CoChat
Sign Up
Cited by 4Hate Speech and Cyberbullying DetectionSocial Media and PoliticsMisinformation and Its Impacts

The Effects of Group Sanctions on Participation and Toxicity: Quasi-experimental Evidence from the Fediverse

Open access
Colglazier, Carl, TeBlunthuis, Nathan, Shaw, Aaron·2024·Proceedings of the International AAAI Conference on Web and Social Media
DOI 10.1609/icwsm.v18i1.31316
Open this paper in CoChat — free accountOpen PDF

Abstract

Online communities often overlap and coexist, despite incongruent norms and approaches to content moderation. When communities diverge, decentralized and federated communities may pursue group-level sanctions, including defederation (disconnection) to block communication between members of specific communities. We investigate the effects of defederation in the context of the Fediverse, a set of decentralized, interconnected social networks with independent governance. Mastodon and Pleroma, the most popular software powering the Fediverse, allow administrators on one server to defederate from another. We use a difference-in-differences approach and matched controls to estimate the effects of defederation events on participation and message toxicity among affected members of the blocked and blocking servers. We find that defederation causes a drop in activity for accounts on the blocked servers, but not on the blocking servers. Also, we find no evidence of an effect of defederation on message toxicity.

AI snapshot

Written from the paper’s full text.

TL;DR

This quasi-experimental study examines server-level defederation in the Fediverse, where Mastodon or Pleroma administrators block another server. It finds that defederation reduces posting activity for accounts on blocked servers but not for accounts on blocking servers, with no detectable change in post toxicity.

Key findings

  • Across 214 defederation events involving 275 servers and 661 accounts, blocked-server accounts showed a larger activity drop than matched controls: median decline 135.5 posts (21.9%) vs 18 posts (9.5%). The difference-in-differences estimate was β5 = -0.24, SE = 0.07; median activity fell from 24.9 to 19.6 posts per week (21.4% reduction), while matched controls averaged 22.8 posts per week after treatment.
  • Blocking-server accounts showed no measurable activity effect: difference-in-differences β5 = -0.08, SE = 0.08, with the 95% credible interval containing zero; the non-parametric change was 3.0 posts (p = 0.421).
  • Toxicity did not change detectably for either group. For blocked accounts, the median toxicity change was -0.005 (p = 0.072) relative to controls; difference-in-differences credible intervals for treatment effects contained zero for both blocked and blocking groups.
  • The final matched analysis included 258 blocked accounts from 71 home servers and 150 blocking accounts from 52 home servers, drawn from 7,445 defederation records and about 104 million public posts.

Methods

  • Observational quasi-experiment using public Mastodon and Pleroma API data from April 2, 2021 to May 31, 2022; defederation events were identified from public server block lists, with 91-day pre/post windows and weekly aggregation.
  • Outcomes were account posting activity (posts per week) and post toxicity from Perspective API scores (0–1). Accounts were included if they had prior cross-server replies and met activity thresholds; accounts exposed to multiple defederation dates were excluded.
  • Coarsened exact matching created control groups on pre-treatment posting, replies, server engagement, and server size; analyses combined visualization, Wilcoxon signed-rank tests, and difference-in-differences models (negative binomial for activity, beta regression for toxicity).

Limitations

  • The authors state that data cover only public posts, so private posts and direct messages are missing; some servers do not publish defederation records; and Perspective API accuracy may vary by language or context and may not capture Fediverse-specific harms.
  • The authors also note that matching assumes selection on observables, difference-in-differences assumes parallel trends and as-if random exposure, possible pre-announcements could affect behavior, and findings may not generalize beyond this context or longer time horizons.
  • Evident from the design: defederation events are not randomized, the sample is restricted to accounts with prior cross-server replies and sufficient activity, and the study excludes accounts exposed to multiple defederation dates, which may limit representativeness.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1609/icwsm.v18i1.31316

Cites

33 of 33 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

7 of 15 cited sources checked · 7 supported · 0 flagged

  • Premisecitation

    Prior empirical studies have found that group-level sanctions sometimes lead to increases in anti-social behavior or rebellious non-compliance.

    “Although sanctions are typically intended to promote normative compliance, group-level sanctions in some prior empirical studies have been met with increases in anti-social behavior or rebellious non-compliance (Mitts, Pisharody, and Shapiro 2022;Ribeiro et al. 2021;Russo et al. 2023).”
    Quote in the paperSubstantial claim3 of 3 sources checked
    • cites Mitts, Pisharody, and Shapiro 2022 source
    • cites Ribeiro et al. 2021 source
    • cites Russo et al. 2023 source
  • Interpretationcitation

    Defederation appears similar to individual and collective sanctions such as account banning, quarantine, and sub-community removal.

