基于信息可替代性的评价指标筛选研究

提出一种指标筛选方法,旨在筛选出对评价结果影响显著、反映信息重叠程度低的评价指标。以相对离散系数作为指标的信息含量,依据提出的累计信息贡献率标准删除信息含量明显偏小的指标进行指标的初筛,克服现有研究仅利用相对离散系数进行指标筛选的不足。借鉴聚类分析的思想,取初筛后保留下来的一个指标与其余各指标构成Person相关系数平方的均值,反映该指标的信息可被其余全部指标替代的程度。通过信息可替代性标准剔除信息可替代性较大的指标,保证最终被保留的指标间反映的信息重叠程度低,克服现有研究仅通过两个指标间的相关性筛选指标难以有效降低评价指标集信息重叠的不足。此外,针对剔除信息重叠的指标与剔除对评价结果影响不显著的指标何者优先为宜的问题,提出了显著再相关的指标筛选标准。最后,通过一个实例说明指标筛选方法的可行性。

Study of Evaluation Indices Screening based on Information Substitutability

An indices screening method is proposed for screening indicators of impact on the evaluation results significantly and indicators with low information overlap.The coefficient of variation of an indicator is used as its information content.According to the proposed standard of accumulative information contribution rate to delete the indices with significantly smaller information content in the initial screening indices.And this method overcomes the defect that only using coefficient of variation is unable to screen indices.This paper takes the ideas of cluster analysis,uses the mean value of Person correlation coefficient squared between an index and the other all indices to reflect the degree that information of the index can be replaced all other indicators.The indices with the greater information substitutability are deleted based on the standard of information substitutability.This guarantees that the information overlap degree among the final reserved indices is low,and overcomes the deficiency that the existing research is difficult to reduce information overlap effectively of an evaluation indices set only by means of the correlation between any two indices.In addition,deleting indicators with high information overlap and deleting indices of impact on the evaluation results weakly should all be carried out,but there is a problem of what priority. In order to solve this problem,the significance-correlation standard is proposed.Finally,a numerical illustration is presented to illustrate the effectiveness of the indices screening method.