Blocking of counter-partisan accounts drives political assortment on Twitter

Cameron Martel; Mohsen Mosleh; Qi Yang; Tauhid Zaman; David Rand

doi:10.1145/3631478

In brief

Using field experiments and surveys on Twitter, users were 11 times more likely to block counter-partisan than co-partisan accounts, and Democrats were 26 times more likely than Republicans to block opposing accounts. The primary stated reason for blocking (27%) was avoiding seeing the blocked user's content, not political identity per se, suggesting users employ blocking as a tool for feed curation rather than purely political avoidance. This blocking asymmetry appears driven in part by differences in content quality and partisan slant between partisan accounts, revealing how selective tie prevention—not just tie formation—drives political assortment online.

Contributions

Demonstrates empirically that blocking (social tie prevention) is a powerful but understudied mechanism driving political assortment and echo chambers, comparable to or larger than effects of preferential follow-back (tie formation).
Provides first causal evidence via field experiments on Twitter that users exhibit strong preferential blocking of counter-partisans (11–4.3× more likely, depending on experiment and party).
Documents stark partisan asymmetry in blocking: Democrats 26 times more likely to block Republicans; Republicans 4 times more likely to block Democrats in field experiments, with reduced asymmetry (~5.5×) in survey experiments.
Reveals the primary motivation for blocking is informational (avoiding unwanted content; 27% cite this explicitly), not affective polarization or fear of harassment per se, complicating narratives that blocking is driven primarily by partisan animosity.
Shows asymmetry correlates with content differences: Republican bot accounts in field experiments shared lower-quality, more politically slanted, and more toxic content than Democratic accounts, suggesting asymmetric blocking may reflect content-driven curation rather than pure partisan identity.

Method

Field Experiments

Two field experiments on Twitter (2020–2022) created human-looking bot accounts that varied partisan identification:

Field Experiment 1:

Created 10 bot accounts (5 Democrat-identified, 5 Republican-identified) with profile bios indicating partisan preference.
Each account retweeted content from partisan news outlets aligned with its indicated party (e.g., MSNBC, Washington Post for Democrat accounts; Fox News, NRO for Republican accounts) 3× per week.
Followed a politically balanced sample of Twitter users and measured their blocking/follow-back responses.
Outcome: Users 11× more likely to block counter-partisan bots vs. co-partisan bots.

Field Experiment 2:

Similar design but added a politically neutral control account (retweeting neutral outlets like Reuters, AP, BBC World).
Outcomes: Users 4.32× more likely to block counter-partisan vs. neutral accounts; no significant difference between co-partisan and neutral blocking.

Survey Experiments

Recruited Twitter users (N > 1,240 for main experiment) and showed them profiles of Democratic, Republican, or neutral accounts (no actual posted content shown, only follower counts and partisan cues).
Asked whether participants would follow-back, ignore, or block the account.
Survey results replicated field findings: 3× more likely to block counter-partisan than neutral profiles; partisan asymmetry smaller in surveys than field experiments, suggesting content differences matter.

Supplementary Analysis

Quantified content quality/slant of bot accounts: Republican accounts shared lower quality content (quality score 0.409 vs. 0.647), more politically slanted content (+0.696 vs. -0.210), and slightly more toxic language.
Survey probing blocking motivations (N=96 participants who chose to block) revealed: 27% cited "Never see their content," 19% "Likely a Bot," 19% "Never Engage," 10% "Disagree Politically."

Results

Field Experiment 1:

Users 11× more likely to block counter-partisan bot accounts compared to co-partisan accounts (b=−0.048, SE=0.010, p<.001).
Replication in Field Experiment 2: 4.32× more likely to block counter-partisan than neutral accounts.
Partisan asymmetry: Democratic users 26× more likely to block Republican accounts than Republican users to block Democratic accounts.

Field Experiment 2 (with neutral control):

Users 4.32× more likely to block counter-partisan accounts vs. neutral; no significant difference in blocking co-partisan vs. neutral accounts (p=.497).
Partisan asymmetry: Democratic users 5× more likely than Republicans to block counter-partisan accounts (interaction coefficient: b=−0.044, SE=0.008, p<.001).

Survey Experiment:

3× more likely to block counter-partisan profiles than neutral (b=0.089, SE=0.027, p<.001).
Smaller partisan asymmetry in survey vs. field (~5.5× vs. 26×), attributable to content differences.
Blocking motivation: "Never see their content" is the dominant reason (27%), significantly more common than random blocking (5.2%) or blocking out of partisan disagreement alone.

Content Analysis:

Republican bot accounts: lower quality (0.409), more slanted (+0.696), higher toxicity (0.0782).
Democratic bot accounts: higher quality (0.647), less slanted (−0.210), lower toxicity (0.0645).
This difference in content quality/slant correlates with observed asymmetry in blocking rates across parties.

Notes

Strengths: Large-scale field experiments on a real platform with real users provide strong causal evidence; replication across two field experiment designs and survey experiments builds confidence; novel focus on tie prevention as complement to tie formation; direct measurement of blocking motivations provides mechanism insight; content analysis explains partisan asymmetries, ruling out pure identity-driven explanations.

Limitations: Bot accounts portrayed as White, male profiles; prior work shows bot race/gender affects treatment effect. Field experiments conducted 2020–2022; Twitter's subsequent rebranding as X and policy changes (easier blocking UI, bot account moderation) may reduce generalizability. Survey did not assess long-term effects or downstream consequences of blocking on network structure or information exposure. Blocking motivation probed post-hoc; causal claim about informational vs. affective motivation relies on self-report. Does not examine blocking on other platforms (Facebook, Reddit, LinkedIn) where blocking mechanics and user populations differ.

Policy implications: Current Twitter/X interface makes blocking a high-friction action; field results suggest some users still actively block counter-partisans despite friction, indicating blocking is important enough to do despite interface friction. Easier blocking may increase political assortment. Conversely, findings that blocking is driven primarily by informational avoidance (not animosity) suggest interventions focused on content curation options may reduce blocking without addressing affective polarization directly. Platform researchers and designers should monitor blocking as a signal of user-driven assortment and consider how interface design (effort required, visibility, recommendations) shapes its prevalence.