学术评价中一级指标测度方法研究——结构方程降维法  

Structural Equation Dimensionality Reduction:A New Method of Dimensionality Reduction for Academic Evaluation Indexes

作　　者：俞立平 Yu Liping(Zhejiang Gongshang University,School of Statistics and Mathematics,Hangzhou 310018)

机构地区：[1]浙江工商大学统计与数学学院,杭州310018

出　　处：《信息资源管理学报》2020年第5期76-84,65,共10页Journal of Information Resources Management

基　　金：国家社科基金项目:学术评价与创新绩效评价问题研究(19FTQB011);浙江省一流学科A类项目(浙江工商大学统计学,管理科学与工程)。

摘　　要：由于学术评价指标众多,导致指标分类复杂,赋权困难,降维可以解决这个问题,但现有主成分分析、因子分析降维是非线性的,会破坏原始指标中包含的大量信息,对评价是不利的。本研究采用聚类分析、因子分析辅助进行指标分类,然后采用结构方程模型建模,对显变量与潜变量之间的回归系数进行归一化处理得到权重,进而计算得到潜变量即一级指标来进行降维,并以JCR2015经济学期刊以及TOPSIS评价方法为例进行实证研究,比较了降维前后评价结果的差异。研究发现,结构方程降维法具有线性降维、方便赋权、降低一级指标之间的相关性、计算方式客观唯一等优点,体现了学术评价的系统性思想;结构方程的稳定性对评价具有重要影响,可适当降低对结构方程的统计检验要求。Due to the large number of academic evaluation index,it is difficult to classify and weight them.Dimensionality reduction can solve this problem.However,the principal component analysis and factor analysis are nonlinear dimension reduction,which will destroy the large amount of information contained in the original index,and will be disadvantageous to evaluation.In order to solve the problem,this paper suggests using cluster analysis and factor analysis to classify index,and using structural equation to model.Then,the regression coefficients between explicit variables and latent variables are normalized to obtain weights,and latent variables which named first-level indexes are calculated to reduce the dimension.Taking the JCR2015 economics journal and TOPSIS evaluation method as an examples,the paper makes an empirical study and compares the differences between the evaluation results before and after the dimensionality reduction.The result shows that the structural equation dimensionality reduction has the advantages of linear dimension reduction,convenient weighting,lowering the correlation between the first-level indexes,and the objective of the calculation method,which reflects the systematic thought of academic evaluation.The stability of structural equation has an important influence on evaluation,and the requirements of statistical test can be appropriately reduced.This paper is based on JCR economics journals for related research,and other disciplines need to be further explored.This method has further popularization value in the therelatively mature evaluation mechanism.

关 键 词：评价指标 降维 聚类分析 结构方程 因子分析 学术评价 

分 类 号：G302[文化科学]