基于似然比的多步多重比较检验  

A STEP-DOWN TEST BASED ON THE LIKELIHOOD RATIO FOR MULTIPLE COMPARISONS WITH A CONTROL

作　　者：勾建伟 王金德[1] 

机构地区：[1]南京大学数学系,南京210093

出　　处：《南京大学学报（数学半年刊）》2008年第2期131-140,共10页Journal of Nanjing University(Mathematical Biquarterly)

基　　金：Supported by NSFC Project(No.10671089);SRFDP(No.20060254006)

摘　　要：似然比检验是假设检验中一个重要方法,不少最优检验都是似然比检验.文中对带有对照组的多重比较问题提出了一种新的逐步检验法,该检验法正是建立在似然比检验的基础上的.同原有的方法相比,它既能对带对照组的多重比较问题作出更详细的检验结果,推断出究竟哪些新的处理方法效果真正优于对照组的;而且这种逐步检验法只需要至多k-1次假设检验,而它的热却逼近需要2^(k-1)次检验的完全检验法的势.Likelihood ratio is very powerful for hypothesis testing.In this paper we propose a new test procedure based on LRT for multiple comparisons with a control.It not only gives a more detailed conclusion to identify better treatments,such as which of these treatments are better than the control,but also needs only at most k-1 hypothesis tests.Moreover the power approximates the closed testing procedure's.

关 键 词：带对照组的多重比较 ■分布 似然比检验 总体错误推断率 势函数 

分 类 号：O212.1[理学—概率论与数理统计]