    “In this sense, defederation appears similar to individual and collective sanctions analyzed in prior studies such as account banning, quarantine, and sub-community removal (Chandrasekharan et al. 2022(Chandrasekharan et al. , 2017;;Jhaver et al. 2021).”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Chandrasekharan et al. 2022 source
    • cites Chandrasekharan et al. 2017 source
    • cites Jhaver et al. 2021 source
  • Findingown data

    Defederation has no detectable effect on post toxicity among affected accounts on either blocked or blocking servers.

    “Also, we find no evidence of an effect of defederation on message toxicity.”
    • Also stated in: § Introduction · plain
      “By contrast, we find that defederation has no effects on post toxicity on either the blocked or blocking servers.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Matched controls are assumed to be unaffected by defederation yet equally likely to have been exposed as treated accounts.
    • The authors state this as an assumption: The treatment and control groups' outcome trends were parallel prior to treatment.
    • The authors state this as an assumption: The trends are linear.
  • Findingown data

    For a median blocked-server account, activity dropped from 24.9 posts per week before treatment to 19.6 after (21.4% reduction), versus 22.8 for its matched control.

    “For a user account with median pre-treatment activity levels on the blocked server, this corresponds to a reduction from 24.9 posts per week immediately before treatment to 19.6 posts per week immediately after treatment (a 21.4% reduction), compared to 22.8 posts per week after treatment for its matched control.”
    Quote in the paperSubstantial claim
  • Findingown data

    No change in toxicity was observed among posts from blocking-server accounts.

    “We do not observe even such a small change in toxicity among posts from accounts on the blocking server.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Language model limitations, such as Perspective API errors, are time invariant at least in the short run, so they occur on both sides of a defederation event.
  • Classified as probably not a substantial claim. 10
  • May combine two claims. 17
  • The quote only partly states this claim. 2

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Prior deplatforming research has reported toxicity rises following certain deplatforming actions.

    “Some past studies of deplatforming have found that toxicity increases after some deplatforming events (Ali et al. 2021;Buntain et al. 2023;Chandrasekharan et al. 2017).”
    Quote in the paperNot a substantial claim2 of 3 sources checked
    • cites Ali et al. 2021 source
    • cites Buntain et al. 2023 source
    • cites Chandrasekharan et al. 2017 source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation

    Additional studies report that group-level sanctions either lower toxicity or leave it unchanged.

    “However, other studies have found that group-level sanctions decrease or do not change toxicity (Jhaver et al. 2021;Chandrasekharan et al. 2022).”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites Jhaver et al. 2021 source
    • cites Chandrasekharan et al. 2022 source

  • May combine two claims.

    Findingown data

    After defederation, activity levels of accounts on blocked servers diverged quickly from matched controls, although all groups reduced activity somewhat.

    “Although all groups reduced activity somewhat after defederation, Figure 6 shows the activity levels on the blocked servers quickly diverged from their matched controls after defederation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In non-parametric tests, median toxicity for posts from blocked-server accounts declined by 0.005, but the difference was not statistically distinguishable from null (p = 0.072).

    “For example, the median toxicity score for posts from these accounts declined by 0.005, but we lack sufficient statistical power to distinguish this from the null hypothesis that there was no difference in changes between treatment and control groups on the blocked server (p = 0.072).”
    Quote in the paperNot a substantial claim
    • The authors state this as an assumption: Language model limitations, such as Perspective API errors, are time invariant at least in the short run, so they occur on both sides of a defederation event.

  • May combine two claims.

    Findingown data

    Following defederation, users in the blocked group decreased their activity, while changes among users in the blocking group were minor or noisy.

    “Our visualization of the time series of affected accounts compared to matched controls, non-parametric tests, and difference-in-differences analysis all indicate that blocked group users decreased their activity following defederation while changes among blocking group users were minor or noisy.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Matching assumes that selection into treatment or control occurred due to observable variables incorporated into the matching process.

  • May combine two claims.

    Findingown data

    Difference-in-differences results show decreased activity among users on blocked servers and no change among users on blocking servers.

    “Our DiD results corroborate the same pattern: decreased activity among users on the blocked servers, and no change among users on the blocking servers.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For treated accounts on blocked servers, median post count was 18.5 one week before defederation and 13 one week after, compared to 20.5 both before and after for matched controls.

    “For instance, while the median post count for treated accounts on blocked servers (U 0 ) was 18.5 posts one week before defederation, this dropped to 13 posts one week after defederation.Compare this to the analogous median post count among matched controls which was 20.5 posts per week both before and after defederation.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Defederation reduces posting activity among affected accounts on blocked servers, while leaving activity among affected accounts on blocking servers unchanged.

