Malhotra, N., Settle, J., Wojcieszak, M., Bergeron-Boutin, O., Guo, F., Franco, A., Herdagdelen, A., Mason, W., Allcott, H., Barbera, P., Brown, T., Crespo-Tenorio, A., Freelon, D., Gentzkow, M., Gonzalez-Bailon, S., Guess, A., Iyengar, S., Kim, Y. M., Lazer, D., Moehler, D., Nyhan, B., Pan, J., Rivera, C. V., Thomas, D., Thorson, E., Tromble, R., Wilkins, A., Xiong, B., de Jonge, C. K., Stroud, N. J., & Tucker, J. (2026). Political unfriending on Facebook: Evidence from over 180 million users. The Journal of Politics. https://doi.org/10.1086/744929

View paper

Summary

This paper provides the first large-scale behavioral evidence on political “unfriending” on Facebook, using aggregate day-level data from over 180 million adult U.S. users combined with a linked six-wave panel survey of roughly 72,000 respondents during the 2020 U.S. election and its aftermath. The central argument is that while severing online ties is rare overall, users disproportionately cut cross-cutting (politically dissimilar) friendships rather than aligned ones — and that this behavior is dynamic, spiking around contentious political events and concentrated among strong partisans and politically engaged users. The authors reframe political homophily as a process of tie dissolution rather than a static network property, and they validate that survey self-reports of unfriending meaningfully track actual on-platform behavior.

Key Contributions

  • First large-scale behavioral distinction between severing of aligned versus cross-cutting online ties, drawing on 180M+ users.
  • Shifts the study of political homophily from static network composition toward the dynamic process of tie dissolution.
  • Maps fine-grained temporal fluctuations of unfriending across a contentious election period, benchmarked against New Year’s “network cleaning” spikes.
  • Methodological validation that survey self-reports of unfriending correlate with real on-platform behavior, countering concerns about expressive responding.
  • Documents heterogeneity in unfriending by ideological strength and civic engagement.

Methods

The authors analyzed aggregate day-level behavioral data from 183,883,696 monthly active U.S. adults (each with at least one liberal and one conservative friend) spanning June 2020 to February 2021. A linked, post-stratified six-wave panel survey of 71,930 consenting users matched self-reports to on-platform logs. User and friend ideology were estimated via an internal classifier (liberal ≤0.4, conservative ≥0.6; strong ideologues <0.2 or >0.8), supplemented by survey party ID and ideology. Tie severing was measured as the proportion of pre-study cross-cutting vs. aligned friends unfriended, normalized by network size, with risk ratios comparing the two. Analyses included temporal plots of daily unfriending, heterogeneity across ideological strength and civic-engagement quintiles, logistic regressions with demographic controls, and validation of self-reports over a 90-day window. The study was part of the preregistered U.S. 2020 Facebook and Instagram Election Study (FIES), with academics holding final publication control.

Findings

  • Liberals severed 2.66% of conservative vs. 2.14% of liberal ties (~24% higher cross-cutting); conservatives severed 2.80% of liberal vs. 2.09% of conservative ties (~34% higher).
  • Pre-study networks were more aligned than cross-cutting, especially for conservatives (60.5% conservative friends), but still politically diverse.
  • Unfriending spiked around Election Day, January 6th, and Inauguration Day; conservatives quadrupled unfriending of liberals on Election Day and also cut co-partisans more, while liberals showed more modest spikes and moderates were largely unaffected.
  • Strong conservatives showed the most extreme cross-cutting unfriending (risk ratio 1.52); heterogeneity by ideological strength was clearer among Republicans than Democrats.
  • Civic engagement strongly predicted cross-cutting unfriending (liberal rates rose from 1.5% in the lowest quintile to 3.2% in the highest) and was a stronger predictor than ideological strength.
  • Self-reports showed meaningful concordance: 62% reporting unfriending had done so; 84% reporting none had not; reported cross-cutting unfriending was less accurate (54%).
  • Motivations included offensive content (65.3%), January 6th (49.8%), harassment (46.8%), political disagreement (44.2%), and over-posting politics (37.5%).

Connections

This paper sits within the broader affective-polarization and echo-chamber literature, and it is a companion output of the 2020 Facebook and Instagram Election Study, directly related to Bergeron-Boutin2026-yh (sharing a lead author) and to other FIES platform-experiment work such as Guess2023-ur, Gonzalez-Bailon2023-uy, and Nyhan2023-gb. Its treatment of partisan network composition and selective exposure extends earlier Facebook network studies like Bakshy2015-rn and Messing2014-jc, while its focus on affective polarization and partisan identity connects to Iyengar2019-jj, Bail2018-fk, and Osmundsen2021-et.