Pith. sign in

REVIEW 3 major objections 6 minor 142 references

Changes to the Facebook Algorithm Decreased News Visibility Between 2021-2024

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Facebook's algorithm changes cut reactions to news posts by 78 percent between 2021 and 2024.

desk verdict A large, carefully analyzed observational dataset showing a striking decline in reactions to news on Facebook after 2021, with a plausible but not airtight causal claim. read the letter →

arxiv 2507.19373 v1 pith:THIJLOKM submitted 2025-07-25 cs.SI

Szymon Talaga , Erin Wertz , Dominik Batorski , Magdalena Wojcieszak This is my paper · ORCID
classification cs.SI
keywords Facebookalgorithmicnewsfeedvisibilityuserreactionsdifference-in-differencechangepointdetectionqualityplatformgovernance
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper tries to establish that Facebook's own algorithmic choices, not outside forces, caused a sharp drop in how often Americans engaged with news on the platform between 2021 and 2024. Drawing on 5.2 million posts from 40 U.S. news outlets and nearly 8 billion reactions, it reports that average reactions per news post fell by roughly 78 percent while reactions to posts from sports, retail, and entertainment pages rose by a comparable amount. It argues the decline cannot be explained by news outlets posting less, by users leaving Facebook, or by falling interest in news, and that the January 2025 reversal of the policy produced an immediate rebound. If right, the finding shows that a platform's feed design can quietly redraw the boundaries of public attention to news.

What carries the argument

The load-bearing object is the difference-in-difference contrast $\tau(t) = \psi(t|\text{news})/\psi(t|\text{non-news})$, a ratio of ratios that uses engagement with 21 non-news pages as the counterfactual for what news engagement would have been without algorithmic intervention. Around it sits a pipeline: user reactions validated as a proxy for visibility (correlation with views about 0.79), a Bayesian changepoint detector that splits the decade into 12 epochs, and negative binomial mixed-effects models that estimate per-post reaction counts while absorbing outlet-level and daily variation.

What would settle it

Look for an internal Facebook metric of news feed impressions or reach: if impressions of news posts stayed flat or rose while reactions fell by 78 percent, the 'decreased visibility' claim would fail because the decline would be in engagement propensity rather than exposure. Conversely, if reach fell with reactions, the claim is supported. A second check would be to match each news outlet to non-news pages by post type and posting volume; if the divergence disappears or shrinks sharply, the counterfactual comparison was not valid.

Watch

Extended reading notes

Core claim

On the paper's own terms, the discovery is that Facebook's 'War on News' worked as announced: after the February 2021 announcement that civic and political content would be deprioritized, user reactions to news posts entered a steady decline, reaching a minimum of about 526 reactions per post in mid-2023 to early 2024 versus peaks above 3,000 during election periods, a cumulative drop of about 78 percent. The decline was largest for low-quality outlets (about 86 percent), which had historically drawn the most reactions, while all quality tiers converged to similarly depressed levels. Crucially, 21 large non-news pages showed no such decline; their reactions rose by about 78 percent in the same interval, and a difference-in-difference contrast finds the two statistically significant epoch-to-epoch drops (2022 and 2023) were specific to news. The paper also reports a partial rebound after January 2025, concentrated among low- and medium-quality outlets, and rules out shrinking platform use, falling news supply, and reduced overall news interest as explanations.

Load-bearing premise

The whole causal story rests on the assumption that, absent algorithmic changes, news and non-news pages would have kept parallel reaction trends; the pre-2021 test supports this but uses only 5 epochs and 21 non-news pages, so it cannot rule out that the two groups were headed for different trajectories anyway.

Editorial extensions

If this is right

  • If the central claim is right, Facebook's 2021-2024 feed changes reduced on-platform reactions to news by about 78 percent while leaving non-news engagement intact, implying the effect was content-specific rather than a general engagement slump.
  • Low-quality outlets, the original target of the policies, were hit hardest (about 86 percent), meaning the algorithm acted as a quality-filtering mechanism during that period.
  • The January 2025 reversal caused an immediate increase in reactions to news, with the rebound concentrated in low- and medium-quality outlets, suggesting the earlier decline was reversible rather than a structural loss of audience.
  • The absence of a parallel decline in off-platform traffic and in news interest surveys implies that on-platform engagement metrics, not actual news consumption, were the thing the algorithm moved.
  • Election spikes in 2024 were markedly weaker than in 2016 and 2020, so even peak-interest events could not overcome the algorithmic suppression.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: if the parallel-trends comparison is trusted, the 78 percent rise in non-news reactions implies that attention freed from news was partly reallocated to entertainment and retail content, not just lost; that reallocation is the paper's implicit mechanism but is not directly measured.
  • Beyond the paper: a natural extension would test the same design on other content types the policy targeted, such as posts by politicians and parties, which the paper notes it did not include; finding a similar divergence there would widen the claim beyond news organizations.
  • Beyond the paper: the analysis cannot distinguish which of the roughly 399 announced changes did the work; a stricter test would exploit staggered rollout dates or platform-internal impression logs to isolate each intervention.
  • Beyond the paper: because reactions track views only partially (r ≈ 0.79), internal impression data from the platform could either confirm or revise the magnitude; the paper's proxy assumption is testable in principle.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper analyzes 5.2 million Facebook posts from 40 U.S. news outlets and 396,468 posts from 21 non-news pages between January 2016 and February 2025, using reaction counts as a proxy for news visibility. It reports that reactions to news posts declined by about 78% between 2021 and 2024, while reactions to non-news posts increased by a similar amount, and that low-quality outlets were especially suppressed. The authors use Bayesian changepoint detection (BEAST) to define epochs, negative binomial mixed-effects models to estimate epoch-level reaction means, and a difference-in-difference comparison against non-news pages to support a causal interpretation that Meta's algorithmic 'War on News' drove the decline. They also present robustness checks against alternative explanations including user-base changes, general news-interest trends, and off-platform traffic.

Significance. If the causal claim holds, this is an important contribution: it provides large-scale, longitudinal, on-platform evidence that a major platform's algorithmic choices substantially reduced public exposure to news, with differential effects on low-quality outlets and a reversal when the policy ended. The paper is unusually strong in descriptive scope, with nearly a decade of posting and reaction data, a plausible control series, external quality ratings anchored outside the authors' own work, and a policy reversal that fits the intervention hypothesis. The changepoint detection is data-driven rather than constructed from announcement dates, which mitigates circularity. The central weakness is that two load-bearing assumptions—that non-news pages are a valid counterfactual and that reactions track visibility stably over time—are not directly tested at the temporal resolution required for the causal claim.

