pith. sign in

arxiv: 2411.03241 · v3 · submitted 2024-11-05 · 💰 econ.TH

Troll Farms

Philipp Denter , Boris Ginzburg This is my paper

Pith reviewed 2026-05-23 17:52 UTC · model grok-4.3

classification 💰 econ.TH
keywords disinformationelectionsinformation designpolarizationsignal precisionmanipulationvotingtroll farms
0
0 comments X

The pith

Disinformation campaigns manipulate elections more effectively when voters have precise independent signals but less so when the electorate is polarized.

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

The paper develops a model in which a sender sends uninformative messages that mimic voters' own exogenous signals to influence election results. The model shows that the sender's ability to shift outcomes rises as voters' independent signals become more precise. This ability falls as the electorate grows more polarized. When sending messages carries a cost, the sender stops targeting voters who are only slightly opposed and reduces the extremism of messages sent to core supporters.

Core claim

In the constrained information design model, the sender targets initially opposed voters with favorable uninformative messages and supporters with unfavorable ones to dilute adverse information. The sender's manipulation ability increases with greater precision of voters' independent signals but decreases with polarization. When messaging is costly, the sender may stop targeting marginally opposing voters while moderating message extremism among supporters.

What carries the argument

Constrained information design model in which uninformative messages mimic exogenous informative signals and are targeted by each voter's type and prior belief.

If this is right

  • Greater precision in voters' independent signals allows the sender to achieve larger shifts in election outcomes.
  • Higher polarization in the electorate limits the sender's overall influence.
  • Costly messaging leads the sender to ignore marginally opposed voters.
  • The sender moderates the extremism of messages directed at supporters to preserve their beliefs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Platforms could limit manipulation by making it easier for voters to distinguish fabricated messages from genuine signals.
  • The model suggests that interventions reducing polarization may have limited effect if signal precision remains high.
  • Extensions could examine how partial voter awareness of possible disinformation alters the sender's optimal strategy.

Load-bearing premise

Voters treat the sender's uninformative messages as if they were their exogenous informative signals and update beliefs accordingly, and the sender can perfectly observe and target based on each voter's type.

What would settle it

A lab experiment or field study that measures whether voters update beliefs from fabricated messages in the same manner as from genuine signals, or checks whether real targeting patterns match the model's prediction of selective focus under costly messaging.

read the original abstract

We study how coordinated disinformation campaigns affect elections. We develop a constrained information design model in which a sender deploys uninformative messages that mimic voters' exogenous informative signals. Voters initially opposed to the sender's preferred outcome receive favourable messages, while those in favour are targeted with unfavourable messages to dilute adverse information. The sender's ability to manipulate political outcomes increases with greater precision of voters' independent signals, but decreases with polarisation. When messaging is costly, the sender may stop targeting marginally opposing voters while moderating message extremism among supporters.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper develops a constrained information design model in which a sender deploys uninformative messages that mimic voters' exogenous informative signals to influence election outcomes. Opposed voters receive favorable messages and supporters receive unfavorable ones to dilute adverse information. The central claims are that the sender's manipulation ability increases with the precision of voters' independent signals but decreases with polarization; when messaging is costly, the sender stops targeting marginally opposing voters and moderates message extremism among supporters.

Significance. If the results hold, the model offers a novel theoretical account of troll-farm-style disinformation campaigns, with counterintuitive comparative statics linking signal precision and polarization to manipulation effectiveness. The constrained information-design framing is a useful addition to the literature on political persuasion.

major comments (2)
  1. [Model setup] Model setup (voter updating rule): The directional claims on precision and polarization rest on voters treating the sender's fabricated messages exactly as if they were drawn from the exogenous signal distribution, with no Bayesian correction for the existence or targeting rule of the campaign. The manuscript does not solve or report an equilibrium in which voters assign positive probability to mimicry and update on the sender's strategy; relaxing this assumption would reduce the effective information content of the messages and the comparative statics need not survive.
  2. [Model setup] Sender information and commitment: The sender is assumed to observe every voter's type and prior perfectly and to commit to a message for each voter. No robustness check is provided for the case of imperfect observation or for equilibria in which the sender cannot commit to the full targeting plan; either change would alter the feasible manipulation set and the reported results on costly messaging.
minor comments (1)
  1. [Abstract] The abstract states the main results without reference to equilibrium definitions or robustness checks; the introduction or model section should preview the key assumptions and where the derivations appear.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive report and the opportunity to clarify the modeling choices. We address each major comment below, defending the assumptions as appropriate for the troll-farm context while agreeing to add explicit discussion of their implications.

