独立逆抽样下优势比检验统计量的构造  

Construction of Test Statistics for Odds Ratio Under Independent Inverse Sampling

作　　者：古丽斯坦·库尔班尼牙孜 田茂再[1,2,3] Gulistan Kurbanyaz;Tian Maozai(School of Statistics and Data Science,Xinjiang University of Finance and Economies,Urumqi 830012,China;Center for Applied Statistics,Renmin University of China,Beijing 100872,China;School of Statistics,Renmin University of China,Beijing 100872,China)

机构地区：[1]新疆财经大学统计与数据科学学院,乌鲁木齐830012 [2]中国人民大学应用统计科学研究中心,北京100872 [3]中国人民大学统计学院,北京100872

出　　处：《统计与决策》2022年第5期5-10,共6页Statistics & Decision

基　　金：教育部人文社会科学研究规划基金项目(21XJJA910001);国家自然科学基金资助项目(71864034);新疆财经大学科研研究基金(2017XYB017)。

摘　　要：优势比(OR)是医学研究中常用的重要指标,它反映的是疾病和暴露的关联强度。文章在独立逆抽样下采用五种方法构造优势比的检验统计量,即对数-Wald检验统计量(基于样本方差和基于原假设下的方差)、得分检验统计量、似然比检验统计量和基于Filler定理的检验统计量;通过蒙特卡洛模拟对五种检验统计量在控制第一类错误的能力与检验功效方面的表现性能进行评价。结果表明:在独立逆抽样下对优势比而言,基于样本方差的对数-Wald检验统计量是最有效的检验统计量,在相同参数设置下其既能保证犯第一类错误的概率最接近显著性水平,又能使得功效达到最大;其他四种检验统计量在控制第一类错误与保证检验功效方面都存在一定的不足。Odds ratio(OR)is an important index commonly used in medical research.It reflects the correlation strength between disease and exposure.This paper uses five methods to construct the test statistic for odds ratio under independent inverse sampling:logarithm-Wald test statistic(based on sample variance and variance based on null hypothesis),score test statistic,likelihood ratio test statistic and test statistic based on Filler theorem.Through Monte Carlo simulation,the performance of five test statistics in controlling the type I error and power of the test is evaluated.The results show that the logarithm-Wald test statistic based on sample variance is the most effective test statistic for odds ratio under independent inverse sampling,that under the same parameter setting,it can not only ensure that the probability of making the first kind of error is closest to the significance level,but also maximize the efficiency,and that the other four test statistics have some shortcomings in controlling the type I error and ensuring power of the test.

关 键 词：优势比 逆抽样 假设检验 第一类错误 检验功效 蒙特卡洛模拟 

分 类 号：O212.1[理学—概率论与数理统计]