    “We find that defederation causes a drop in activity for accounts on the blocked servers, but not on the blocking servers.”
    • Also stated in: § Introduction · plain
      “We find an asymmetric impact on posting activity: Accounts on blocked servers reduce their activity, but not accounts on blocking servers.”
    Quote in the paperSubstantial claim
    • The authors state this as an assumption: Matched controls are assumed to be unaffected by defederation yet equally likely to have been exposed as treated accounts.
    • The authors state this as an assumption: The treatment and control groups' outcome trends were parallel prior to treatment.
    • The authors state this as an assumption: The trends are linear.

  • May combine two claims. The quote only partly states this claim.

    Interpretationown data

    Defederation has mixed effectiveness as a group-level sanction: blocking communities do not lose activity, while blocked accounts reduce activity and their posts do not become less toxic.

    “These findings suggest that defederation, although a common group-level sanction on the Fediverse, has mixed effectiveness: Despite the risks of severing communication channels, communities implementing group-level sanctions do not lose activity.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Defederation has asymmetric effects on activity: activity decreases for affected accounts on blocked servers but not on blocking servers.

    “In sum, we observe asymmetric effects of defederation: activity decreases for affected accounts on blocked servers, but not on blocking servers.”
    Quote in the paperSubstantial claim

  • The quote only partly states this claim.

    Study design and conductown data

    Before defederation, trends and weekly median activity levels were comparable between treated and matched control accounts on both blocked and blocking servers.

    “Before defederation, the trends and weekly median activity levels remained comparable for accounts on both the blocked and blocking servers.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    Asymmetric effects on posting activity suggest administrators may find defederation an effective response to undesirable behavior on other servers without negative impacts on activity by affected user accounts on their own server.

    “These asymmetric effects of defederation on posting activity suggest that administrators may find defederation an effective response to undesirable behavior on other servers without negative impacts on activity by affected user accounts on their own server.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based). May combine two claims.

    Study design and conductnone given

    The defederation events in the study did not occur at random and were likely undertaken with awareness of the actors involved.

    “As noted above, the defederation events in our study did not occur at random and were likely undertaken with some awareness of the various actors involved.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductown data

    Matching yielded 258 blocked-account units from 397 candidates and 150 blocking-account units from 213 candidates.

    “The blocked accounts had 258 matched units from 397 potential units; the blocking accounts had 150 from 213 potential units.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Broadly, earlier research indicates that restricting particular contribution types from particular actor types produces immediate declines in the targeted behavior among those actors.

    “In general, this prior work finds that barriers to specific kinds of contributions from specific kinds of actors cause immediate decreases in the targeted activity among the targeted accounts.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Study design and conductreasoning

    Some servers choose not to publish defederation information, so effects emerging outside public posts could not be captured.

    “In addition, some servers choose not to publish defederation information.If the effects of defederation emerge anywhere other than public posts, our analysis could not capture them.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Some group-level sanctions have improved compliance with widely-held norms.

    “Some of these group-level sanctions have even improved compliance with widely-held norms (Chandrasekharan et al. 2017).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Chandrasekharan et al. 2017 source

  • Classified as probably not a substantial claim. May combine two claims.

    Premisereasoning

    When one Fediverse server blocks another, the blocked server stops propagating messages to the blocking server, reducing the audience reachable from the blocked server and potentially decreasing the utility of posting there.

    “When one Fediverse server blocks another, the blocked server no long propagates messages to the blocking server.This reduces the size of the audience that can be reached from the blocked server and thereby may decrease the utility of posting on that server.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductown data

    The final dataset includes 258 blocked accounts from 71 home servers and 150 blocking accounts from 52 home servers, across 214 defederation events.

    “After dropping some additional accounts in the matching process (described below), the resulting dataset includes 258 accounts whose 71 home servers were defederated and 150 accounts whose 52 home servers defederated another server.This included a total of 214 defederation events.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisereasoning

    Decentralized group-level sanctions differ from central platform sanctions because the targeted sub-community remains part of the larger network.

    “These decentralized group-level sanctions are distinct in that the targeted sub-community remains part of the larger network.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The results indicate that defederation events produce asymmetric effects on activity for affected accounts on blocked versus blocking servers, with no increase in toxicity for any groups.

    “The results indicate that such events produce asymmetric effects on activity for affected accounts on blocked servers versus those on blocking servers with no increase in toxicity for any groups.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Premisenone given

    In a decentralized system, defederation is among the limited instruments administrators possess to guard against bad actors or enforce external norms.

    “Defederation is one of the few tools administrators in a decentralized system have to protect against bad actors or enforce norms from beyond their own servers.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductcitation

    The Perspective API's accuracy may vary across languages and contexts, and its errors may affect statistical results.