read point-by-point responses

  1. Referee: [Model setup] Model setup (voter updating rule): The directional claims on precision and polarization rest on voters treating the sender's fabricated messages exactly as if they were drawn from the exogenous signal distribution, with no Bayesian correction for the existence or targeting rule of the campaign. The manuscript does not solve or report an equilibrium in which voters assign positive probability to mimicry and update on the sender's strategy; relaxing this assumption would reduce the effective information content of the messages and the comparative statics need not survive.

    Authors: The model deliberately assumes that fabricated messages are treated as draws from the exogenous distribution to isolate the effect of mimicry in disinformation campaigns. This reflects settings where troll-farm messages are crafted to be indistinguishable and voters do not detect or Bayesian-update on the campaign's existence. A full rational-expectations equilibrium with positive probability on mimicry would indeed alter the information content and potentially the comparative statics, but that constitutes a distinct modeling exercise. We will add a dedicated paragraph in the introduction and a remark in the conclusion discussing this assumption, its empirical motivation, and the scope for future extensions. revision: partial

  2. Referee: [Model setup] Sender information and commitment: The sender is assumed to observe every voter's type and prior perfectly and to commit to a message for each voter. No robustness check is provided for the case of imperfect observation or for equilibria in which the sender cannot commit to the full targeting plan; either change would alter the feasible manipulation set and the reported results on costly messaging.

    Authors: Perfect observation and commitment are maintained to characterize the sender's maximum feasible influence under the constrained information-design approach, which is standard for deriving optimal targeting. The costly-messaging results on selective targeting and message moderation follow directly from this benchmark. While imperfect observation or limited commitment would shrink the manipulation set, the qualitative pattern of prioritizing high-impact voters may survive under noisy type estimates. We will insert a short discussion of these assumptions and their role as an upper-bound benchmark in the model section. revision: partial

Circularity Check

0 steps flagged

Theoretical model derivation is self-contained; no circular steps identified

full rationale

The paper constructs a constrained information design model from stated assumptions on voter updating (treating sender messages as exogenous signals) and sender targeting capabilities. Comparative statics on manipulation increasing with signal precision and decreasing with polarization follow directly from solving the model under those assumptions. No equations or claims reduce by construction to fitted parameters, self-definitions, or load-bearing self-citations. The provided abstract and description contain no references to prior author work used to justify uniqueness or ansatzes. This is a standard theoretical construction that is self-contained against its own primitives.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

Review based on abstract only; specific free parameters, axioms, and invented entities cannot be audited in detail. The model appears to rest on standard Bayesian updating and sender-receiver information-design assumptions common to the field.

axioms (2)
  • domain assumption Voters update beliefs using Bayes' rule when receiving messages that mimic their exogenous signals
    Core to any information-design model; invoked implicitly by the abstract's description of message effects.
  • domain assumption The sender can perfectly target messages conditional on each voter's type and prior
    Required for the targeting strategy described in the abstract.

pith-pipeline@v0.9.0 · 5593 in / 1378 out tokens · 26783 ms · 2026-05-23T17:52:31.523780+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages

  1. [1]

    Ozdaglar, and J

    Acemoglu, D., A. Ozdaglar, and J. Siderius (2023): A Model of Online Misinformation , The Review of Economic Studies, rdad111

    work page 2023

  2. [2]

    Lazarus, and M

    Augenblick, N., E. Lazarus, and M. Thaler (2024): Overinference from Weak Signals and Underinference from Strong Signals , The Quarterly Journal of Economics, qjae032

    work page 2024

  3. [3]

