理论与数据双驱动的社会分层研究  

Theory and Data Driven Research of Social Stratification

作　　者：梁玉成[1] 贾小双[1] LIANG Yucheng;JIA Xiaoshuang(School of Sociology and Anthropology,Sun Yat-sen University,Guangzhou 510275,China)

机构地区：[1]中山大学社会学与人类学学院,广东广州510275

出　　处：《西安交通大学学报（社会科学版）》2022年第1期25-37,共13页Journal of Xi'an Jiaotong University：Social Sciences

基　　金：国家社会科学基金项目(15ZDB172)。

摘　　要：以往社会分层研究的理论和方法可以归纳为理论和数据驱动两种范式,通过回顾和比较发现其都存在不可避免的局限性,为此提出一种结合二者优势的理论与数据双驱动的社会分层研究框架。该框架将社会阶层看作由类别和等级参数所构成的高维社会空间中聚集的子群体,基于分层理论所提出的阶层测量指标建构“社会阶层空间”,并使用机器学习算法识别出空间中的不同群体,从而进行阶层划分。使用这一框架对中国综合社会调查2017年的数据进行阶层划分,发现其既能区分出地位一致性高的、边界清晰的阶层,也能对地位不一致的、还未形成阶层的利益群体进行准确识别。此外还发现:(1)将当前中国社会划分为三个阶层是最好的划分方式,三个社会阶层在经济、声望、文化等维度上的特征分布都存在高、中、低的等级差异;(2)分层的指标并不是越多越好,对中国当前社会阶层划分和对个体阶层测量最有意义的指标是单位类型,其次是职业社会经济地位。The theories and methods of previous social stratification research can be summarized into two paradigms:theory-driven research and data-driven research.Comparison and review of these two paradigms revealed that both of them have inevitable limitations.Thus,a framework that combines the advantages of these two paradigms is proposed,which is called theory-and-data-driven social stratification research framework.To define social classes,this framework regards social class as subgroups gathered in a high-dimensional social space composed of nominal and graduated parameters,and constructs the space of social class using dimensions which are selected from the criteria proposed by the social stratification theories,and uses machine learning algorithms to identify the subgroups in this space.When using this framework to identify the social classes of samples from CGSS 2017,it s found that this framework can not only distinguish social classes that have high consistencies and clear boundaries,but also accurately identify subgroups with inconsistent status which have not yet formed a social class.In addition,it was found that:(1)Chinese society can be stratified into three social classes which can be graded into high,medium or low level on economy,prestige,culture and other feature dimensions;(2)Using as many dimensions as possible is not necessary,the most useful criterion for social stratification is Danwei,followed by ISEI.

关 键 词：社会分层 理论驱动 数据驱动 机器学习 社会阶层空间 阶层测量方法 

分 类 号：D013[政治法律—政治学]