一种相依极小p值统计量概率分布的近似计算方法

Approximating the distribution of minimal dependent p-values

极小p值统计量常用于多重检验问题,其显著性在各个检验不独立的情形下计算复杂度过高.在单个假设的检验统计量联合服从多元正态分布,且检验统计量之间非负相关的假设下,检验统计量之间的相关系数的均值可以作为检验统计量相关性的度量.基于这个度量给出了一个计算极小p值统计量的分布函数的近似方法.模拟结果表明该方法计算简便,具有较高的逼近精度.

The minimal p-value Calculation of its significance is statistic is often used as a test usually computationally intensive statistic in a multiple testing problem. when the tests are correlated. Under the assumptions that the test statistics for all hypotheses are jointly normally distributed and non-negatively correlated, the average correlation coefficient can be used as a measure of the correlation between test statistics. A simple approximation method for calculating the distribution of the minimal p-value statistic was proposed based on this measure. Simulation results show that this method is computationally efficient and has high approximation accuracy.