逐对剔除的相关系数检验方法及应用

The pair-wise deletion correlation coefficient testing method and its applications

为了定量考察样本相关系数的真实性和稳定性,基于x^2检验方法,通过逐对剔除样本中的元素并构造相关系数组的方式,提出了逐对剔除的相关系数检验方法,并用理想数据样本和真实气候数据验证了该方法的正确性和可行性。结果表明,此方法可以客观、定量地找出样本中影响相关系数真实性和稳定性的极端值,这些极端值的存在会使样本的相关系数发生较大变化。本方法的检验过程简单、物理意义清楚,并且,检验过程被定量化,克服了传统方法(散点图、滑动相关等)的一些不足。

To quantitatively investigate the stability and authenticity of the statistical correlation coefficient, we propose a pair-wise deletion correlation coefficient （PWDCC） testing method which is based on χ2 test, through pair-wise deleting elements of a data sample and constructing a correlation coefficient series. The analysis using both ideal data sample and real climate data confirms the validity and feasibility of the PWDCC. The results show that, this method can objectively and quantitatively determine whether there exist extreme values which lead to large variation of the sample correlation coefficient. Furthermore, the method is simple and has clear physical significance, and it also overcomes some weakness of the traditional methods （the scatter diagram, the slide correlation method, etc）.