概率法 截点 

分 类 号：O212.1[理学—概率论与数理统计]