样本率多重比较方法的模拟研究  

Simulation Study of Pair-wise Comparisons in Proportions

作　　者：钱俊[1] 陈平雁[1] 

机构地区：[1]广州南方医科大学公共卫生学院生物统计系,510515

出　　处：《中国卫生统计》2009年第2期131-134,共4页Chinese Journal of Health Statistics

摘　　要：目的采用MonteCarlo方法模拟研究样本率多重比较方法特性。方法用SAS9.13软件编程,模拟估算不同参数条件下各种样本率多重比较方法的总Ⅰ型错误率、检验效能和判对率。结果不同方法各具特色,能较好控制总Ⅰ型错误率且检验效能较高的方法有Ryan、Perm、Boot和Hommel、Hoch、StepS、StepB方法。结论如需严格控制总Ⅰ型错误率在α水准内,Hommel、Hoch、StepS、StepB效能较高;如允许总Ⅰ型错误率略溢出(1.15α以内),Ryan、Perm和Boot法效能较高。Objective To explore the characteristics of pair - wise comparisons procedures in proportions using Monte Carlo methods. Methods FWER,Power and true model of each procedure are compared using the technology of simulation sampling in SAS 9. 13 programs. Results Each method has their characteristics. Some procedures are sensitive and can control FWER on the whole,such as Ryan,Perm,Boot,Hommel etc. Conclusion If FWER is required strictly control, Hommel, Hoch, StepS, StepB can provide adequate power. If FWER is allowed slightly overflow （within 1.15α） ,Ryan,Perm and Boot are sensitive methods.

关 键 词：多重比较 总Ⅰ型错误率 检验效能 判对率 

分 类 号：R195[医药卫生—卫生统计学]