Schulz, W., Robertson, C. E., Sjåstad, H., Ashokkumar, A., Oswald, L., Yang, J., Munzert, S., Bavel, J. J. V., & Guess, A. M. (2026). Online Discourse is Falsely Polarized by Self-Selection. https://doi.org/10.31234/osf.io/7z69m_v1
Summary
This paper examines false polarization on social media—the mistaken perception that public opinion is more divided than it actually is—and locates its cause in users’ self-selection into political posting. Combining over a decade of linked behavioral data, controlled experiments, and a 40-country survey, the authors argue that people with more extreme ideological views are disproportionately likely to post about politics. The visible discourse is therefore a skewed, non-representative sample of the underlying population, systematically exaggerating the appearance of division. The contribution is less about discovering a new phenomenon than about rigorously quantifying its magnitude and mechanism across multiple data sources and national contexts.
Key Contributions
- Comprehensive, multi-method evidence quantifying the magnitude and scope of false polarization online.
- Links large-scale behavioral data (over 1.7 million posts) to individual-level ideological attitudes.
- Isolates selective posting as a specific generative mechanism, distinguishing apparent from actual polarization.
- Establishes cross-national generalizability through a 40-country survey (combined N = 49,222).
Methods
The authors triangulate across three designs: (1) observational analysis of more than a decade of social media behavior linked to individual ideological attitudes, spanning over 1.7 million posts; (2) a series of controlled experiments testing the posting-selection mechanism; and (3) a global survey covering 40 countries. The convergence of observational, experimental, and survey evidence is central to the paper’s claim that the pattern is robust and not an artifact of any single data source.
Findings
- Individuals with stronger or more extreme ideological attitudes are more inclined to post political content, producing a skewed visible discourse.
- The polarization observed in online discourse exceeds the actual polarization of opinions in the underlying population.
- The pattern holds consistently across observational data, experiments, and the cross-national survey.
Connections
This paper contributes a self-selection mechanism to the broader debate over whether social media genuinely polarizes publics or merely distorts perceptions of them, connecting to large-scale platform-effect studies such as Guess2023-ai, Guess2023-ur, Nyhan2023-gb, and Allcott2025-jb, and to the exposure-and-curation literature represented by Bakshy2015-rn and Barbera2015-fw. Its emphasis on misperceived rather than actual division complements work on affective polarization and partisan identity like Iyengar2019-jj and Osmundsen2021-et, and speaks to the echo-chamber and selective-exposure claims examined in Bail2018-fk and Flaxman2016-lm.
Podcast
A research-radio episode discusses this paper: 🎧 MP3 · Spotify