Communication as Voting

arXiv:2505.14639 · econ.TH · Submitted 2025-05-20 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.

Ruth: We're looking at the paper: "Communication as Voting".

Warren: Here's the gist of the paper. The paper investigates information aggregation in informal elections using a cheap-talk model involving multiple senders and one receiver, establishing an equilibrium with a no-conflict property where communication only reveals information inducing the same preferred policy for both parties.

Ruth: The extent of aggregation is fundamentally tied to the nature of the conflict between senders and the receiver.

Warren: The model incorporates two distinct types of conflict: one arising from differences in payoff magnitudes in agreement states (L and H), which appears only under uncertainty, and another stemming from opposing policy preferences in a disagreement state (M), which persists even with full information.

Ruth: The presence of the disagreement state M prevents complete learning as the number of senders grows indefinitely.

Paper summary: Ruth: So, to recap what we've covered, the paper "Communication as Voting" looks at information aggregation in informal elections using a cheap-talk model with multiple senders and one receiver. The main idea is that communication can only reveal information that induces the same preferred policy for both sides >

Warren: And they establish this no-conflict property, showing it depends not just on the magnitude of the conflict between senders and receiver but fundamentally on what kind of conflict it is >

Iris: They break down that conflict into two types: one from differences in payoff magnitudes under uncertainty, and another from opposing policy preferences even with full information >

Ruth: And that second type, the disagreement state M, is crucial because it stops complete learning as the number of senders grows without bound >

Warren: What they claim is a discontinuity relative to binary benchmarks: if there's any positive probability of that disagreement state M, then communication can unravel as the population size increases sufficiently >

Iris: That's because they have conflicting requirements: uncertainty demands enough information to solve uncertainty, but full-information conflict makes highly informative communication unsustainable >

Ruth: They also introduce a mediator who commits ex ante to a mechanism that collects messages and recommends an outcome, which helps restore efficiency by enhancing transmission >

Warren: The paper demonstrates that as the number of senders grows large, simple step mechanisms can asymptotically achieve any outcome on the Pareto frontier between what the receiver's first best is and what senders could get with optimal Bayesian persuasion >

Iris: The central conclusion they draw is that information fails to aggregate in cheap-talk communication, and communication can unravel entirely >

Ruth: But mediated communication shows that it can asymptotically implement any Pareto-efficient outcome on the frontier between the receiver’s first best and the senders’ optimal Bayesian persuasion outcome >

Warren: They also suggest a centralized benchmark where under certain conditions, transmission fails in a single fully informed sender setting but persists in the decentralized setting with dispersed information >

Iris: It's about how those sources of conflict determine persistence; it depends on both types of conflict being small, and the presence of M sharpens those conditions >

Ruth: The paper also notes limitations on sender welfare, showing there's a finite population size N star beyond which more senders don't improve welfare >

Warren: They also extend it to nonbinary signal and message spaces, where monotone equilibria behave effectively like binary ones >

Conclusion: Ruth: So we wrap up this look at the paper "Communication as Voting." The authors are Kailin Chen, and they really focus on showing how the nature of conflict drives whether information aggregation works or fails >

Warren: It boils down to this: in cheap-talk communication, information doesn't aggregate well and communication can unravel entirely unless you have specific conditions on the conflict types being met >

Iris: The main takeaway is that while basic cheap-talk communication struggles, introducing a mediator changes things; it allows for asymptotic implementation of any Pareto-efficient outcome on the frontier between the receiver's best first choice and the senders' optimal Bayesian persuasion outcome >

Ruth: And they also show that this happens even in a multi-sender setting with dispersed information, where players actually get better outcomes than if they were centralized >

Warren: So for the listener, it means that when you're dealing with complex group decisions or elections, the way you structure the communication—whether it's centralized or decentralized—matters a lot for whether people actually end up getting a good result >

Aalto University

econ.TH

Submitted: 2025-05-20

Updated: 2026-11-05

License: http://creativecommons.org/licenses/by-nc-nd/4.0/

Importance score: 92/100

The gist: The paper investigates information aggregation in informal elections using a cheap-talk model involving multiple senders and one receiver…

Key concepts

Cheap-Talk Model
This model describes how parties communicate without direct monetary transfers. It examines how senders try to convey information about their preferences to a receiver in an informal setting, focusing on the equilibrium where communication only reveals information that both parties prefer.
Conflict Types (L/H and M)
The paper identifies two main conflicts: one related to payoff differences under uncertainty (L and H states) and another stemming from opposing policy preferences in a disagreement state (M). The presence of the disagreement state M is crucial as it prevents complete learning as more senders are added.
Mediated Communication
A mediator acts as an intermediary who commits beforehand to a mechanism that collects messages and recommends an outcome. This intervention enhances transmission efficiency, allowing simple step mechanisms to eventually achieve any Pareto-efficient result between the receiver's best and the senders' optimal outcomes.

