两个百分数差别的精确法检验  

The Exact Test Method to Test Difference Between Proportions

作　　者：詹绍康[1] 

机构地区：[1]上海医科大学

出　　处：《中国卫生统计》1990年第4期31-34,共4页Chinese Journal of Health Statistics

摘　　要：对两个百分数的差别作统计学检验,在许多统计著作中介绍了以超几何分布为基础的Fisher精确法。作者认为,只有当边际合计都固定时,Fisher精确法才是合适的。事实上,有不少四格表资料并不具有“边际合计都固定”这一性质。如果只有两总体率相等π_1=π_2及两样本含量n_1和n_2固定的条件时,应该以两个二项分布的联合概率为基础来作检验。上述两种方法的理论依据不一样,所作的检验假设也不相同。由于两个二项分布联合概率所具有的特点,作者认为应对拒绝域再作一番研究。在传统的统计学检验中,差数极端法和概率极端法是一致的。而对于本文所提到的方法,却产生了这个以前没有遇到过的问题,作者认为值得作进一步探讨。Fisher's exact test method based on hypergeometric distribution has been introduced to test difference between two proportions in many statistical books. This method is appropriate only when all marginal totals are considered fixed.Unfortunatelly this is not true for quite a few of 2×2 tables. Statistical test should be based on the products of two binomial distributions with parameters (n_1, Π) and (n_2, Π_2) provided that only n_1 and n_2 are fixed and null hypothesis is Π_1 =Π_2, The two methods previously introduced have different theoratical basis and different null hypothesis. Being use of products of two binomial probabilities a further study on rejection region is necessary, There is no difference between the difference-extreme method and the probability-extreme method in most statistical tests.It is worth while further studying the problem which has never occurred before and arisen when the method discribed in this paper is applied.

关 键 词：四格表 精确法检验 超几何分布 

分 类 号：R195.1[医药卫生—卫生统计学]