Source-linked AI summary
Viral Visualizations: How Coronavirus Skeptics Use Orthodox Data Practices to Promote Unorthodox Science Online
Crystal Lee, Tanya Yang, Gabrielle Inchoco, Graham M. Jones, Arvind Satyanarayan
TL;DR
The paper asks how people who distrust scientific institutions use COVID data visualizations to contest public-health authority. Using Twitter network and image analysis alongside Facebook ethnography, it finds that anti-mask groups deploy polished, orthodox visualization practices to support heterodox conclusions, revealing a deeper conflict over science’s role in public life.
Problem
Existing literacy-oriented approaches do not fully explain how groups can use sophisticated data practices to promote conclusions opposed to scientific orthodoxy.
Method
The paper combines quantitative analysis of COVID visualizations circulating on Twitter with qualitative digital ethnography of anti-mask Facebook groups.
Results
Anti-mask groups produce popular, polished counter-visualizations using orthodox methods to argue that public-health measures are unnecessary and that the crisis is over.
Takeaways & Limitations
COVID visualizations function as a battleground over expertise, reflecting an epistemological rift about how data should guide public life.
Takeaways & Limitations
The analysis is bounded by an emic approach that uses community members’ own terminology and avoids labeling them “anti-science.”
Abstract
from arXiv · showhide
Controversial understandings of the coronavirus pandemic have turned data visualizations into a battleground. Defying public health officials, coronavirus skeptics on US social media spent much of 2020 creating data visualizations showing that the government's pandemic response was excessive and that the crisis was over. This paper investigates how pandemic visualizations circulated on social media, and shows that people who mistrust the scientific establishment often deploy the same rhetorics of data-driven decision-making used by experts, but to advocate for radical policy changes. Using a quantitative analysis of how visualizations spread on Twitter and an ethnographic approach to analyzing conversations about COVID data on Facebook, we document an epistemological gap that leads pro- and anti-mask groups to draw drastically different inferences from similar data. Ultimately, we argue that the deployment of COVID data visualizations reflect a deeper sociopolitical rift regarding the place of science in public life.
1 INTRODUCTION
The paper examines how anti-mask groups use orthodox data-visualization practices to challenge public-health narratives, mobilize policy opposition, and expose conflicting interpretations of COVID-19 data.
- Counter-visualizations: Anti-mask groups create “counter-visualizations” that use orthodox scientific methods to make unorthodox arguments against pandemic urgency and public-health measures.They frame these practices through data-driven decision-making and calls to “follow the data.”
- Research approach: The study combines quantitative analysis of nearly half a million tweets with a six-month observational study of anti-mask Facebook groups.Twitter analysis examines visualization circulation and design, while Facebook ethnography examines how groups discuss scientific rigor and COVID data.
- Visualization practices: Anti-mask users produce polished visualizations that resemble those found in scientific papers, health-department reports, and financial journalism.Their visual sophistication contrasts with the assumption that misinformation necessarily reflects weak data literacy.
- Visualization practices: These groups debate government data access and visualization components, including raw data, statistical transformations, mappings, marks, and colors.They use such critiques to argue that social-distancing mandates are unnecessary or ill-advised.
- Epistemological conflict: The paper argues that maskers and anti-maskers can use the same data while reaching radically different conclusions because data interpretation is culturally and socially situated.The disagreement reflects a deeper epistemological conflict rather than simply unequal access to datasets or visualization skills.
2 RELATED WORK
The related work frames data and visualization literacy as socially situated practices, challenging deficit-based solutions and motivating a critical analysis of COVID counter-visualizations as contests over power and expertise.
- Data and visualization literacies: Visualization and media literacy research often treats literacy as interpreting or creating graphs, but social context shapes which visualizations people prioritize and how they understand them.This literature motivates examining users’ political and social contexts rather than treating visualization interpretation as purely individual skill.
