定性数据的多重比较问题  

Multiple comparisons in categorical data analysis

作　　者：曹瑞[1] 钱俊[2] 陈平雁[1] 

机构地区：[1]南方医科大学公共卫生与热带医学学院生物统计学系,广东广州510515 [2]南方医科大学生物医学工程学院,广东广州510515

出　　处：《南方医科大学学报》2010年第1期118-120,共3页Journal of Southern Medical University

基　　金：国家自然科学基金(30972554);广东省自然科学基金(915180200400001)

摘　　要：目的阐述定性数据多重比较方法,提出构成比多重比较检验水准校正方法。方法针对构成比多重比较的特殊性,应用Bonferroni法原理,并通过Monte Carlo方法在SAS9.13环境编程模拟验证,提出检验水准的调整方法。结果对于多个率或构成比资料,如果将其分解为若干四格表进行多重比较而不调整检验水准,会导致扩大I类错误的后果。对于构成比的多重比较,校正数并非一般多重比较的两两比较组合数,而是组合数减1,该结果得到了模拟结果的验证。结论应正确运用定性数据的多重比较方法,构成比的多重比较因自由度约束其校正数为组合数减1。Objective To discuss the method for multiple comparisons of categorical data and propose an approach to deal with the percentage data. Methods The method of multiple comparisons for percentages was verified based on Bonferroni methodology and Monte Carlo method using SAS 9.13 software. Results The type I error could be enlarged if the statistical tests were conducted without adjustment of the significant level after dividing the data of several categories or percentages into several four-fold tables. For the percentage data, the correction of adjustment of the significant level was the number of pairwise comparison minus one, as supported by the results of Monte Carlo simulation. Conclusion Multiple comparisons of categorical data should be applied appropriately. Multiple comparisons of percentages data need to be conducted with the number of pairwise comparison minus one.

关 键 词：构成比 多重比较 MONTE CARLO方法 I类错误 

分 类 号：R195[医药卫生—卫生统计学]