major comments (3)
  1. [Sec. 4.4, Eq. (12)] The pre-period parallel-trends test (chi-square(4)=3.88, p=0.423) is based on only five epochs and a control group of 21 highly heterogeneous pages (Netflix, NBA, Walmart, Chipotle, etc.). This test has low power, and it cannot rule out divergent trends driven by content-type shifts such as the growth of video and entertainment or by platform changes that disproportionately favored non-news content. Since the non-news series increased by roughly 78% over the same period in which news declined, the difference-in-difference estimate in Eq. (1) may partly reflect a relative shift toward entertainment content rather than targeted suppression of news. Please add a direct test using a control set restricted to non-entertainment pages, or a model with post-type-by-time interactions, and report the parallel-trends test at higher temporal resolution.
  2. [Appendix B.2] The validation of reactions as a visibility proxy is cross-sectional: the correlation between views and reactions (r=0.791) is computed across posts at a single point in time. The paper's central claim is about changes in visibility over time, and the reaction-to-view ratio may itself have changed as the algorithm altered which users see news posts (e.g., leaving only highly engaged loyal followers). If the ratio is not stable over time, a 78% decline in reactions could overstate or understate the true visibility decline. Please estimate the reactions-per-view ratio by epoch for the Content Library subset and show that it is stable, or qualify the visibility claims accordingly.
  3. [Sec. 2.2, Fig. 2] The assignment of detected changepoints to causes is partly ambiguous, which is acknowledged in Sec. 4.4. For example, changepoint 5 at 2021-03-01 is matched to both the Biden inauguration and the February 10, 2021 algorithmic announcement, and changepoint 8 has no identified cause. Because the causal claim depends on the alignment of epochs with algorithmic interventions, please provide a formal sensitivity analysis—such as placebo changepoints or a comparison of changepoint density around algorithmic versus political events—to demonstrate that the matching is not post-hoc selection.
minor comments (6)
  1. [Sec. 3] In the Discussion, 'we identifying the feed algorithm changes' should read 'we identify the feed algorithm changes'.
  2. [Sec. 3] In the Discussion, 'making them more susceptibility to misinformation' should read 'making them more susceptible to misinformation'.
  3. [Appendix F.1] The word 'algoirthmically' is a typo and should be 'algorithmically'.
  4. [Table A.1] The header 'Ad Fonts Bias' should be 'Ad Fontes Bias' for consistency with the cited rating system.
  5. [Sec. 4.1 and Appendix E] The main text says reaction counts were missing for approximately 10% of posts in the Content Library data, while Appendix E says the preliminary model was used for imputing about 1.1% of missing reaction counts; please clarify that these percentages refer to different denominators.
  6. [Fig. 2b] Changepoint 8 (2022-10-31) is listed without an event description; consider adding a brief note that no proximal algorithmic event was identified.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: changepoints are detected from the outcome data without using announcement dates as inputs, the non-news counterfactual is an external control, and reactions-as-visibility is validated against view counts.

full rationale

The paper's derivation chain is self-contained. Epochs are identified by BEAST applied to the weekly news-reaction time series with no announcement dates fed into the detector (Sec. 4.2), so the subsequent qualitative alignment of changepoints with Meta's publicly announced changes is a post-hoc match rather than a fitted prediction. The 78% decline is estimated from observed reaction counts through a mixed-effects negative binomial model (Sec. 4.3, Appendix G), and the causal interpretation uses non-news pages as an external counterfactual in a difference-in-difference contrast (Eq. 1, Sec. 4.4), not as a quantity derived from the news outcome. The reactions-as-visibility proxy is validated against Content Library view counts (Appendix B.2, r = 0.791), and alternative explanations are checked with external comScore, user-base, and survey data (Appendix J). The manuscript itself flags its limits: the correlational design, the inability to pinpoint causes for every changepoint (Appendix F.1, changepoint 8), and the proxy nature of reactions (Sec. 3). These are statistical-validity and interpretation concerns, not circular reductions. No load-bearing argument reduces to a self-citation or to a parameter fitted to the target claim.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim does not introduce theoretical free parameters or new entities; the listed free parameters are analytical choices and fitted statistical coefficients. The key assumptions are empirical: reactions proxy visibility, non-news pages identify the counterfactual, and external ratings classify quality. No new forces, particles, or mechanisms are postulated.

free parameters (4)
  • Quality tier percentile cutoffs = 33rd and 67th percentiles of Lin et al. quality scores
    Divided 40 outlets into low, medium, and high quality groups; this choice affects all differential-suppression conclusions.
  • BEAST changepoint hyperparameters = max changepoints=30, min separation=13 weeks, peak height pmin=0.5, smoothing window l=2, runs=1000
    Chosen by the authors; these determine the epoch boundaries that define the 78% decline and all epoch contrasts.
  • Reaction imputation regression coefficients = Adjusted R2=0.6827
    Fitted on non-missing posts to impute roughly 10% of reaction counts from views, outlet identity, and video type; if biased, this propagates to all downstream estimates.
  • Negative binomial model coefficients and variance components = Estimated via ML in glmmTMB
    Fixed effects, random effects, and dispersion parameters are fitted to the data and used to derive marginal means and causal contrasts.
assumptions (5)
  • domain assumption Reactions are a valid proxy for post visibility
    Validated by a view-reaction correlation of r=0.791 and R2=0.626 in a subset with view data (Appendix B.2), but this leaves substantial unexplained variance in views.
  • domain assumption Non-news pages provide a valid counterfactual under parallel trends
    Difference-in-differences in Sec. 4.4 assumes news and non-news reaction trends would be equiproportional absent the algorithmic changes; the pre-period test cannot verify this for the post-intervention period.
  • domain assumption External quality ratings classify outlets reliably
    Quality tiers are taken from Lin et al. (24), which aggregates external raters; errors in these ratings propagate to low-versus-high suppression conclusions.
  • domain assumption Data collection is complete and missing reactions are ignorable
    The analysis assumes Sotrender and Content Library capture all posts and that missing reactions can be imputed using views, outlet, and video type without systematic bias.
  • standard math Asymptotic normality of MLE and simultaneous inference
    Confidence intervals and p-values rely on asymptotic Gaussian approximations for GLMM estimators and joint contrasts, as is standard but not exact in finite samples.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Changes to the Facebook Algorithm Decreased News Visibility Between 2021-2024." pith.science (2026). https://pith.science/paper/THIJLOKM

@misc{pith2026250719373,
  author       = {Pith},
  title        = {Pith review of: Changes to the Facebook Algorithm Decreased News Visibility Between 2021-2024},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/THIJLOKM}},
  note         = {Machine review of arXiv:2507.19373}
}
read the original abstract

Platforms, especially Facebook, are primary news sources in the US. In its widely criticized "War on News," Meta algorithmically deprioritized news and political content. We use data from 40 news organizations (5,243,302 Facebook posts, 7,875,372,958 user reactions) and 21 non-news pages (396,468 posts; 1,909,088,308 reactions) between January 1, 2016 and February 13, 2025 to examine how these changes influenced news visibility on the platform. Reactions to news declined by 78% between 2021 and 2024 while reactions to non-news pages increased, indicating targeted suppression of news visibility. Low-quality sources were especially suppressed, yet the 2025 end to "War on News" increased user reactions to news, especially low-quality ones. These changes do not reflect decreased news supply, Facebook user base, or interest in news over this period.

Figures

Figures reproduced from arXiv: 2507.19373 by the authors.