- Data and visualization literacies: Anthropological approaches define literacy as historically embedded and locally meaningful, supporting an analysis of anti-mask practices without imposing an expert-prescribed standard.The paper therefore describes what these practices do in context rather than normatively assessing them.
- Critiques of literacy solutionism: Calls for more media literacy can become solutionism, backfire by weaponizing critical thinking, or prove insufficient for addressing culturally situated alternative facts.This literature shifts attention from individual education toward the political and historical conditions surrounding data interpretation.
- Critical approaches to visualization: Critical visualization scholarship treats visualizations as representations of power and emphasizes reflexivity about exclusions, situated knowledge, and alternative analytic practices.These approaches connect visualization design and interpretation to systemic inequalities and political advocacy.
- Critical approaches to visualization: The paper extends critical visualization work by showing that anti-mask groups also identify political power in government data and seek alternatives to official data practices.Their counter-visualizations demonstrate why visualization analysis must account for competing interpretations of expertise and authority.
3 METHODS
The paper pairs computational analysis of Twitter with interpretive digital ethnography of Facebook groups to connect visualization circulation with users’ social interactions and meanings.
- Mixed-methods design: The mixed-methods design combines computer vision and network analysis of Twitter data with digital ethnography of Facebook Groups.This pairing links computational patterns in visualization circulation to interpretive analysis of social-media discussions.
3.1 Twitter data and quantitative analysis
The Twitter study identifies COVID visualization patterns by filtering a large tweet corpus, clustering image features, and constructing a user-interaction network.
- Twitter corpus: The source dataset contained over 390M COVID-related tweets collected from January 21 through July 31, 2020.The researchers hydrated tweet IDs from a dataset assembled using keywords and public-health institution accounts.
- Twitter corpus: The researchers retained tweets with images and explicit chart-related terms after broader data-analysis keywords produced excessive noise.The final keyword set included chart, plot, map, dashboard, vis, viz, and visualization variants.
- Twitter corpus: The filtering process yielded almost 500,000 tweets containing over 41,000 images.Tweets and metadata were stored in SQLite, while images were downloaded to the file system.
- Image analysis: Because the computer-vision model classified only 30% of images, the researchers extracted 4096-dimensional embeddings and clustered them after reducing the space to 100 dimensions.They evaluated k-means solutions for k from 5–40 and manually inspected the outputs.
- Network analysis: The interaction network contained almost 400,000 users and over 583,000 edges, with an average degree of 2.9.Edges represented mentions, replies, retweets, or quote-tweets between users.
3.2 Facebook data and qualitative analysis
The study combines digital ethnography with a case-study approach to examine how anti-mask Facebook groups develop and discuss data practices. Researchers followed five groups, archived posts and interactions, and used grounded theory to synthesize themes while complementing large-scale Twitter analysis.
- Data collection: Researchers followed five Facebook groups with 10K–300K followers from their earliest dates through September 2020.The observation period covered the first six months of the pandemic, from March to September 2020.
- Data collection: They archived posts, comments, Facebook Live streams, and photos of in-person events while taking field notes.CrowdTangle was used to collect posts containing terms for “coronavirus” and “visualization.”
- Mixed-methods design: The Twitter and Facebook analyses serve complementary purposes: statistical representativeness for Twitter and granular social understanding within a singular Facebook community.The authors use the two analyses as foils, pairing large-scale interaction analysis with close study of social dynamics.
- Qualitative analysis: Grounded theory was used to tag qualitative data, identify analytically pertinent themes, and group codes into higher-level concepts.The approach is systematic but flexible, allowing themes to respond dynamically to changing empirical phenomena.
- Terminology: The study uses “anti-mask” as an emic term for heterogeneous beliefs including pandemic exaggeration and support for reopening schools.The authors avoid labeling these groups “anti-science” because that would prevent examining what members mean by “science.”
4 CASE STUDY
The case study examines how COVID visualizations circulate across Twitter communities and asks how groups using similar visualization methods can reach divergent interpretations. It then turns to Facebook ethnography to analyze anti-mask knowledge-making and the epistemological rift underlying those interpretations.
