Research
Network Echo: How Networks Shape the Evolution of Linguistic Culture in Alt-Right Echo Chambers
Inducing cultural change in online communities is a key problem for platforms and in digitization policy as concerns mount about echo chambers that foment toxic cultures. Yet the evidence on the role that echo chambers play in producing and reinforcing extreme online communities is mixed. While the prior literature has prioritized understanding the role of the infrastructure of social media, this study prioritizes understanding the role of social network structure through the construct of network echo: reflective, within-group communication that structurally insulates a subgroup of a community from the broader community’s culture. I hypothesize that reducing the network echo of a community reduces the linguistic culture expressed by those in the echo chamber. I test this hypothesis in the setting of online communities by leveraging a large-scale natural experiment on the platform Reddit. The natural experiment shocked the networks of an Alt-Right community, transforming it from a high-echo, insular network to a low-echo, integrated network. Across several ways of measuring language use, including measures derived from natural language processing embedding techniques, I find that this shock causes a decrease in Alt-Right language imported to other communities by Alt-Right members (including a decrease in hate speech). This result overturns the conventional wisdom that breaking up echo chambers will simply displace problematic conversations.
From Contestation to Camaraderie: Structural Similarity Dampens Derogatory Discourse in Polarized Social Groups
Partisan animosity has been on the rise around the world and has been linked to such consequential outcomes as political violence. It is especially prevalent in online social interactions and frequently results in derogatory discourse. The authors develop a novel theoretical account of the network-structural antecedents of derogatory language use. In polarized online groups, they hypothesize that the greater the structural similarity between two individuals, the less will be their likelihood of using derogatory language with one another. They further argue that this relationship will be moderated by the degree of group polarization. Using a node embedding algorithm (i.e., node2vec) to derive an omnibus measure of interpersonal structural similarity, they find support for the theory using a dataset that encompasses more than 25 million comments made by over 1.7 million users in six polarized communities on Reddit. They discuss implications for research on partisan animosity, group polarization, the measurement of structural similarity, and the interplay of structure and culture.
True Believers and New Believers: The Impact of Tactic Efficacy Beliefs on Social Movement Diffusion
When does exposure to social movement activity in one domain catalyze acts of protest elsewhere? Prior research on the diffusion of protest has overlooked how the collective beliefs of dormant populations—those not yet engaged in activism—may influence their participation. We show that the susceptibility of a region to the spread of social movements depends on the region's tactic efficacy beliefs---shared attitudes about the efficacy of social movement tactics. Using 1.6 million posts from 29,000 users on the largest online forum for ride-sharing drivers, we show that counterintuitively, regions with strong tactic efficacy beliefs witness lower levels of ride-share mobilization when exposed to other movements. By contrast, efforts to organize in regions with weaker tactic efficacy beliefs are more likely to result in protest activity growth. We employ a shift-share instrument (SSIV) to estimate the causal impact of social movement mobilization exposure on real-world labor movement mobilization activity based on 2.6 million news stories across 14 social movements. Our conceptualization of tactic efficacy beliefs contributes by developing a novel explanation for the spread of social movements.
Consonance Versus Dissonance: How Exposure to Unfamiliar Colleagues Within and Across Network Communities Affects Social Belonging and Network Change
Organizations vary in the degree to which their members experience social belonging. Interventions designed to boost belonging have typically focused on changing individuals’ mindsets. We instead develop a structural intervention that seeks to foster belonging by exposing people to unfamiliar colleagues---ones they are not in regular contact with. We consider two forms of such exposure: consonant, to colleagues from the same network community as the focal actor; and dissonant, to colleagues from different network communities. We hypothesize that consonant exposure engenders more group solidarity, more persistent relationships, and enhanced social belonging. We test these expectations in a pre-registered field experiment at a non-profit organization. Participants (N=213) engaged in a facilitated professional development experience with unfamiliar colleagues and were randomly assigned to either consonant or dissonant groups. Although the anticipated solidarity advantage of consonant exposure was only marginally significant when assessed immediately following the intervention, consonant-condition participants maintained more intervention-group ties and reported greater social belonging three months after the intervention. Yet, pointing to the potential tradeoffs of consonant versus dissonant exposure, dissonant-condition participants experienced steeper declines in network constraint and greater increases in betweenness and closeness centrality. We discuss implications for research on social networks, workplace belonging, and organizational interventions.
Moving the Goal Posts: Exposure to AI-made Products Changes Evaluation Criteria for Human-made Products
A central question in economic sociology concerns how creative output is valued. The question has gathered particular significance as AI becomes increasingly adept at a range of cognitively complex tasks, most notably creative endeavors. Each demonstration of AI’s increased capability adds to concerns about AI replacing humans across economic domains, resulting in de-skilling and unemployment, even for highly skilled workers. Underlying the debate about AI’s impact on employment—whether it will create new jobs or replace human labor—is a fundamental question: what constitutes valuable work? How do advances in AI’s capabilities affect the criteria used to value work? We propose that as the capabilities of AI approach those of humans, people place more importance on evaluative criteria that cannot be applied to AI, effectively “moving the goalposts” to favor distinctively human qualities. We test this argument in two studies: a natural experiment of art-related corpuses and a randomized experiment. In discussions on Reddit groups related to art and critical reviews on ArtNet, we find an increase in the relative frequency of comments discussing evaluative criteria not applicable to AI (versus criteria applicable to AI) and a more positive stance towards these criteria after the widespread introduction of OpenAI’s ChatGPT. In an online experiment, we find that participants who evaluate an AI-made sculpture before a human-made one are more likely to list AI-incompatible criteria as being more important to their evaluation and to rate the human-made sculpture higher on these criteria, relative to a control group of participants who only evaluate human-made sculptures.
CULTURAL STORAGE: THE ORGANIZATIONAL SHIP OF THESEUS
We investigate the ways that culture, and especially language, can encode organizational priorities and enable organizational memory. Culture's ability to encode priorities creates differences between organizations in which types of knowledge can be efficiently communicated. Like an organizational Ship of Theseus, we ask how much of the language of the original members remains when they have all been replaced. We find that by iteratively replacing every member of the organization, a substantial percentage of the updated language reflects the founding members’ priorities. Common language is a type of limited codifiability that makes certain (possibly tacit) routines and situations more efficient to describe within an organization. We test the model using data from a popular online strategic communication cooperation game. By exploiting a panel of partially overlapping teams, we find that players bring communicative routines with them when they join a new team, and are able to transfer this knowledge to the new team.
PEERING INTO CAREER CHOICE: PEER EFFECTS, ETHICAL HOMOPHILY AND SOCIAL PERSUASION
We seek to unpack the mechanisms by which social learning operates, focusing specifically on ethical homophily. Using the random assignment of MBA students into learning teams, we evaluate the impact of prior industry composition of learning teams on students’ industry choices following the MBA program and find that peer effects are in general negligible, except in the case of ethical homophily (measured as LDA-based topical similarity on an ethical dilemma essay) where peer effects on industry choice become large and significant. The proposed mechanism is that ethical homophily makes communication either more efficient by means of a common language around ethics and values, or more persuasive because those values are shared.