Figure 1
Figure 1. Time series of weekly post and average reaction counts across news quality tiers and non-news posts smoothed using 4-week rolling mean. Vertical bars highlighted in yellow mark the U.S. presidential elections. Thin vertical bars with annotations correspond to other important events for aiding the interpretation. 0.985, and lack of correlation between post counts and reaction counts in all quality tiers as well as no… view at source ↗
Figure 2
Figure 2. Detected changepoints. a Two components of the signal used for changepoint detection (weekly-averaged relative expectations and coefficients of variation). Gray vertical lines denote point estimates for locations while gray bounds correspond to interval estimates. b Point and interval estimates for changepoint locations. The ’Event’ and subsequent columns catalogue and briefly describe the most likely algorithmic ch… view at source ↗
Figure 3
Figure 3. Dynamics of reactions in time and throughout the detected changepoints and epochs (intervals between changepoints). a Es￾timated average numbers of reactions per post by epochs and news outlet quality tiers. Round points represent outlet means and their sizes are proportional to the square root of the number of posts in a given epoch. Horizontal arrows denote significant sequential changes and vertical arrows signif… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Comparison of reaction levels in news and non-news data. a Estimated marginal means in epochs for news and non-news posts (news means were derived by averaging over the quality tiers) with outlet-epoch averages denoted by dots. Up/down arrows and p-values at the top co…

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

142 extracted references · 70 canonical work pages

  1. [1]

    Statista, Topic: Social media usage in the United States, https://www.statista.com/topics/3196/social-media-usage-in-the-united- states/ (2024)

    2024

  2. [2]

    URL https://www.pewresearch.org/journalism/fact-sheet/ social-media-and-news-fact-sheet/

    Pew Research Center, Social media and news fact sheet (2024). URL https://www.pewresearch.org/journalism/fact-sheet/ social-media-and-news-fact-sheet/

    2024

  3. [3]

    Caplan, D

    R. Caplan, D. Boyd, Isomorphism through algorithms: Institutional de- pendencies in the case of Facebook, Big Data & Society 5 (1) (2018) 2053951718757253. doi:10.1177/2053951718757253

    doi:10.1177/2053951718757253 2018

  4. [4]

    Barber ´a, A

    P. Barber ´a, A. Casas, J. Nagler, P. J. Egan, R. Bonneau, J. T. Jost, J. A. Tucker, Who leads? who follows? measuring issue attention and agenda setting by legislators and the mass public using social media data, Amer- ican Political Science Review 113 (4) (2019) 883–901

    2019

  5. [5]

    Newman, R

    N. Newman, R. Fletcher, C. Robertson, A. Arguedas, R. K. Nielsen, Reuters institute Digital News Report 2024 (2024). URL https://reutersinstitute.politics.ox.ac.uk/sites/ default/files/2024-06/RISJ_DNR_2024_Digital_v10%20lr. pdf

    2024

  6. [6]

    URL https://www.pewresearch.org/journalism/2024/06/12/ how-americans-get-news-on-tiktok-x-facebook-and-instagram/

    Pew Research Center, How americans get news on TikTok, x, facebook and instagram (2024). URL https://www.pewresearch.org/journalism/2024/06/12/ how-americans-get-news-on-tiktok-x-facebook-and-instagram/

    2024

  7. [7]

    Kaplan, More Speech and Fewer Mistakes (Jan

    J. Kaplan, More Speech and Fewer Mistakes (Jan. 2025). URL https://about.fb.com/news/2025/01/ meta-more-speech-fewer-mistakes/

    2025

  8. [8]

    Mosseri, Bringing people closer together | meta (2018)

    A. Mosseri, Bringing people closer together | meta (2018). URL https://about.fb.com/news/2018/01/ news-feed-fyi-bringing-people-closer-together/

    2018

Show all 142 references
  1. [9]

    Stepanov, Reducing political content in news feed (2021)

    A. Stepanov, Reducing political content in news feed (2021). URL https://about.fb.com/news/2021/02/ reducing-political-content-in-news-feed/

    2021

  2. [10]

    Bailo, J

    F. Bailo, J. Meese, E. Hurcombe, The institutional impacts of algorithmic distribution: Facebook and the australian news media, Social Media + Society 7 (2) (2021) 20563051211024963, publisher: SAGE Publications Ltd. doi:10.1177/20563051211024963. URL https://doi.org/10.1177/2...

    2021 doi

  3. [11]

    McNally, M

    N. McNally, M. Bastos, The news feed is not a black box: A longitudinal study of facebook’s algorithmic treatment of news, Digital Journalism (2025) 1–20

    2025

  4. [12]

    Horwitz, K

    J. Horwitz, K. Hagey, E. Glazer, Facebook Wanted Out of Politics. It Was Messier Than Anyone Expected., Wall Street Journal (Jan. 2023). URL https://www.wsj.com/articles/ facebook-politics-controls-zuckerberg-meta-11672929976

    2023

  5. [13]

    G. A. Fowler, Don’t say ‘vote’: How Instagram hides your political posts, The Washington Post (Oct. 2024). URL https://www.washingtonpost.com/technology/2024/10/ 16/instagram-limits-political-content-shadowban-election-posts/

    2024

  6. [14]

    M. X. Delli Carpini, In search of the informed citizen: What Americans know about politics and why it matters, The Communication Review 4 (1) (2000) 129–164

    2000

  7. [15]

    Prior, News vs

    M. Prior, News vs. entertainment: How increasing media choice widens gaps in political knowledge and turnout, American Journal of Political Science 49 (3) (2005) 577–592

    2005

  8. [16]

    Altay, R

    S. Altay, R. K. Nielsen, R. Fletcher, News can help! the impact of news media and digital platforms on awareness of and belief in misinformation, The International Journal of Press/Politics 29 (2) (2024) 459–484

    2024

  9. [17]

    Mont’Alverne, A

    C. Mont’Alverne, A. Ross Arguedas, S. Banerjee, B. To ff, R. Fletcher, R. Nielsen, The electoral misinformation nexus: How news consumption, platform use, and trust in news influence belief in electoral misinforma- tion, Public Opinion Quarterly (2024). doi:10.1093/poq/nfae019

    2024 doi

  10. [18]

    Altay, E

    S. Altay, E. Hoes, M. Wojcieszak, Following News on Social Media Boosts Knowledge, Belief Accuracy, and Trust, Nature Human Behaviour (2025) Forthcomingdoi:10.31234/osf.io/hq5ru_v2

    2025 doi

  11. [19]

    Humprecht, F

    E. Humprecht, F. Esser, P. Van Aelst, Resilience to online disinformation: A framework for cross-national comparative research, The international journal of press/politics 25 (3) (2020) 493–516

    2020

  12. [20]

    Taylor, Meta says facebook cannot solve media industry’s ‘issues’ as it defends ending payments for news in australia, The Guardian (2024)

    J. Taylor, Meta says facebook cannot solve media industry’s ‘issues’ as it defends ending payments for news in australia, The Guardian (2024). URL https://www.theguardian.com/media/2024/mar/14/ meta-facebook-news-media-bargaining-code

    2024

  13. [21]

    Evershed, J

    N. Evershed, J. Taylor, ‘news on facebook is dead’: memes replace aus- tralian media posts as meta turns off the tap, The Guardian (2024). URL https://shorturl.at/dAmXG

    2024

  14. [22]

    Taylor, Meta claims news is not an antidote to misinformation on its platforms, The Guardian (2024)

    J. Taylor, Meta claims news is not an antidote to misinformation on its platforms, The Guardian (2024). URL https://www.theguardian.com/media/article/2024/ jul/09/meta-facebook-australia-news-ban-misinformation

    2024

  15. [23]

    Bruns, D

    A. Bruns, D. Angus, Facebook’s Australian news ban: threat, impact, and aftermath, in: Selected Papers in Internet Research 2021: Research from the Annual Conference of the Association of Internet Researchers, AoIR - Association of Internet Researchers, United States of Americ...