- 4 CASE STUDY: The Twitter analysis identifies communities of users who share or engage with pandemic visualizations through retweets and other interactions.The fourth-largest network consists of users promoting heterodox scientific positions about the pandemic.
- 4 CASE STUDY: Together, the analyses frame divergent readings of similar data as evidence of a fundamental epistemological rift about knowledge.The quantitative overview establishes circulation patterns, while the qualitative study investigates interpretive practices within an anti-mask community.
- 4 CASE STUDY: The Facebook analysis addresses how opposing groups can use similar visualization methods yet reach different interpretations of COVID data.The ethnography examines anti-mask interactions to understand their practices of knowledge-making and data analysis.
4.1 Visualization design and network analysis
Twitter users shared a wide range of pandemic visualization forms across multiple communities, with substantial overlap in chart types. Anti-mask users formed a large, relatively insular network and used visualization styles resembling those of mainstream and expert sources.
- Visualization design: The corpus contained eight major visualization clusters, including line charts, area charts, bar charts, tables, maps, dashboards, and images.Line charts were the largest cluster, while maps were the second largest.
- Visualization design: Line charts commonly depicted exponential case growth, cross-country comparisons, or US state comparisons, often using log-scales in early-pandemic charts.Prominent examples came from the Financial Times and Our World in Data.
- Visualization design: Area, bar, and map visualizations used annotations, labels, shading, and contextual text to emphasize trends, events, categories, or geographic differences.Area charts often marked lockdowns, reopening, peaks, troughs, and rolling averages; bar charts frequently included explainer text.
- User networks: The anti-mask network comprised over 2,500 users, or 9% of the network graph, and was anchored by several prominent skeptic and political accounts.The community also included accounts such as COVID-19 Tracking and Ohio Governor Mike DeWine, who was often targeted by anti-mask protests.
- User networks: British media had the highest in-network retweet percentage at 89.32%, while American politics and right-wing media had the highest original-tweet percentage at 44.75%.The British media network also had 22.92% verified users and averaged 94 engagements per visualization.
- User networks: 82.17% of anti-mask retweets were in-network, while the community had 37.12% original tweets, indicating substantial internal circulation alongside original production.The anti-mask community had roughly similar user and verified-account proportions to other networks.
4.2 Anti-mask discourse analysis
Anti-mask groups present themselves as rigorous data users who distrust mainstream sources and create original visualizations to challenge public-health interpretations. Their discussions focus on missing metrics, data construction, context, visualization choices, bias, and ways to validate or teach their analyses.
- Discourse and data practices: Anti-maskers use data-driven narratives and orthodox visualization forms to justify heterodox beliefs and advocate policies such as reopening schools and businesses.The Twitter analysis alone cannot explain how these groups invoke data and scientific reasoning in policy discussions.
- Emphasis on original content: Some Facebook groups prohibit non-original content and encourage members to follow data without mainstream news or models influencing discussion.Original-content rules can also reflect grassroots efforts to provide local data where official dashboards or open-data portals are absent.
- Emphasis on original content: Members share raw datasets, troubleshoot files, and collectively develop ways to obtain pandemic information not consistently shared across state and local governments.Users exchanged local data and helped one another resolve problems with CSV files and death-date datasets.
- Critically assessing data sources: Members debate which metrics should guide policy, with some prioritizing deaths over cases because they view case rates as sensitive to testing and asymptomatic detection.They also complain that death data are not consistently reported or used in policy decisions.
- Critically assessing data sources: Users question whether governments withhold underlying data and whether coding, cleaning, and aggregation introduce subjective or politically motivated definitions.They argue that absent generally accepted definitions, researchers can define datasets in many ways.
- Critically assessing data sources: They argue that hospitalization and infection rates require contextual interpretation because testing practices and case definitions affect how representative and reliable the measures are.Some users view random sampling as necessary for estimating true infection rates and distinguish symptomatic from asymptomatic cases.
