逆抽样条件下风险差的检验

Testing of Risk Difference with the Inverse Sampling

建立了不同的统计量对逆抽样条件下病例对照组实验中的风险差进行检验.这些统计量分别是Wald统计量,Score统计量和似然比统计量.然后采用Fisher-Score的方法求解感兴趣参数的方差.采用近似的方法计算各统计量条件下犯第1类错误的概率和功效.多数情况下,Score统计量比较好,特别是针对感兴趣的样本量小于30的条件下,Score统计量犯第1类错误的概率能控制在置信水平以下.

Inverse sampling is considered to be a more appropriate sampling scheme than the usual binomial sampling scheme when the subjects arrive sequentially and when the underlying response of interest is acute. This article studies various test statistics for testing the ratio in the case study of the inverse sampling. They include the Wald, unconditional score, likelihood ratio test statistics. The Fisher-score is used to get the variance of interest parameter. Then we get the type Ⅰ error and power by Monte Carlo simulation. In most cases, Score statistic is the best, especially when the number of interest parameter is less than 30, and the type Ⅰ error under unconditional score is less than the confidence level.