一种基于MOVER法中介效应置信区间的构造方法

A MOVER-based Method to Construct the Confidence Interval of Mediating Effect

目的提出一种基于MOVER法中介效应置信区间的构造方法,并与Sobel法、百分位数bootstrap法、偏差校正的百分位数bootstrap法、乘积分布法和Monte Carlo法进行比较。方法借鉴MOVER法的思想,提出一种新的构造中介效应置信区间的MMA方法。在不同参数设置下,通过计算机模拟,比较MMA法与现有方法的第一类错误率、检验效能、覆盖率及区间宽度。同时,将MMA法与现有方法应用于多发性硬化患者数据,对这些方法做进一步的比较。结果 MMA法在样本量大于25时均能控制第一类错误率,偏差校正的百分位数bootstrap法的第一类错误率可能会膨胀;除了偏差校正的百分位数bootstrap法,MMA法的检验效能明显高于其余方法;随着样本量的增大,MMA法的覆盖率逐渐趋近0.95;在部分模拟背景下,MMA法的区间宽度最窄。结论 MMA法是一种稳健且检验效能有明显提高的中介效应置信区间构造方法。

Objective To propose a MOVER-based approach for constructing the confidence interval(CI)of mediating effect and then compare the performance of the proposed MMA method with Sobel,percentile bootstrap,bias-corrected percentile bootstrap,distribution of product and Monte Carlo methods.Methods Borrowing the idea of MOVER,we proposed a novel MMA method to construct the CI of mediating effect.Simulation was carried out to evaluate the type I error rate,test power,coverage probability and confidence width of MMA and the existing methods under various simulation settings.Further,the MMA together with the existing methods were applied to a real example.Results Compared with the existing methods,MMA can generally control the type I error rate well when the sample size is larger than 25 and its test power is higher than other existing methods except for the bias-corrected percentile bootstrap method.However,the bias-corrected percentile bootstrap method may have the inflated type I error rate.The coverage probability of MMA is close to 0.95 as the sample size increases and its confidence width is the smallest in some cases.Conclusion The proposed MMA method is a robust and powerful method.