摘要
目的在实际中经常遇到选择哪种非参数多重比较方法的问题,本文考察了五种非参数多重比较方法的性能,并对实际中采用合适的方法提出建议。方法本文考察了Dunn_z,扩展的t检验法,以及秩次转换后的Bonferroni(R_BON),SNK(R_SNK)和LSD(R_LSD)这五种方法,用MonteCarlo模拟来考察五种方法的第一类错误、第二类错误以及判对率,用SAS宏功能编程实现。结果扩展的t检验法与R_LSD等价,R_LSD、R_SNK、R_BON以及Dunn_z犯第一类错误的概率依次减小,但检验效能也依次减小。四种方法受样本量和组数的影响不同。结论 R_BON和R_SNK是两种较优的方法。在实际中,当组数较小时,或组数较大且样本量也较大时,可以选用R_BON;而如果组数较大,样本量较小时,可以选用R_SNK。
Objective How to choose nonparametric multiple comparison procedure (MCP) is frequently encountered in practice.The performances of five methods were investigated and how to choose appropriate multiple comparison method in different situations were discussed.Methods Five methods of Dunn_z,Extended t test and Rank transformed Bonferroni (R_BON),SNK (R_SNK) and LSD (R_SNK) were selected and evaluated in Type I error,Type II error and Correct Decision Rate using Monte Carlo simulation.All above were implemented with SAS Macro.Results Extended t test is equivalent with R_LSD.R_LSD,R_SNK,R_BON and Dunn_z decrease in sequence in probability of committing Type I error,but the power also reduce in this order.Four methods are influenced by sample size and the number of groups in different way.Conclusion R_BON and R_SNK are two optimum methods.The recommended MCP would be R_BON when the number of groups is small or the number of groups is large and sample size is also large.R_SNK would be recommended when the number of groups is large but sample size is small.
出处
《中国卫生统计》
CSCD
北大核心
2011年第5期501-503,506,共4页
Chinese Journal of Health Statistics