    “While this tool provides a useful proxy, its accuracy may vary across languages and contexts and its errors may affect our statistical results (TeBlunthuis, Hase, and Chan 2023).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites TeBlunthuis, Hase, and Chan 2023

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationown data

    The results highlight the potential of decentralized social networks and mechanisms like defederation in providing communities with tools to manage content moderation and other aspects of online interactions.

    “The results also highlight the potential of decentralized social networks and their unique mechanisms, such as defederation, in providing communities with tools to manage content moderation and other aspects of online interactions.”
    Quote in the paperNot a substantial claim
24 of 24

  • The paper relies on this without support.

    The effects of decentralized group-level sanctions such as defederation have not been empirically studied before.

    “To our knowledge, the effects of such sanctions remain unexplored empirically.”

  • The paper relies on this without support.

    The bulk of earlier research on group-level sanctions examines actions taken by central platform actors, including commercial social media staff.

    “Most prior work on group-level sanctions focuses on sanctions applied by central actors such as commercial social media platform staff.”

  • The authors state this as an assumption.

    Matched controls are assumed to be unaffected by defederation yet equally likely to have been exposed as treated accounts.

    “The matching procedure should result in control groups of accounts that were not affected by defederation, but were just as likely to have been affected as the accounts that were”

  • The authors state this as an assumption.

    The treatment and control groups' outcome trends were parallel prior to treatment.

    “the two groups' trends in the outcome variables should be parallel prior to treatment”

  • The authors state this as an assumption.

    The trends are linear.

    “and the trends should be linear.”

  • The paper relies on this without support.

    The date of each defederation event can be accurately inferred from the day a server first appears in the public block list.

    “We check these records each day and infer the date that defederation occurred by the day a server first appears in the list.”

  • The paper relies on this without support.

    The block lists that servers make public are assumed to be complete and accurate.

    “We identify defederation from the public records servers publish of which other servers they block.”

  • The paper relies on this without support.

    The observable pre-treatment attributes used for matching account for potential confounders.

    “we use observable, pre-treatment attributes of accounts that capture how they engage with the wider Fediverse through posting replies to external servers.”

  • The paper relies on this without support.

    The study provides the first evidence concerning the impacts of defederation in Fediverse communication networks.

    “Our results provide the first evidence concerning the impacts of defederation in Fediverse communication networks.”

  • The authors state this as an assumption.

    Matching assumes that selection into treatment or control occurred due to observable variables incorporated into the matching process.

    “Matching entails an assumption that selection into treatment or control occurred due to observable variables incorporated into the matching process.”

  • The authors state this as an assumption.

    Language model limitations, such as Perspective API errors, are time invariant at least in the short run, so they occur on both sides of a defederation event.

    “Although language models like those utilized in the Perspective API have significant limitations, we believe that these are likely to be time invariant, at least in the short-run.That is, they would happen on both sides of a defederation event.”
11 of 11

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • In ActivityPub, independent Server administrators act as moderators, with significant discretion over local content policies and inter-server relationships.

    “In ActivityPub, independent Server administrators act as moderators, with significant discretion over local content policies and inter-server relationships [27,82].”

    From Seeing the Politics of Decentralized Social Media Protocols (2026) · cites this paper as [27]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • The activity model mislabels a group trend as post-treatment only

    Minor

    The model defines β6 as a group-by-time term with no post-treatment indicator, so it applies before as well as after defederation. The accompanying description instead lists β6 among the post-treatment terms.

  • Conflicting descriptions of the toxicity test hypothesis

    Minor

    The methods identify zero change as the null hypothesis, but Table 4's caption calls zero change the alternative hypothesis.

2 of 2

Cites

Flagged first, then by citations

Cited by

Sorted by citations

Related

Authors
Colglazier, C., TeBlunthuis, N., & Shaw, A.
Affiliations
  • Northwestern University (US)
  • University of Michigan (US)
ORCID
Venue
Proceedings of the International AAAI Conference on Web and Social Media
Publisher
Association for the Advancement of Artificial Intelligence (AAAI)
Published
2024-05-28
Volume
18
Pages
315-328
ISSN
  • 2162-3449 (ISSN-L)
  • 2334-0770
DOI
10.1609/icwsm.v18i1.31316
OpenAlex
W4399205583
Cited by
3
References
27
Citations by year
2026: 1 · 2025: 2
FWCI
0.97
Type
conference paper
Language
en
Open access
Locations
Proceedings of the International AAAI Conference on Web and Social Media (publishedVersion)
Topics
  • Physical Sciences › Computer Science › Artificial Intelligence › Hate Speech and Cyberbullying Detection
  • Social Sciences › Communication › Social Media and Politics
  • Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
Keywords
Fediverse, decentralized social networks, Mastodon, Pleroma, online community participation, content moderation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-06.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-06.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review