Terminology

Summary

The paper investigates information aggregation in informal elections using a cheap-talk model involving multiple senders and one receiver, establishing an equilibrium with a no-conflict property where communication only reveals information inducing the same preferred policy for both parties. p. 0

Where the paper says it

Communication as Voting Kailin Chen∗ September 16, 2026 Abstract This paper studies information aggregation in informal elections by examining a cheap-talk model with multiple senders and one receiver. Each sender observes a noisy private signal about an unknown state and sends a message; the receiver observes the message tally and chooses the final policy choice.

The extent of aggregation is fundamentally tied to the nature of the conflict between senders and the receiver. p. 0
Where the paper says it

We establish an equilibrium no-conflict property according to which communication can reveal only information that induces the same preferred policy for both the receiver and the senders. Building on this property, we show that whether information aggregation and complete learning occur depends not only on the magnitude of the conflict of interest between the senders and the receiver but more fundamentally on the nature of that conflict.

The model incorporates two distinct types of conflict: one arising from differences in payoff magnitudes in agreement states (L and H), which appears only under uncertainty, and another stemming from opposing policy preferences in a disagreement state (M), which persists even with full information. p. 3

Where the paper says it

Our model thus features two qualitatively distinct forms of conflict: one stems from differences in payoff magnitudes in the agreement states, while the other stems from opposing policy preferences in the disagreement state. The first form of conflict arises only under uncertainty; by contrast, the second persists even under full information.

The presence of the disagreement state M prevents complete learning as the number of senders grows indefinitely. p. 0
Where the paper says it

Building on this property, we show that whether information aggregation and complete learning occur depends not only on the magnitude of the conflict of interest between the senders and the receiver but more fundamentally on the nature of that conflict. In particular, the presence of a disagreement state, in which the senders and the receiver have opposing policy preferences, precludes full information aggregation and complete learning even as the number of senders grows without bound.

A key finding is a discontinuity relative to binary benchmarks: for certain parameter ranges where aggregation occurs in the binary case, assigning any positive probability to the disagreement state causes communication to unravel as population size increases sufficiently. p. 4

Where the paper says it

for which information is aggregated in the binary benchmark, assigning the disagreement state with any positive probability causes communication to unravel once the population becomes sufficiently large. Hence, the binary benchmark’s prediction of informative communication is not robust to conflict that persists under full information.

This discontinuity stems from conflicting requirements on informativeness: uncertainty-based conflict demands enough information to resolve uncertainty, while full-information conflict makes highly informative communication unsustainable. p. 4
Where the paper says it

As conflict that arises only under uncertainty intensifies, communication must become sufficiently informative to resolve uncertainty and mitigate the effects of the conflict. By contrast, conflict that persists even under full information renders highly informative communication unsustainable: even if the disagreement state has only a tiny positive probability, such communication would reveal it at some approval tallies, at which the receiver and the senders prefer opposing policies.

The paper introduces a mediator who can enhance transmission and restore efficiency by committing ex ante to a mechanism that collects messages and recommends an outcome. p. 4

Where the paper says it

Since cheap-talk communication is subject to sharp limitations, we ask whether introducing a mediator with commitment power can improve information transmission and restore efficiency. The mediator does not observe the senders’ signals but can commit to a communication mechanism that collects their messages and provides a recommendation to the receiver.

It demonstrates that as the number of senders grows large, simple step mechanisms can asymptotically achieve any outcome on the Pareto frontier between the receiver’s first best and the senders’ optimal Bayesian-persuasion outcome. p. 4
Where the paper says it

The mediator can be interpreted as a petition platform or advocacy organization in petitions and protests, a shareholder representative in nonbinding shareholder voting, or a survey platform, polling firm, or statistical agency in polls and surveys. We show that, as the number of senders grows large, simple and intuitive step mechanisms can asymptotically implement any outcome on the Pareto frontier between the receiver’s first best and the senders’ optimal Bayesian-persuasion outcome.