- Visualization and bias: Members debate whether absolute counts, per-capita rates, or untransformed tables provide the most faithful representation of pandemic data.This disagreement reflects broader mistrust of mediation through data transformations and visualizations.
- Visualization and bias: Users acknowledge that their visualizations represent partial perspectives and can contain bias, while invoking controls, source checking, and comparison studies as safeguards.They also scrutinize institutional and pharmaceutical profit motives when evaluating pandemic information.
5 DISCUSSION
The discussion shows that anti-mask communities use sophisticated data practices to challenge scientific authority, while interpreting pandemic evidence through distrust, lived experience, and anti-establishment politics. These practices make better data access alone insufficient for building consensus about scientific findings.
- 5 DISCUSSION: Anti-mask communities treat data literacy as a marker of membership, prestige, expertise, and political action.Their practices include accessing, interpreting, critiquing, visualizing, and sharing data within a community of practice.
- 5 DISCUSSION: Counter-visualizations combine orthodox scientific methods with opposition to scientific authority and anti-establishment ideals.Members use scientific analysis to assert independence from central government, business, and liberal academia.
- 5 DISCUSSION: Anti-mask groups treat pandemic statistics as unreliable when categories such as deaths and cases appear subjective or manipulable.They particularly question whether reported deaths reflect deaths from COVID rather than deaths with COVID.
- 5 DISCUSSION: Limited lived experience and missing hyperlocal data lead these groups to seek more granular evidence and interpret information gaps as possible government or media suppression.National statistics may not reflect users’ local experiences, reinforcing distrust of coronavirus data reporting.
- 5 DISCUSSION: These groups view science as a process of critical scrutiny rather than an institution whose consensus should be accepted.They invoke falsification and paradigm change to frame anomalous evidence as potentially capable of overturning established scientific conclusions.
- 5 DISCUSSION: Anti-maskers’ interpretive framework reflects resentment toward a paternalistic elite perceived as usurping scientific knowledge and demanding intellectual subservience.This account connects their views to a broader partisan “deep story” about elites and authority.
- 5 DISCUSSION: The paper argues that improving data access or visualization quality alone will not produce consensus because anti-mask protesters already use visualizations effectively.Counter-visualization and anti-masking are presented as complementary practices within a culturally situated political movement.
6 IMPLICATIONS AND CONCLUSION
The paper argues that anti-mask counter-visualizations use orthodox methods to advance unorthodox science, revealing a deeper epistemological rift over data, expertise, and public life. It concludes that addressing this rift requires engagement with the social and political worlds in which visualizations are made and interpreted.
- Implications and conclusion: Anti-mask users are prolific and skilled creators of popular counter-visualizations that use orthodox methods to argue public health measures are unnecessary.These visualizations circulate as participatory, heterodox forms of information sharing.
- Implications and conclusion: Coronavirus skeptics treat science as a personal practice emphasizing rationality and autonomy rather than knowledge certified by expert institutions.This helps explain why calls for individual data or scientific literacy can misdiagnose coordinated information campaigns as individual ignorance.
- Implications and conclusion: Deficit-focused media literacy approaches that center individual responsibility can obscure larger structural problems and coordinated disinformation systems.The paper connects this concern to skeptical narratives amplified through collective interaction and influential platforms.
- Implications and conclusion: Visualization research should address social and political dimensions from the outset and recognize knowledge produced through visualization systems as multiple, subjective, and socially constructed.The authors propose shifting from positivist toward interpretivist frameworks.
- Implications and conclusion: The paper also highlights uncertainty communication and inconsistent public messaging as factors that can undermine trust in science.It specifically discusses the CDC’s initial mask messaging and subsequent reversal.
- Implications and conclusion: Data visualizations function as a battleground over the contested role of expertise in modern American life, not merely as tools for understanding epidemiological events.The paper therefore challenges the assumption that better visualization tools necessarily improve public understanding.