    2021

  16. [25]

    J. Roth, P. H. Sant’Anna, A. Bilinski, J. Poe, What’s trending in difference-in-differences? A synthesis of the recent econometrics litera- ture, Journal of Econometrics 235 (2) (2023) 2218–2244.doi:10.1016/ j.jeconom.2023.03.008

    2023

  17. [26]

    Cornia, A

    A. Cornia, A. Sehl, D. A. L. Levy, R. K. Nielsen, Private sector news, social media distribution, and algorithm change (2018)

    2018

  18. [27]

    Reuning, A

    K. Reuning, A. Whitesell, A. L. Hannah, Facebook algorithm changes may have amplified local republican parties, Research & Politics 9 (2) (2022). URL https://journals.sagepub.com/doi/full/10.1177/ 20531680221103809

    2022

  19. [28]

    Bandy, N

    J. Bandy, N. Diakopoulos, Facebook’s news feed algorithm and the 2020 US election, Social Media + Society 9 (3) (2023) 20563051231196898, publisher: SAGE Publications Ltd. doi:10. 1177/20563051231196898. URL https://doi.org/10.1177/20563051231196898

    2023 doi

  20. [29]

    D. A. Broniatowski, J. R. Simons, J. Gu, A. M. Jamison, L. C. Abroms, The efficacy of facebook’s vaccine misinformation policies and architec- ture during the COVID-19 pandemic, Science Advances 9 (37) (2023) eadh2132, publisher: American Association for the Advancement of Sci...

    2023 doi

  21. [30]

    Meese, E

    J. Meese, E. Hurcombe, Facebook, news media and platform de- pendency: The institutional impacts of news distribution on so- cial platforms, New Media & Society 23 (8) (2021) 2367–2384. doi:10.1177/1461444820926472. URL https://journals.sagepub.com/doi/10.1177/ 1461444820926472

    2021 doi

  22. [31]

    Lazer, R

    D. Lazer, R. Kennedy, G. King, A. Vespignani, The parable of google flu: Traps in big data analysis, Science 343 (6176) (2014) 1203–1205. doi:10.1126/science.1248506. URL https://www.science.org/doi/10.1126/science. 1248506

    2014 doi

  23. [32]

    Schi ff, An Update on the Georgia Runoff Elections (Dec

    S. Schi ff, An Update on the Georgia Runoff Elections (Dec. 2020). URL https://about.fb.com/news/2020/12/ update-on-the-georgia-runoff-elections/

    2020

  24. [33]

    Jin, Keeping People Safe and Informed About the Coronavirus (Dec

    K.-X. Jin, Keeping People Safe and Informed About the Coronavirus (Dec. 2020). URL https://about.fb.com/news/2020/12/coronavirus/ 13

    2020

  25. [34]

    URL https://about.fb.com/news/2023/06/ changes-to-news-availability-on-our-platforms-in-canada/

    Meta, Changes to news availability on our platforms in canada (2023). URL https://about.fb.com/news/2023/06/ changes-to-news-availability-on-our-platforms-in-canada/

    2023

  26. [35]

    M. M. Waldrop, The genuine problem of fake news, Proceedings of the National Academy of Sciences 114 (48) (2017) 12631–12634

    2017

  27. [36]

    URL https://www.weforum.org/publications/ global-risks-report-2024/digest/

    World Economic Formul, Global Risks Report 2024 (2024). URL https://www.weforum.org/publications/ global-risks-report-2024/digest/

    2024

  28. [37]

    Guess, J

    A. Guess, J. Nagler, J. Tucker, Less than you think: Prevalence and pre- dictors of fake news dissemination on facebook, Science Advances 5 (1) (2019) eaau4586. doi:10.1126/sciadv.aau4586. URL https://www.science.org/doi/10.1126/sciadv.aau4586

    2019 doi

  29. [38]

    Grinberg, K

    N. Grinberg, K. Joseph, L. Friedland, B. Swire-Thompson, D. Lazer, Fake news on twitter during the 2016 us presidential election, Science 363 (6425) (2019) 374–378

    2019

  30. [39]

    Peysakhovich, K

    A. Peysakhovich, K. Hendrix, Further reducing clickbait in feed (2016). URL https://about.fb.com/news/2016/08/ news-feed-fyi-further-reducing-clickbait-in-feed/

    2016

  31. [40]

    URL https://about.fb.com/news/2020/04/ covid-19-misinfo-update/

    Rosen, An update on our work to keep people informed and limit misinformation about COVID-19 (2020). URL https://about.fb.com/news/2020/04/ covid-19-misinfo-update/

    2020

  32. [41]

    URL https://about.fb.com/news/2020/10/ preparing-for-election-day/

    Rosen, Preparing for election day (2020). URL https://about.fb.com/news/2020/10/ preparing-for-election-day/

    2020

  33. [42]

    URL https://about.fb.com/news/2019/04/ protecting-eu-elections-from-misinformation/

    Woodford, Protecting the EU elections from misinformation and expand- ing our fact-checking program to new languages (2019). URL https://about.fb.com/news/2019/04/ protecting-eu-elections-from-misinformation/

    2019

  34. [43]

    comScore, comScore Media Metrix Description of Methodology (2013)

    2013

  35. [44]

    E. L. Lehmann, G. Casella, Theory of Point Estimation, 2nd Edition, Springer Texts in Statistics, Springer, New York, 1998

    1998

  36. [45]

    Hothorn, F

    T. Hothorn, F. Bretz, P. Westfall, Simultaneous Inference in General Parametric Models, Biometrical Journal 50 (3) (2008) 346–363. doi: 10.1002/bimj.200810425

    2008 doi

  37. [46]

    K. Zhao, M. A. Wulder, T. Hu, R. Bright, Q. Wu, H. Qin, Y . Li, E. Toman, B. Mallick, X. Zhang, M. Brown, Detecting change-point, trend, and sea- sonality in satellite time series data to track abrupt changes and nonlinear dynamics: A Bayesian ensemble algorithm, Remote Sensin...

    2019 doi

  38. [47]

    Gelman, J

    A. Gelman, J. Hill, Data Analysis Using Regression and Multi- level/Hierarchical Models, 23rd Edition, Analytical Methods for Social Research, Cambridge Univ. Press, Cambridge, 2021

    2021

  39. [48]

    Haugsgjerd, R

    A. Haugsgjerd, R. Karlsen, Election campaigns, news consumption gaps, and social media: equalizing political news use when it matters?, The International Journal of Press/Politics 29 (2) (2024) 507–529

    2024

  40. [49]

    Newman, R

    N. Newman, R. Fletcher, A. Schulz, S. Andı, C. T. Robertson, R. K. Nielsen, The Reuters Institute digital news report 2021 (2021). doi:10.60625/RISJ-7KHR-ZJ06 . URL https://reutersinstitute.politics.ox.ac.uk/ digital-news-report/2021

    2021 doi

  41. [50]

    Newman, A

    N. Newman, A. R. Arguedas, C. T. Robertson, R. K. Nielsen, R. Fletcher, The Reuters Institute digital news report (2025)

    2025

  42. [51]

    URL https://www.pewresearch.org/internet/fact-sheet/ social-media/

    Pew Research Center, Social media fact sheet (2024). URL https://www.pewresearch.org/internet/fact-sheet/ social-media/

    2024

  43. [52]

