基于广义回归神经网络的社会保障待遇公平感模型

Model of the Sense of Fairness in Social Security Based onGeneralized Regression Neural Network

对养老等社会保障主观感受的研究越来越受关注,但所用统计方法都是基于研究对象与影响因素之间的线性关系。由于人及人类社会关系复杂,变量之间往往呈现非线性关系。采用广义回归神经网络(GRNN)算法,建立养老等社会保障待遇公平性主观感受(公平感)非线性模型。将来源于中国社会科学院2015年的“中国社会状况综合调查”(CSS2015)的“养老等社会保障待遇”公平感分为“不公平”与“公平”两类。GRNN模型对训练集、测试集预测整体精度分别是81.4%、90.0%,分别优于支持向量机(SVM)模型的预测精度(80.1%、89.5%),以及二元逻辑回归(BLR)预测结果(79.4%、88.5%)。基于5335份调查样本的研究表明,养老等社会保障待遇公平感与4个影响因素(教育程度、医疗保障满意度、就业保障满意度、总体社会保障状况满意度)之间存在非线性关系。因此,应用GRNN建立养老等社会保障待遇公平感非线性模型是成功的。

Increasing attention is being focused on the sense of fairness in social security such as pension,by applying the statistical methods based on the linear relationships between the research objects and the influencing factors.Because of the complexity of the relationships among human and human society,there are often non-linear relationships between variables.This paper,for the first time,reports that the generalized regression neural network(GRNN)algorithm is used to establish non-linear models for the sense of fairness in social security such as pension.The data of the sense of fairness in social security were taken from the Chinese Social Survey of the Chinese Academy of Social Sciences in 2015(CSS2015),and were divided into two categories:unfairness and fairness.The prediction accuracies of GRNN model for training set and test set is 81.4%and 90.0%respectively,which are better than those of support vector machine(SVM)model(80.1%and 89.5%)and binary logistic regression(BLR)model(79.4%and 88.5%).Based on 5335 survey samples,the investigation shows that there are non-linear relationships between the sense of fairness in social security such as pension and the influencing factors used in this paper(education level,medical security satisfaction,employment security satisfaction,overall social security status satisfaction).Therefore,the application of GRNN to establish a non-linear model of the sense of fairness in social security such as pension is successful.