The central conclusion is that in cheap-talk communication, information fails to aggregate, and communication may unravel entirely. p. 0

Where the paper says it

Communication as Voting Kailin Chen∗ September 16, 2026 Abstract This paper studies information aggregation in informal elections by examining a cheap-talk model with multiple senders and one receiver. Each sender observes a noisy private signal about an unknown state and sends a message; the receiver observes the message tally and chooses the final policy choice.

However, mediated communication can asymptotically implement any Pareto-efficient outcome on the frontier between the receiver’s first best and the senders’ optimal Bayesian-persuasion outcome. p. 4
Where the paper says it

The mediator can be interpreted as a petition platform or advocacy organization in petitions and protests, a shareholder representative in nonbinding shareholder voting, or a survey platform, polling firm, or statistical agency in polls and surveys. We show that, as the number of senders grows large, simple and intuitive step mechanisms can asymptotically implement any outcome on the Pareto frontier between the receiver’s first best and the senders’ optimal Bayesian-persuasion outcome.

The centralized benchmark suggests that under certain conditions (when conflict persists under full information), transmission fails in a single fully informed sender setting, but it persists in the decentralized setting with dispersed information. p. 3
Where the paper says it

Information aggregation and complete learning fail because fully revealing communication would violate the equilibrium no-conflict property: full revelation would make some approval tallies identify the disagreement state, while at such tallies, the receiver and the senders would prefer opposing policies. Intuitively, when conflict persists under full information, sufficiently informative communication is no longer harmless to the senders: it may enable the receiver to choose a policy unfavorable to them, thereby creating incentives for the senders to misrepresent their information and undermining information transmission.

The final remark highlights that decentralizing information across a crowd can outperform centralization in specific scenarios. p. 1

Where the paper says it

1 Introduction Information is often dispersed across a population. Formal elections can aggregate such information effectively.

Centralization corresponds to one sender observing multiple signals, which amplifies adverse selection and reduces receiver trust when the conflict of interest is large. p. 4
Where the paper says it

We show that, when the number of senders is sufficiently large, the senders endogenously pool all but one signal realization on a common message, so that these equilibria effectively reduce to equilibria with binary signals and binary messages. Our proof develops a novel large-deviations approach that may be of independent interest to theorists studying communication and voting.

Conversely, in the multi-sender model with dispersed information, noise created by failed aggregation can act as a commitment device to mitigate conflict and improve communication, as shown by Blume et al. p. 3
Where the paper says it

Hence, the conflict that persists under full information places an endogenous bound on the informativeness of equilibrium communication. Having shown that the disagreement state precludes complete learning and full information aggregation, we next characterize when informative communication can nevertheless be sustained and when communication unravels completely.

(2007).

Information transmission persistence depends on both sources of conflict—captured by US(L)US(H)UR(H)UR(L) and q 0M —being small. p. 7

Where the paper says it

This conflict is captured by q 0 M, the prior probability. Note that the first form of conflict arises only under uncertainty, whereas the second persists even under full information.

The presence of the disagreement state M sharply alters these conditions, creating a discontinuity relative to the binary benchmark. p. 3
Where the paper says it

We show that the disagreement state fundamentally alters the conditions for sustaining informative communication. In particular, the presence of the disagreement state generates a discontinuity relative to the binary benchmark of LMB.

Specifically, if q 0M > 0, no informative equilibrium exists for sufficiently large N, and babbling becomes the unique outcome. p. 4
Where the paper says it

for which information is aggregated in the binary benchmark, assigning the disagreement state with any positive probability causes communication to unravel once the population becomes sufficiently large. Hence, the binary benchmark’s prediction of informative communication is not robust to conflict that persists under full information.

A threshold q hat exists such that transmission persists if q 0M q hat, where q hat depends on the conflict ratio. p. 7
Where the paper says it

The first conflict arises from different payoff magnitudes in the agreement states L and H, in which the preferences align. This conflict is captured by the ratio of the senders’ threshold of doubt to the receiver’s threshold of doubt, −US(L)/US(H) −UR(L)/UR(H) = US(L) US(H) · UR(H) UR(L) ≥ 1.

The analysis also establishes limitations on sender welfare, showing a finite population size N* beyond which increasing the number of senders does not improve sender welfare. p. 4

Where the paper says it

When the first form of conflict reaches an intermediate level, the opposing requirements from two forms of conflict become incompatible, generating a discontinuity between zero and any positive probability of the disagreement state. Conflict that persists under full information places a further limit on sender welfare: in the presence of the disagreement state, sender welfare is maximized at a finite population size, so adding further senders—and hence further private signals—cannot raise sender welfare above the level already attained.

