基于趋势距离的面板数据聚类方法探讨  

Discussion on Panel Data Clustering Based on Trend Distance

作　　者：王泽东 邓光明[1,2] Wang Zedong;Deng Guangming(College of Science, Guilin University of Technology, Guilin Guangxi 541006, China;Institute of Applied Statistics, Guilin University of Technology, Guilin Guangxi 541006, China)

机构地区：[1]桂林理工大学理学院,广西桂林541006 [2]桂林理工大学应用统计研究所,广西桂林541006

出　　处：《统计与决策》2019年第8期35-38,共4页Statistics & Decision

基　　金：国家社会科学基金资助项目(13BTJ009);国家自然科学基金资助项目(61563013)

摘　　要：鉴于面板数据同时具有时间维度和截面维度的特征,文章引入趋势距离概念,在利用主成分分析法计算数据在各个时间点的主成分综合得分基础上,构造了一种能够体现数据动态变化趋势的综合趋势距离,再利用系统聚类法实现面板数据的聚类。实证结果表明新的方法克服了以往聚类方法中无法体现数据动态发展趋势相似性的问题,聚类结果较为稳定且有很好的可视化效果。In view of the fact that panel data has the characteristics of both time dimension and section dimension, this paper firstly introduces the concept of trend distance, and then uses principal component analysis method to calculate the integrated score of principal components at various time points, on the basis of which is constructed a comprehensive trend distance that can reflect the data dynamic change trend. Finally the paper employs the system clustering method to cluster the panel data. The empirical results show that the new method overcomes the problem that the clustering method can not reflect the similarities of data dynamic development trend, and that the clustering results are stable and have very good visual effects.

关 键 词：面板数据 趋势距离 主成分分析 聚类分析 

分 类 号：C812[社会学—统计学]