    Knight Foundation, How media habits relate to voter participation for younger adults (25-29), avail- able at https://knightfoundation.org/reports/ how-media-habits-relate-to-voter-participation/ (2020)

    2020

  44. [53]

    Majid, Search vs social: How referral tra ffic to news sites has changed in five years (Apr

    A. Majid, Search vs social: How referral tra ffic to news sites has changed in five years (Apr. 2023). URL https://pressgazette.co.uk/ media-audience-and-business-data/media_metrics/ news-referral-traffic-breakdown/

    2023

  45. [54]

    Majid, Facebook’s referral tra ffic for publishers down 50% in 12 months (2024)

    A. Majid, Facebook’s referral tra ffic for publishers down 50% in 12 months (2024). URL https://pressgazette.co.uk/ media-audience-and-business-data/media_metrics/ facebooks-referral-traffic-for-publishers-down-50-in-12-months/

    2024

  46. [55]

    Pecile, N

    G. Pecile, N. Di Marco, M. Cinelli, W. Quattrociocchi, Mapping the global election landscape on social media in 2024, PLOS ONE 20 (2) (2025) e0316271. doi:10.1371/journal.pone.0316271

    2025 doi

  47. [56]

    A. M. Guess, N. Malhotra, J. Pan, P. Barber ´a, H. Allcott, T. Brown, A. Crespo-Tenorio, D. Dimmery, D. Freelon, M. Gentzkow, et al., How do social media feed algorithms a ffect attitudes and behavior in an election campaign?, Science 381 (6656) (2023) 398–404

    2023

  48. [57]

    A. M. Guess, N. Malhotra, J. Pan, P. Barber ´a, H. Allcott, T. Brown, J. A. Tucker, How do social media feed algorithms affect attitudes and behavior in an election campaign?, Science 381 (6656) (2023b) 398–404

    2023

  49. [58]

    Nyhan, J

    B. Nyhan, J. Settle, E. Thorson, M. Wojcieszak, P. Barber ´a, A. Y . Chen, H. Allcott, T. Brown, A. Crespo-Tenorio, D. Dimmery, D. Freelon, M. Gentzkow, S. Gonz´alez-Bail´on, A. M. Guess, E. Kennedy, Y . M. Kim, D. Lazer, N. Malhotra, D. Moehler, J. Pan, D. R. Thomas, R. Tromb...

    2023

  50. [59]

    A. M. Guess, N. Malhotra, J. Pan, P. Barber ´a, H. Allcott, T. Brown, J. A. Tucker, Reshares on social media amplify political news but do not de- tectably affect beliefs or opinions, Science 381 (6656) (2023a) 404–408

    2023

  51. [60]

    X. Yu, M. Haroon, E. Menchen-Trevino, M. Wojcieszak, Nudging rec- ommendation algorithms increases news consumption and diversity on youtube, PNAS nexus 3 (12) (2024) pgae518

    2024

  52. [61]

    Bates, M

    D. Bates, M. M ¨achler, B. Bolker, S. Walker, Fitting Linear Mixed-Effects Models Using lme4, Journal of Statistical Software 67 (1) (2015). doi: 10.18637/jss.v067.i01

    2015 doi

  53. [62]

    M. E. Brooks, K. Kristensen, K. J. van Benthem, A. Magnusson, C. W. Berg, A. Nielsen, H. J. Skaug, M. M ¨achler, B. M. Bolker, glmmTMB Balances Speed and Flexibility Among Packages for Zero-inflated Gen- eralized Linear Mixed Modeling, The R Journal 9 (2) (2017) 378. doi: 10.3...

    2017 doi

  54. [63]

    Pennycook, D

    G. Pennycook, D. Rand, Fighting misinformation on social media us- ing crowdsourced judgments of news source quality, PNAS Nexus 116 (2019). doi:10.1093/pnasnexus/pgad286. URL https://doi.org/10.1093/pnasnexus/pgad286

    2019 doi

  55. [64]

    Lasser, S

    J. Lasser, S. T. Aroyehun, A. Simchon, F. Carrella, D. Garcia, S. Lewandowsky, Social media sharing by political elites: An asymmet- ric american exceptionalism, arXiv (arXiv:2207.06313) (2022). arXiv: 2207.06313, doi:10.48550/arXiv.2207.06313. URL http://arxiv.org/abs/2207.06313

    2022 doi arXiv

  56. [65]

    URL https://www.similarweb.com/

    Similarweb: AI-Powered Digital Data Intelligence Solutions (2025). URL https://www.similarweb.com/

    2025

  57. [66]

    J. C. Donnelly, BuzzFeed News Never Built the Brand Loyalty It Needed (Apr. 2023). URL https://www.amediaoperator.com/newsletter/ buzzfeed-news-never-built-the-brand-loyalty-it-needed/

    2023

  58. [67]

    Virtanen, R

    P. Virtanen, R. Gommers, T. E. Oliphant, M. Haberland, T. Reddy, D. Cournapeau, E. Burovski, P. Peterson, W. Weckesser, J. Bright, S. J. Van Der Walt, M. Brett, J. Wilson, K. J. Millman, N. Mayorov, A. R. J. Nelson, E. Jones, R. Kern, E. Larson, C. J. Carey,˙I. Polat, Y . Feng...

    2020

  59. [68]

    Backstrom, Helping Make Sure You Don’t Miss Stories from Friends 14 (Jun

    L. Backstrom, Helping Make Sure You Don’t Miss Stories from Friends 14 (Jun. 2016). URL https://about.fb.com/news/2016/06/ news-feed-fyi-helping-make-sure-you-dont-miss-stories-from-friends/

    2016

  60. [69]

    A. Lada, J. Li, S. Ding, New Signals to Show You More Authentic and Timely Stories (Jan. 2017). URL https://about.fb.com/news/2017/01/ news-feed-fyi-new-signals-to-show-you-more-authentic-and-timely-stories/

    2017

  61. [70]

    Su, New Test With Related Articles (Apr

    S. Su, New Test With Related Articles (Apr. 2017). URL https://about.fb.com/news/2017/04/ news-feed-fyi-new-test-with-related-articles/

    2017

  62. [71]

    Mosseri, Helping Ensure News on Facebook Is From Trusted Sources (Jan

    A. Mosseri, Helping Ensure News on Facebook Is From Trusted Sources (Jan. 2018). URL https://about.fb.com/news/2018/01/trusted-sources/

    2018

  63. [72]

    Parkland high school shooting, page Version ID: 1296272133 (Jun. 2025). URL https://en.wikipedia.org/w/index.php?title= Parkland_high_school_shooting&oldid=1296272133

    2025

  64. [73]

    Murder of George Floyd, page Version ID: 1296879981 (Jun. 2025). URL https://en.wikipedia.org/w/index.php?title=Murder_ of_George_Floyd&oldid=1296879981

    2025

  65. [74]

    Brown, Prioritizing original news reporting on facebook (2020)

    C. Brown, Prioritizing original news reporting on facebook (2020). URL https://about.fb.com/news/2020/06/ prioritizing-original-news-reporting-on-facebook/

    2020

  66. [75]

    Sethuraman, J

    R. Sethuraman, J. Vallmitjana, J. Levin, Using Surveys to Make News Feed More Personal (May 2019). URL https://about.fb.com/news/2019/05/ more-personalized-experiences/