    Ba, C., J. A. Bohren, and A. Imas (2023): Over-and Underreaction to Information, SSRN Working Paper 4274617

    work page 2023

  4. [4]

    Bardhi, A. and Y. Guo (2018): Modes of persuasion toward unanimous consent, Theoretical Economics, 13, 1111--1149

    work page 2018

  5. [5]

    BBC (2018): Russia trolls 'spreading vaccination misinformation' to create discord, Accessed on 10 April 2024

    work page 2018

  6. [6]

    Bergemann, D. and S. Morris (2016): Information Design, Bayesian Persuasion, and Bayes Correlated Equilibrium, American Economic Review, 106, 586--91

    work page 2016

  7. [7]

    Shadmehr, and K

    Boleslavsky, R., M. Shadmehr, and K. Sonin (2021): Media Freedom in the Shadow of a Coup, Journal of the European Economic Association, 19, 1782--1815

    work page 2021

  8. [8]

    Gentzkow, and J

    Boxell, L., M. Gentzkow, and J. M. Shapiro (2022): Cross-country trends in affective polarization, Review of Economics and Statistics, 1--60

    work page 2022

  9. [9]

    Cai, H. and J. T.-Y. Wang (2006): Overcommunication in strategic information transmission games, Games and Economic Behavior, 56, 7--36

    work page 2006

  10. [10]

    Gupta, F

    Chan, J., S. Gupta, F. Li, and Y. Wang (2019): Pivotal persuasion, Journal of Economic Theory, 180, 178--202

    work page 2019

  11. [11]

    (2024): Motivated Reasoning and the Political Economy of Climate Change Inaction, arXiv:2410.20982

    Denter, P. (2024): Motivated Reasoning and the Political Economy of Climate Change Inaction, arXiv:2410.20982

    work page arXiv 2024

  12. [12]

    Dumav, and B

    Denter, P., M. Dumav, and B. Ginzburg (2021): Social Connectivity, Media Bias, and Correlation Neglect , The Economic Journal, 131, 2033--2057

    work page 2021

  13. [13]

    Ottaviani, and P

    Di Tillio, A., M. Ottaviani, and P. N. S rensen (2021): Strategic sample selection, Econometrica, 89, 911--953

    work page 2021

  14. [14]

    Ederer, F. and W. Min (2022): Bayesian persuasion with lie detection, NBER Working Paper 30065

    work page 2022

  15. [15]

    (2013): Information manipulation, coordination, and regime change, Review of Economic studies, 80, 1422--1458

    Edmond, C. (2013): Information manipulation, coordination, and regime change, Review of Economic studies, 80, 1422--1458

    work page 2013

  16. [16]

    Edmond, C. and Y. K. Lu (2021): Creating confusion, Journal of Economic Theory, 191, 105145

    work page 2021

  17. [17]

    Escud \'e , M. and L. Sinander (2023): Slow persuasion, Theoretical Economics, 18, 129--162

    work page 2023

  18. [18]

    Euronews (2023): Meet the trollhunters who got 600 climate change deniers banned from Twitter, Accessed on 10 April 2024

    work page 2023

  19. [19]

    Varol, C

    Ferrara, E., O. Varol, C. Davis, F. Menczer, and A. Flammini (2016): The rise of social bots, Communications of the ACM, 59, 96--104

    work page 2016

  20. [20]

    Freedom House (2017): Manipulating Social Media to Undermine Democracy, Accessed on 26 April 2021

    work page 2017

  21. [21]

    --- -.1pt --- -.1pt --- (2019): The Crisis of Social Media, Accessed on 26 April 2021

    work page 2019

  22. [22]

    Gehlbach, S. and K. Sonin (2014): Government control of the media, Journal of Public Economics, 118, 163--171

    work page 2014

  23. [23]

    (2019): Optimal information censorship, Journal of Economic Behavior & Organization, 163, 377--385

    Ginzburg, B. (2019): Optimal information censorship, Journal of Economic Behavior & Organization, 163, 377--385

    work page 2019

  24. [24]

