基于配对实验设计的2×2列联表中相对差的统计推断

Statistical Inference for Relative Difference in Matched-pair Design Correlated 2×2 Table

为了消除混杂现象中讨厌参数的影响,更有效地研究因素间的关联性,2×2配对实验设计是一种简单且行之有效的方法.在对2×2列联表的相对差进行极大似然估计时可能会出现相对差分母为零的情况,从而使相对差的估计出现无限大的偏差并且方差无限.为避免出现上述情况,本文在抽样过程中采用逆抽样对相对差进行极大似然估计,利用Wald统计量及Score统计量,通过Monte carlo模拟得出结论:依据覆盖概率和平均区间宽度,Wald统计量和Score的效果基本相同且令人满意,相比之Wald统计量比Score统计量效果稍差一点.

In order to eliminate the influence of the hated parameters in the mixed phenomena and study the relationship among factors more effectively, the 2×2 matched-pair design sampling is a simple and effective method. When the maximum likehood estimation of relative difference is perform , the denominator of relative difference may be zero. In order to avoid the above situation, the inverse sampling method for MIE of the relative difference in the process of sampling was used. Wald statstic and Score statistic were used for test. After Monte Carol simulation, the following conclusions were obtained： In terms of coverage probability and average interval width, the effect of Watd statistic and Score statistic are similar and satisfactory, but the former inferior to the latter.