    2019

  67. [76]

    troop withdrawal from Afghanistan, page Version ID: 1296897229 (Jun

    2020–2021 U.S. troop withdrawal from Afghanistan, page Version ID: 1296897229 (Jun. 2025). URL https://en.wikipedia.org/w/index.php?title=2020% E2%80%932021_U.S._troop_withdrawal_from_Afghanistan& oldid=1296897229

    2020

  68. [77]

    Acquisition of Twitter by Elon Musk, page Version ID: 1296764074 (Jun. 2025). URL https://en.wikipedia.org/w/index.php?title= Acquisition_of_Twitter_by_Elon_Musk&oldid=1296764074

    2025

  69. [78]

    Meta, An Update on Facebook News (Mar. 2024). URL https://about.fb.com/news/2024/02/ update-on-facebook-news-us-australia/

    2024

  70. [79]

    Andersen, A

    K. Andersen, A. Shehata, M. Skovsgaard, J. Str ¨omb¨ack, Selective news avoidance: Consistency and temporality, Communication Research 0 (0) (2024) 00936502231221689. arXiv:https://doi.org/10.1177/ 00936502231221689, doi:10.1177/00936502231221689. URL https://doi.org/10.1177/0...

    2024 doi

  71. [80]

    Statista, U.S.: Facebook users 2019-2028, https://www.statista.com/statistics/408971/number-of-us-facebook-users/ (2025)

    2025

  72. [81]

    Newsroom (Jun. 2025). URL https://about.fb.com/news/

    2025

  73. [82]

    Meta, Timeline, archived at Wayback Machine (https: //web.archive.org/) > https://www.facebook.com/formedia/mjp/timeline (Aug. 2023). URL https://web.archive.org/web/20230803114746/https: //www.facebook.com/formedia/mjp/timeline

    2023

  74. [83]

    URL https://transparency.meta.com/reports/ community-standards-enforcement/

    Meta, Community Standards Enforcement (2025). URL https://transparency.meta.com/reports/ community-standards-enforcement/

    2025

  75. [84]

    URL https://transparency.meta.com/en-gb/features/ approach-to-ranking/cdgs-changes-corrections

    Meta, CDG Corrections and adjustments (2025). URL https://transparency.meta.com/en-gb/features/ approach-to-ranking/cdgs-changes-corrections

    2025

  76. [85]

    Company, Integrity Timeline 2016-2021 (2025)

    F. Company, Integrity Timeline 2016-2021 (2025). URL https://about.fb.com/wp-content/uploads/2021/09/ FB_Integrity-Timeline.pdf

    2025

  77. [86]

    URL https://www.oversightboard.com/

    Oversight Board, running Time: 107 (2025). URL https://www.oversightboard.com/

    2025

  78. [87]

    URL https://transparency.meta.com/en-gb/metasecurity/ threat-reporting

    Meta’s threat disruptions (2025). URL https://transparency.meta.com/en-gb/metasecurity/ threat-reporting

    2025

  79. [88]

    URL https://about.fb.com/wp-content/uploads/2021/12/ Meta-Response_Philippines-Human-Rights-Impact-Assessment

    Meta Response: Philippines Human Rights Impact Assessment (2021). URL https://about.fb.com/wp-content/uploads/2021/12/ Meta-Response_Philippines-Human-Rights-Impact-Assessment. pdf

    2021

  80. [89]

    Sissons, An Independent Due Diligence Exercise into Meta’s Human Rights Impact in Israel and Palestine During the May 2021 Escalation (Sep

    M. Sissons, An Independent Due Diligence Exercise into Meta’s Human Rights Impact in Israel and Palestine During the May 2021 Escalation (Sep. 2022). URL https://about.fb.com/news/2022/09/ human-rights-impact-meta-israel-palestine/

    2021

  81. [90]

    URL https://transparency.meta.com/features/ approach-to-newsworthy-content/

    Meta, Approach to Newsworthy Content (2024). URL https://transparency.meta.com/features/ approach-to-newsworthy-content/

    2024

  82. [91]

    Gonz ´alez-Bail´on, D

    S. Gonz ´alez-Bail´on, D. Lazer, P. Barber ´a, M. Zhang, H. Allcott, T. Brown, A. Crespo-Tenorio, D. Freelon, M. Gentzkow, A. M. Guess, et al., Asymmetric ideological segregation in exposure to political news on facebook, Science 381 (6656) (2023) 392–398. Acknowledgments Fund...

    2023

  83. [93]

    The primary method for collecting data was a real-time integration with the Facebook API, whereby Sotrender retrieved posts and updated interaction metrics on an hourly basis

    Ongoing Data Collection. The primary method for collecting data was a real-time integration with the Facebook API, whereby Sotrender retrieved posts and updated interaction metrics on an hourly basis. This method allowed for close tracking of user engagement in the initial per...

  84. [94]

    For posts not captured through ongoing data collection, Sotrender utilized the Facebook API’s historical data retrieval option

    Historical Data Retrieval. For posts not captured through ongoing data collection, Sotrender utilized the Facebook API’s historical data retrieval option. This method allows for the acquisition of engagement metrics for posts beyond the two-week tracking window. Upon examinati...

    2016

  85. [95]

    Imputing ∼1.1% of missing reaction counts

  86. [96]

    24 Appendix F

    Generating model-based expectations and coe fficients of variations to be used as the signal for the changepoint detection (see Appendix F). 24 Appendix F . Changepoint detection To identify moments of significant changes in dynamics of reactions to news over time, a Bayesian ...

  87. [97]

    Run BEAST on the input time series k times, each time using a different random seed

  88. [98]

    Extract the estimate of the posterior distribution (with weekly time-resolution)

  89. [99]

    p1,w p2,1 p2,2

    Group estimated posterior probabilities in weekly buckets to obtain a k× w array: P =  p1,1 p1,2 ... p1,w p2,1 p2,2 ... p2,w ... ... ... ... pk,1 pk,2 ... pk, w  where pi, j is a posterior probability from i’th run that there is a changepoint ...

  90. [100]

    The purpose of this transformation is to group neighboring posterior probabilities

    Smoothen the rows of P by applying a window function such that: ˜Pi, j = 1− u= j+lY u= j−l 1− pi,u which corresponds to a posterior probability that there is at least one changepoint in the ±l weeks interval around the j’th week. The purpose of this transformation is to group ...

  91. [101]

    Take arithmetic averages over the columns of P to obtain posterior probabilities averaged over multiple BEAST runs

  92. [102]

    Here we use pmin = 1/2

    Detect peaks (using scipy.signal.find peaks command from the scipy package for Python; 67) of the resulting series of posterior probabilities by equating them with local maxima satisfying two additional constraints: 25 (a) Only peaks of height pi, j≥ pmin are selected. Here we...

  93. [103]

    Use peak positions as point estimates of the changepoints’ positions

  94. [104]

    War on News

    Use peak widths as interval estimates of the changepoints’ positions. Appendix F .1. Events Relating to Changepoints To understand what might have led to the identified changepoints, we reviewed our curated list of algorithmic events and iden- tified major algorithmic changes ...

    2016

  95. [106]

    the typical case

    Conditional mean, or “the typical case”. In this approach the pure fixed effects are extracted by conditioning on the modes, or most typical values, of random e ffects. And since random e ffects are, by design, Gaussian and centered, this amounts to removing them from equation...