    --- -.1pt --- -.1pt --- (2024): Comparing Electoral Polarization Levels,

    work page 2024

  25. [25]

    and J.-A

    Ginzburg, B. and J.-A. Guerra (2019): When collective ignorance is bliss: Theory and experiment on voting for learning, Journal of Public Economics, 169, 52--64

    work page 2019

  26. [26]

    Gitmez, A. A. and P. Molavi (2023): Informational Autocrats, Diverse Societies, arXiv:2203.12698

    work page arXiv 2023

  27. [27]

    Gitmez, A. A. and K. Sonin (2023): The Dictator's Dilemma: A Theory of Propaganda and Repression, University of Chicago, Becker Friedman Institute for Economics Working Paper No. 2023-67

    work page 2023

  28. [28]

    Goldstein, J. A., G. Sastry, M. Musser, R. DiResta, M. Gentzel, and K. Sedova (2023): Generative Language Models and Automated Influence Operations: Emerging Threats and Potential Mitigations, arXiv preprint arXiv:2301.04246

    work page arXiv 2023

  29. [29]

    Pham, and O

    Gorodnichenko, Y., T. Pham, and O. Talavera (2021): Social media, sentiment and public opinions: Evidence from\# Brexit and\# USElection, European Economic Review, 136, 103772

    work page 2021

  30. [30]

    Heller, and A

    Gradwohl, R., Y. Heller, and A. Hillman (2022): Social Media and Democracy, arXiv preprint arXiv:2206.14430

    work page arXiv 2022

  31. [31]

    Haustein, S., T. D. Bowman, K. Holmberg, A. Tsou, C. R. Sugimoto, and V. Larivi \`e re (2016): Tweets as impact indicators: Examining the implications of automated ``bot'' accounts on Twitter, Journal of the Association for Information Science and Technology, 67, 232--238

    work page 2016

  32. [32]

    Heese, C. and S. Lauermann (2021): Persuasion and Information Aggregation in Elections, ECONtribute Discussion Paper No. 112

    work page 2021

  33. [33]

    Iyengar, S. and M. Krupenkin (2018): The strengthening of partisan affect, Political Psychology, 39, 201--218

    work page 2018

  34. [34]

    Jin, G. Z., M. Luca, and D. Martin (2021): Is no news (perceived as) bad news? An experimental investigation of information disclosure, American Economic Journal: Microeconomics, 13, 141--73

    work page 2021

  35. [35]

    Kamenica, E. and M. Gentzkow (2011): Bayesian Persuasion, American Economic Review, 101, 2590--2615

    work page 2011

  36. [36]

    Ottaviani, and F

    Kartik, N., M. Ottaviani, and F. Squintani (2007): Credulity, lies, and costly talk, Journal of Economic Theory, 134, 93--116

    work page 2007

  37. [37]

    Pan, and M

    King, G., J. Pan, and M. E. Roberts (2017): How the Chinese government fabricates social media posts for strategic distraction, not engaged argument, American political science review, 111, 484--501

    work page 2017

  38. [38]

    Mylovanov, and A

    Kolotilin, A., T. Mylovanov, and A. Zapechelnyuk (2022): Censorship as optimal persuasion, Theoretical Economics, 17, 561--585

    work page 2022

  39. [39]

    Le Treust , M. and T. Tomala (2019): Persuasion with limited communication capacity, Journal of Economic Theory, 184, 104940

    work page 2019

  40. [40]

    (2012): Campaigns to Track Voters with ``Political Cookies'', MIT Technology Review

    Leber, J. (2012): Campaigns to Track Voters with ``Political Cookies'', MIT Technology Review

    work page 2012

  41. [41]

    Martherus, J. L., A. G. Martinez, P. K. Piff, and A. G. Theodoridis (2021): Party animals? Extreme partisan polarization and dehumanization, Political Behavior, 43, 517--540

    work page 2021

  42. [42]

    Matyskov\' a , L. and A. Montes (2023): Bayesian persuasion with costly information acquisition, Journal of Economic Theory, 211, 105678

    work page 2023

  43. [43]