Furthermore, informative equilibria generated under higher-conflict parameters imply existence under lower-conflict parameters with weakly larger x. p. 8
Where the paper says it

4 2The cutoff Tˆ = 0 means that the receiver implements the proposal for every approval tally, while Tˆ = N + 1 means that she always maintains the status quo. 3The existence of informative equilibria does not depend on the receiver randomizing at Tˆ.

The model is extended to nonbinary signal and message spaces, showing that in monotone equilibria, the signal space is effectively binary. p. 4
Where the paper says it

We extend the preceding results to a natural class of equilibria in which the senders use monotone strategies, in settings with arbitrary numbers of signals and messages. We show that, when the number of senders is sufficiently large, the senders endogenously pool all but one signal realization on a common message, so that these equilibria effectively reduce to equilibria with binary signals and binary messages.

The paper concludes by confirming that mediated communication can asymptotically implement any allocation on the Pareto frontier using a binary voting mechanism and randomized qualified-majority rules. p. 4

Where the paper says it

The mediator can be interpreted as a petition platform or advocacy organization in petitions and protests, a shareholder representative in nonbinding shareholder voting, or a survey platform, polling firm, or statistical agency in polls and surveys. We show that, as the number of senders grows large, simple and intuitive step mechanisms can asymptotically implement any outcome on the Pareto frontier between the receiver’s first best and the senders’ optimal Bayesian-persuasion outcome.

In the multiple-sender environment, a vanishing elicitation cost is compatible with strictly positive commitment. p. 9
Where the paper says it

[ρθxg + (1 − ρθ)xb | {z } Prob of approving ] Tˆ−1 [1 − ρθxg − (1 − ρθ)xb | {z } Prob of rejecting ] N−Tˆ . When a sender receives s ∈ {g, b}, he strictly prefers approving to rejecting if his expected payoff from the proposal, conditional on his signal and on being pivotal, is positive, i.

The centralized case yields payoffs identical to the babbling equilibrium, whereas in the decentralized model with dispersed information, every player obtains a weakly higher expected payoff in any informative equilibrium than in the centralized benchmark. The receiver's best response is always a cutoff rule limiting transmitted information.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Develop a mechanism to dynamically adjust thresholds based on conflict levels to optimize information transmission efficiency across different political or social contexts. This system could implement a cutoff strategy Tˆ that is adaptive, potentially moving between various fixed cutoffs depending on whether the disagreement state M is assigned a positive probability, as suggested by the discontinuity analysis in Theorem 4.

  2. Implement an information aggregation module designed to handle noisy private signals from multiple agents while explicitly modeling the conflict of interest between senders and receivers. This system would be able to determine when communication unravels completely due to persistent disagreement states, allowing it to flag situations where aggregating more information is counterproductive for policy-making.

  3. Create a mediated communication layer that asymptotically implements any Pareto efficient outcome on the frontier between the receiver's first best and the senders' optimal Bayesian persuasion outcome. This system would be capable of generating state-contingent recommendations in a disagreement state, allowing it to achieve better outcomes than direct cheap-talk communication when conflict is present.

  4. Design a large population scaling protocol that informs decision-making on the diminishing returns of increasing the number of senders for private information aggregation. This protocol would quantify when adding more participants leads to vanishingly small individual message informativeness, suggesting limits on how much decentralized information can improve policy outcomes as the population grows indefinitely.

  5. Develop a signal processing pipeline capable of detecting pivotal events in message counts that are exponentially less likely in high-state scenarios compared to low or medium states. This capability would allow the system to prioritize decision-making based on which state is most likely, effectively using the structure of pivotal events to infer the underlying unknown state with greater precision than simple aggregate tallies.

Abstract

This paper studies information aggregation in informal elections by examining a cheap-talk model with multiple senders and one receiver. Each sender observes a noisy private signal about the unknown state and sends a message; the receiver observes the message tally and chooses the final policy. We show that any positive probability of a disagreement state, in which the senders and the receiver have opposing policy preferences, precludes full information aggregation even as the number of senders grows without bound. The key observation is that, in equilibrium, communication can reveal only information that induces the same preferred policy for both the receiver and the senders. Furthermore, there is a discontinuity in the scope for information transmission: for an open set of parameter values, existing models permit full information aggregation, but perturbing them by introducing an arbitrarily small positive probability of a disagreement state leads to no information transmission once the sender population is sufficiently large. Finally, introducing a mediator can improve information transmission and restore efficiency.

Sources

Related papers