实验组与对照组比较的样本率多重比较方法的模拟研究  

Simulation Study on Five Multiple Comparison Methods for Comparison between Experimental Groups and the Control 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卫学院生物统计系,510515

出　　处：《中国卫生统计》2008年第5期500-503,共4页Chinese Journal of Health Statistics

摘　　要：目的研究5种实验组与对照组比较的样本率多重比较方法之特性。方法用SAS软件编制计算机模拟抽样程序,模拟估算各种参数条件下5种方法的总Ⅰ型错误率、检验效能和判对率。结果5种方法各具特色:Brunden法保守,杜养志法灵敏,Dunnett-SNK法和Bootstrap法较稳健,Dunnett-Levy法逊于Dunnett-SNK法。结论5种方法中没有一种方法能在各种条件下都将总Ⅰ型错误率控制在α水平内。相对而言,Dunnett-SNK法最稳健,较其他方法更值得推荐。如需严格控制总Ⅰ型错误率且组数不超过5,可选择Brunden法。如希望结果灵敏,可考虑杜养志法和Bootstrap法。Objective To explore the characteristics of 5 multiple comparisons procedures with the control in proportions.Methods FWER (family-wise error rate),power and true model of five multiple comparison methods in proportions were compared using the Monte Carlo techniques.Results Each method has its characteristic.The Brunden s method is the most conservative while the Du s is the most sensitive.The SNK and Bootstrap methods are robust.The Levy s method is inferior to the SNK.Conclusion The SNK is robust across a...

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]