  96. [107]

    In this approach random e ffects are removed by marginalization, or taking expectation with respect to the random effects

    Marginal mean. In this approach random e ffects are removed by marginalization, or taking expectation with respect to the random effects. At the linear predictor scale, or for linear models (i.e. with identity link function), it yields the same result as the conditional mean a...

  97. [108]

    16.01.01-16.06.20 overall 2042.71 1216.79 3429.23 1.44 1.11 1.86 4.02 0.001 - - - - - - - - - - low 3920.40 1587.48 9681.69 1.58 1.01 2.47 2.92 0.041 - - - - - 1.92 0.98 3.75 2.28 0.058 medium 1841.62 743.55 4561.32 1.56 1.00 2.44 2.85 0.051 - - - - - 0.90 0.46 1.76 -0.36 0.93...

  98. [109]

    16.06.20-17.03.06 overall 3252.84 1940.82 5451.81 2.29 1.78 2.96 9.28 0.000 1.59 1.10 2.31 3.51 0.005 - - - - - low 6196.06 2509.61 15297.70 2.49 1.60 3.90 5.85 0.000 1.58 0.83 3.02 1.98 0.373 1.90 0.98 3.71 2.26 0.061 medium 2960.95 1195.60 7332.92 2.51 1.61 3.93 5.88 0.000 1...

    1940

  99. [110]

    17.03.06-18.03.19 overall 1963.19 1171.46 3290.01 1.38 1.07 1.78 3.63 0.003 0.60 0.42 0.87 -3.83 0.001 - - - - - low 2685.78 1087.99 6630.07 1.08 0.69 1.69 0.50 1.000 0.43 0.23 0.83 -3.62 0.003 1.37 0.70 2.67 1.10 0.514 medium 1986.65 802.25 4919.64 1.68 1.08 2.63 3.33 0.010 0...

    1963

  100. [111]

    18.03.19-20.04.13 overall 1550.80 925.48 2598.65 1.09 0.85 1.41 0.99 0.988 0.79 0.55 1.14 -1.79 0.511 - - - - - low 2699.97 1093.86 6664.32 1.09 0.70 1.70 0.53 1.000 1.01 0.53 1.92 0.02 1.000 1.74 0.89 3.39 1.95 0.126 medium 1283.86 518.48 3179.09 1.09 0.70 1.70 0.54 1.000 0.6...

  101. [112]

    20.04.13-21.03.01 overall 2397.26 1427.91 4024.66 1.69 1.30 2.19 5.79 0.000 1.55 1.07 2.24 3.28 0.011 - - - - - low 4895.55 1983.13 12085.18 1.97 1.26 3.08 4.34 0.000 1.81 0.95 3.46 2.58 0.095 2.04 1.05 3.99 2.50 0.033 medium 1484.57 594.07 3709.93 1.26 0.79 2.00 1.42 0.858 1....

    1983

  102. [113]

    21.03.01-21.07.26 overall 1965.39 1168.35 3306.15 1.38 1.06 1.80 3.54 0.005 0.82 0.56 1.20 -1.47 0.751 - - - - - low 5042.48 2024.24 12561.04 2.03 1.28 3.23 4.35 0.000 1.03 0.53 2.00 0.13 1.000 2.57 1.31 5.03 3.28 0.003 medium 1093.76 437.59 2733.91 0.93 0.58 1.48 -0.46 1.000 ...

    1965

  103. [114]

    21.07.26-22.04.11 overall 1793.40 1068.15 3011.09 1.26 0.98 1.64 2.58 0.111 0.91 0.63 1.33 -0.68 0.998 - - - - - low 3405.12 1379.23 8406.78 1.37 0.88 2.14 2.02 0.405 0.68 0.35 1.31 -1.66 0.611 1.90 0.97 3.71 2.25 0.064 medium 1105.36 442.29 2762.51 0.94 0.59 1.49 -0.40 1.000 ...

  104. [115]

    22.04.11-22.10.31 overall 1185.03 705.75 1989.81 0.83 0.64 1.08 -2.00 0.421 0.66 0.45 0.96 -3.10 0.020 - - - - - low 1749.41 708.53 4319.37 0.70 0.45 1.10 -2.24 0.256 0.51 0.27 0.98 -2.89 0.040 1.48 0.76 2.88 1.36 0.360 medium 908.27 363.40 2270.11 0.77 0.48 1.23 -1.61 0.733 0...

    1989

  105. [116]

    22.10.31-23.06.26 overall 845.07 504.20 1416.38 0.60 0.46 0.77 -5.81 0.000 0.71 0.49 1.04 -2.54 0.105 - - - - - low 1082.44 438.43 2672.40 0.44 0.28 0.68 -5.32 0.000 0.62 0.32 1.18 -2.08 0.309 1.28 0.66 2.50 0.87 0.660 medium 731.21 295.23 1811.02 0.62 0.40 0.97 -3.05 0.027 0....

  106. [117]

    23.06.26-24.04.01 overall 519.57 309.48 872.29 0.37 0.28 0.47 -11.11 0.000 0.61 0.42 0.89 -3.66 0.003 - - - - - low 676.57 274.06 1670.27 0.27 0.17 0.43 -8.32 0.000 0.63 0.33 1.19 -2.04 0.337 1.30 0.67 2.54 0.93 0.624 medium 524.33 209.85 1310.13 0.44 0.28 0.71 -4.98 0.000 0.7...

  107. [118]

    24.04.01-24.12.09 overall 633.11 377.33 1062.26 0.45 0.34 0.58 -8.97 0.000 1.22 0.84 1.77 1.48 0.746 - - - - - low 1344.54 544.59 3319.54 0.54 0.35 0.85 -3.93 0.001 1.99 1.04 3.80 2.98 0.030 2.12 1.09 4.14 2.64 0.023 medium 560.14 224.91 1395.06 0.47 0.30 0.75 -4.64 0.000 1.07...

  108. [119]

    Regression model with epoch e ffects for news and non-news posts Fig

    24.12.09-25.02.13 overall 1192.89 709.85 2004.62 0.84 0.65 1.09 -1.91 0.489 1.88 1.29 2.74 4.73 0.000 - - - - - low 2163.34 875.59 5345.04 0.87 0.56 1.36 -0.88 0.995 1.61 0.84 3.08 2.06 0.323 1.81 0.93 3.54 2.08 0.093 medium 1487.18 594.40 3720.88 1.26 0.79 2.01 1.42 0.858 2.6...

