Collaborative Research: A New Infrastructure for Monitoring Social Class Networks

在过去的15年里，越来越多、越来越多样化的人选择使用社交媒体进行互动，这些媒体记录了他们交流的数字痕迹，这一发展为研究社会阶级关系的网络基础提供了前所未有的机会。尽管对美国社会阶层是否形成良好的研究有着悠久的传统，但它完全是基于调查和人口普查数据，而且必然忽略了阶级结构和形成的网络基础。这项研究利用了社交媒体互动数量的增加，在人口规模上研究了这种网络结构。由此产生的方法将为研究美国社会阶层内部和社会阶层之间的人际互动提供新的研究基础。通过使用来自美国Twitter用户的完整抓取数据，可以衡量人际互动的阶级障碍。该方法的核心是开发方法，使用概要数据、消息内容的词法分析和地理位置用户的住房估值来衡量用户的阶级状况。为了补充和验证这些行为措施，将对网络边缘的随机样本进行调查。对Facebook用户也将进行类似的分析。结果数据将用于完成第一个基于网络的美国阶级结构的范围和模式分析。传统的？静态分析?在阶级结构中，强调了阶级间在行为和态度（例如养育子女的做法、政治态度）方面差异的大小，而忽略了阶级间联系的模式和将阶级联系在一起的网络。因此，关键问题是，所提出的阶级网络分析是否产生了与主导美国几十年阶级研究的静态分析不同的社会阶级结构肖像。与此同时，过去已经尝试了一些基于网络的阶级分析，这些分析依赖于可能通过调查方法识别的特殊网络行为范围(特别是，选择性交配（人们与教育和职业特征相似的人结婚，以及代际社会流动性）。这里进行的分析将揭示，相对于调查数据中显示的面对面网络中的阶级同质性水平（人们与相似的人交往的倾向），社交媒体是否减少了互动中的阶级障碍。这些分析将为一种新的基于网络的阶级结构分析提供基础。更广泛的影响如果阶级障碍在网上互动中相对较弱，那么阶级结构的标准测量将提供一幅越来越具有误导性的公民社会及其包容性的肖像。然而，网络平台强大的搜索算法使人们能够有效地挑选出与自己相似的人，这也是合情合理的。如果事实证明是后者，那就意味着与传统观点相反，新社交媒体的兴起正在加剧阶级同质化和阶级关系的两极分化。这项研究也有方法论上的回报。由于基于网络的社会阶级结构分析需要对媒体用户的阶级状况进行高质量的测量，因此大部分研究将侧重于开发使这种测量成为可能的方法。用户和变更的社会阶层将通过以下方式进行估算：(a)将地理定位用户与其社区和房屋价值联系起来，(b)利用可用的个人资料数据，(c)对消息内容进行词汇分析，以及(d)对用户进行调查。这些方法可以扩展到对种族、性别和其他特征进行类似的归因，将对社会科学、计算机科学、信息科学和其他学科的研究人员有用，在这些学科中，个体特征的直接信息是稀缺的。该项目还将为康奈尔大学和斯坦福大学的研究生和本科生提供新的研究机会。

SES-1357488David GruskyStanford UniversitySES-1357442Michael MacyCornell UniversityOver the last 15 years, an ever larger and more diverse population is choosing to interact using social media that record the digital traces of their communications, a development that opens up unprecedented opportunities to study the network foundation of social class relations. Although there is a long tradition of research examining whether social classes in the United States are well-formed, it has been based exclusively on survey and Census data and, by necessity, has ignored the network foundations of class structure and formation. This research takes advantage of the rising amount of interaction with social media to examine that network structure at population scale. The resulting methods will provide the basis for a new and novel research infrastructure for investigating inter-personal interaction within and between social classes in the United States.By using data from a complete crawl of U.S. Twitter users, it becomes possible to measure class barriers to interpersonal interaction. The centerpiece of this approach is the development of methods to measure the class situation of users with profile data, lexical analysis of message content, and housing valuations for geo-located users. To supplement and validate these behavioral measures, a survey will be administered to a random sample of network edges. A similar analysis of Facebook users will be carried out. The resulting data will be used to complete the first network-based analyses of the extent and patterning of the U.S. class structure. In conventional ?static analyses? of the class structure, the size of inter-class differences in behaviors and attitudes (e.g., childrearing practices, political attitudes) is emphasized, while the patterning of inter-class contact and networks that link classes together is ignored. The key question, therefore, is whether the proposed network analyses of class yield a different portrait of the structure of social classes than the static analyses that have dominated decades of class research in the U.S. At the same time, some network-based analyses of class have been attempted in the past, analyses that have relied on an idiosyncratic range of network behaviors that may be discerned with survey methods (especially, assortative mating where people marry persons with similar education and occupational characteristics, and intergenerational social mobility). The analyses undertaken here will reveal whether social media reduces class barriers to interaction relative to the level of class homophily (the tendency of people to associate with similar people) revealed in face-to-face networks available in survey data. These analyses will provide the foundation of a new network-based analysis of class structure.Broader ImpactsIf class barriers are comparatively weak in on-line interactions, standard measurements of class structure will provide an increasingly misleading portrait of civil society and its inclusiveness. It is also plausible, however, that the powerful search algorithms of online platforms allow people to efficiently cull for alters who are similar to themselves. If the latter proves to be the case, it means that the rise of new social media are, contrary to the conventional view, increasing class homophily and polarizing class relations. The research also has a methodological payoff. Because a network-based analysis of social class structure requires high-quality measurements of the class situation of media users, much of the research will focus on developing the methods that make such measurement possible. The social class of users and alters will be imputed by (a) linking geo-located users to their neighborhoods and housing values, (b) exploiting available profile data, (c) carrying out a lexical analysis of message content, and (d) administering surveys to users. These methods, which may be extended to carry out analogous imputations of race, gender, and other ascribed traits, will be of use to researchers in the social sciences, computer science, information science, and other disciplines facing the stock situation in which direct information on individual traits is scarce. The project will also provide new research opportunities for graduates and undergraduates at Cornell University and Stanford University.