    Rahman, and M

    McCoy, J., T. Rahman, and M. Somer (2018): Polarization and the global crisis of democracy: Common patterns, dynamics, and pernicious consequences for democratic polities, American Behavioral Scientist, 62, 16--42

    work page 2018

  44. [44]

    Mylovanov, T. and A. Zapechelnyuk (2024): Constructive versus Toxic Argumentation in Debates, American Economic Journal: Microeconomics, 16, 262--292

    work page 2024

  45. [45]

    New York Times (2018): Cambridge Analytica and Facebook: The Scandal and the Fallout So Far, Accessed on 11 April 2024

    work page 2018

  46. [46]

    Onuchic, P. and D. Ray (2023): Conveying value via categories, Theoretical Economics, 18, 1407--1439

    work page 2023

  47. [47]

    Patty, J. W. and R. A. Weber (2007): Letting the good times roll: A theory of voter inference and experimental evidence, Public Choice, 130, 293--310

    work page 2007

  48. [48]

    Rossetti, M. and T. Zaman (2023): Bots, disinformation, and the first impeachment of US President Donald Trump, PLoS ONE, 18, e0283971

    work page 2023

  49. [49]

    Schipper, B. C. and H. Y. Woo (2019): Political Awareness, Microtargeting of Voters, and Negative Electoral Campaigning, Quarterly Journal of Political Science, 14, 41--88

    work page 2019

  50. [50]

    Shadmehr, M. and D. Bernhardt (2015): State censorship, American Economic Journal: Microeconomics, 7, 280--307

    work page 2015

  51. [51]

    Zhang, H

    Shen, F., E. Zhang, H. Zhang, W. Ren, Q. Jia, and Y. He (2023): Examining the differences between human and bot social media accounts: A case study of the Russia-Ukraine War, First Monday

    work page 2023

  52. [52]

    Edwards, and S

    Simchon, A., M. Edwards, and S. Lewandowsky (2024): The persuasive effects of political microtargeting in the age of generative artificial intelligence, PNAS Nexus, 3, 1--5

    work page 2024

  53. [53]

    Sisak, D. and P. Denter (2024): Information Sharing with Social Image Concerns and the Spread of Fake News, arXiv:2410.19557

    work page arXiv 2024

  54. [54]

    Strudwicke, I. J. and W. J. Grant (2020): \# junkscience: investigating pseudoscience disinformation in the Russian internet research agency tweets, Public Understanding of Science, 29, 459--472

    work page 2020

  55. [55]

    Tappin, B. M., C. Wittenberg, L. B. Hewitt, A. J. Berinsky, and D. G. Rand (2023): Quantifying the potential persuasive returns to political microtargeting, Proceedings of the National Academy of Sciences, 120, e2216261120

    work page 2023

  56. [56]

    The Guardian (2011): Revealed: US spy operation that manipulates social media, Accessed on 19 April 2021

    work page 2011

  57. [57]

    Tversky, A. and D. Kahneman (1992): Advances in prospect theory: Cumulative representation of uncertainty, Journal of Risk and uncertainty, 5, 297--323

    work page 1992

  58. [58]

    u ller, and J. Pr\

    van Gils, F., W. M\" u ller, and J. Pr\" u fer (2024): Microtargeting, voters' unawareness, and democracy, The Journal of Law, Economics, and Organization, ewae002

    work page 2024

  59. [59]

    Kosinski, and D

    Youyou, W., M. Kosinski, and D. Stillwell (2015): Computer-based personality judgments are more accurate than those made by humans, Proceedings of the National Academy of Sciences, 112, 1036--1040

    work page 2015

  60. [60]

    Dobber, G

    Zarouali, B., T. Dobber, G. D. Pauw, and C. de Vreese (2022): Using a Personality-Profiling Algorithm to Investigate Political Microtargeting: Assessing the Persuasion Effects of Personality-Tailored Ads on Social Media, Communication Research, 49, 1066--1091

    work page 2022

  61. [61]

    Petrova, and R

    Zhuravskaya, E., M. Petrova, and R. Enikolopov (2020): Political effects of the internet and social media, Annual Review of Economics, 12, 415--438

    work page 2020