    2004

  109. [120]

    16.01.01-16.07.11 overall 2951.19 1870.61 4655.98 - - - - - - - - - - non-news 4365.72 1935.36 9848.05 1.06 0.73 1.55 0.47 1.000 - - - - - news 1994.98 1118.51 3558.24 0.94 0.72 1.21 -0.71 0.999 - - - - -

    1935

  110. [121]

    16.07.11-17.03.06 overall 4172.34 2645.67 6579.97 1.41 0.98 2.04 2.64 0.081 1.41 0.98 2.04 2.64 0.081 non-news 5156.94 2288.04 11623.06 1.26 0.86 1.83 1.73 0.615 1.18 0.64 2.17 0.77 0.995 news 3375.73 1893.22 6019.15 1.59 1.23 2.05 5.11 0.000 1.69 1.12 2.56 3.56 0.004

  111. [122]

    17.03.06-18.03.19 overall 2579.90 1636.58 4066.95 0.62 0.43 0.89 -3.68 0.003 0.62 0.43 0.89 -3.68 0.003 non-news 3394.09 1507.47 7641.82 0.83 0.57 1.20 -1.44 0.825 0.66 0.36 1.21 -1.94 0.403 news 1961.03 1099.90 3496.34 0.92 0.71 1.19 -0.90 0.993 0.58 0.38 0.88 -3.68 0.003

    1961

  112. [123]

    18.03.19-20.04.13 overall 2398.75 1522.10 3780.32 0.93 0.64 1.34 -0.56 1.000 0.93 0.64 1.34 -0.56 1.000 non-news 3713.06 1650.32 8354.00 0.90 0.62 1.32 -0.76 0.998 1.09 0.60 2.00 0.42 1.000 news 1549.67 869.23 2762.76 0.73 0.56 0.94 -3.50 0.005 0.79 0.52 1.20 -1.59 0.663

  113. [124]

    20.04.13-21.03.01 overall 3260.32 2075.61 5121.24 1.36 0.95 1.95 2.38 0.161 1.36 0.95 1.95 2.38 0.161 non-news 4438.24 1993.90 9879.11 1.08 0.68 1.73 0.47 1.000 1.20 0.66 2.16 0.85 0.990 news 2395.03 1339.95 4280.88 1.13 0.80 1.58 1.00 0.983 1.55 1.02 2.35 2.93 0.035

    1993

  114. [125]

    21.03.01-21.07.26 overall 2785.99 1767.92 4390.33 0.85 0.59 1.23 -1.22 0.899 0.85 0.59 1.23 -1.22 0.899 non-news 3948.87 1761.90 8850.41 0.96 0.59 1.56 -0.23 1.000 0.89 0.49 1.60 -0.56 1.000 news 1965.56 1096.85 3522.30 0.92 0.66 1.30 -0.66 1.000 0.82 0.54 1.25 -1.31 0.854

    1965

  115. [126]

    21.07.26-22.04.18 overall 2760.01 1757.52 4334.33 0.99 0.69 1.42 -0.07 1.000 0.99 0.69 1.42 -0.07 1.000 non-news 4271.51 1920.29 9501.55 1.04 0.65 1.66 0.24 1.000 1.08 0.60 1.95 0.38 1.000 news 1783.37 997.70 3187.73 0.84 0.60 1.17 -1.49 0.796 0.91 0.59 1.39 -0.64 0.999

    1920

  116. [127]

    22.04.18-22.10.31 overall 2744.58 1747.18 4311.34 0.99 0.70 1.42 -0.04 1.000 0.99 0.70 1.42 -0.04 1.000 non-news 6384.87 2868.25 14213.02 1.56 0.97 2.49 2.67 0.083 1.49 0.84 2.67 1.95 0.393 news 1179.77 659.98 2108.95 0.55 0.40 0.78 -4.98 0.000 0.66 0.43 1.01 -2.76 0.058

  117. [128]

    22.10.31-23.07.03 overall 2138.20 1362.53 3355.47 0.78 0.55 1.11 -1.97 0.384 0.78 0.55 1.11 -1.97 0.384 non-news 5477.42 2462.71 12182.54 1.33 0.83 2.13 1.75 0.595 0.86 0.48 1.53 -0.75 0.996 news 834.69 468.12 1488.29 0.39 0.28 0.55 -8.01 0.000 0.71 0.47 1.07 -2.32 0.182

  118. [129]

    23.07.03-24.03.25 overall 2018.92 1285.84 3169.93 0.94 0.66 1.35 -0.45 1.000 0.94 0.66 1.35 -0.45 1.000 non-news 7879.11 3543.74 17518.36 1.92 1.20 3.07 3.96 0.001 1.44 0.81 2.56 1.77 0.526 news 517.32 289.43 924.64 0.24 0.17 0.34 -11.96 0.000 0.62 0.41 0.94 -3.21 0.014

    2018

  119. [130]

    24.03.25-24.12.09 overall 2234.75 1420.90 3514.74 1.11 0.77 1.59 0.79 0.994 1.11 0.77 1.59 0.79 0.994 non-news 7691.11 3440.01 17195.62 1.87 1.16 3.02 3.74 0.002 0.98 0.54 1.75 -0.12 1.000 news 649.33 363.79 1159.00 0.31 0.22 0.43 -10.11 0.000 1.26 0.83 1.91 1.52 0.717

  120. [131]

    24.12.09-25.02.13 overall 3128.20 1974.70 4955.50 1.40 0.97 2.03 2.55 0.105 1.40 0.97 2.03 2.55 0.105 non-news 8153.36 3581.88 18559.33 1.99 1.20 3.30 3.85 0.001 1.06 0.57 1.96 0.27 1.000 news 1200.20 671.07 2146.53 0.56 0.40 0.79 -4.83 0.000 1.85 1.21 2.81 4.11 0.000 33 Table...

    1974

  121. [132]

    16.01.01-16.07.11 - - - - - 0.88 0.56 1.39 -0.78 0.998

  122. [133]

    16.07.11-17.03.06 1.43 0.69 2.99 1.37 0.820 1.26 0.80 1.99 1.46 0.815

  123. [134]

    17.03.06-18.03.19 0.88 0.42 1.84 -0.48 1.000 1.11 0.71 1.76 0.68 0.999

  124. [135]

    18.03.19-20.04.13 0.72 0.35 1.50 -1.25 0.886 0.81 0.51 1.27 -1.36 0.874

  125. [136]

    20.04.13-21.03.01 1.29 0.63 2.67 1.00 0.969 1.04 0.58 1.86 0.20 1.000

  126. [137]

    21.03.01-21.07.26 0.92 0.45 1.90 -0.31 1.000 0.96 0.53 1.73 -0.19 1.000

  127. [138]

    21.07.26-22.04.18 0.84 0.41 1.73 -0.68 0.998 0.81 0.45 1.44 -1.07 0.973

  128. [139]

    22.04.18-22.10.31 0.44 0.22 0.90 -3.20 0.014 0.36 0.20 0.64 -5.07 0.000

  129. [140]

    22.10.31-23.07.03 0.82 0.40 1.68 -0.76 0.995 0.29 0.17 0.52 -6.06 0.000

  130. [141]

    23.07.03-24.03.25 0.43 0.21 0.88 -3.32 0.009 0.13 0.07 0.23 -10.19 0.000

  131. [142]

    24.03.25-24.12.09 1.29 0.63 2.64 0.98 0.972 0.16 0.09 0.29 -8.86 0.000

  132. [143]

    War on News

    24.12.09-25.02.13 1.74 0.83 3.66 2.10 0.295 0.28 0.15 0.52 -5.88 0.000 34 Appendix I. Total news suppression e ffects by sector and quality tiers Table I.13 presents estimates of the relative total reaction changes between epochs 4 (the peak before the onset of the “War on New...

    2021

  133. [272]

    doi:10.1038/s41592-019-0686-2

    doi

  134. [2021]

    These changes would then be iterated, reevaluated and, eventually removed, leading to updates to the initial announcement that have continued through at least May, 2025 (9)

    These algorithmic changes went through extensive testing during epochs 5 and onwards, matching policy changes, additional algorithmic refinements, changes to core features around how news was displayed, and gradual global deployment that seems to correspond to the May through ...

    2025

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.