Communicating with Anecdotes

Nika Haghtalab, Nicole Immorlica, Brendan Lucier, Markus Mobius, Divyarthi Mohan

Abstract

In many real-world scenarios, experts communicate not through full datasets, but through selected "anecdotes"—individual data points meant to represent a broader reality. This paper models such communication as a sender-receiver game where a biased sender chooses one anecdote from a sample to influence a receiver. We identify a "curse of informedness": as a sender's expertise grows, the temptation to cherry-pick extreme outliers to persuade the receiver can degrade the precision and trustworthiness of the communication, particularly in heavy-tailed distributions.

Introduction

Formal communication often relies on anecdotes—factual snippets or personal experiences—rather than exhaustive data. However, a fundamental tension exists when a sender selects which story to tell. On one hand, the sender wants to inform the receiver so they take a generally appropriate action (the informative motive). On the other hand, the sender may have personal preferences that differ from the receiver's, creating a temptation to nudge the receiver toward a specific outcome (the persuasion temptation). This paper investigates how this selective reporting, or "cherry-picking," affects the precision of communication and why experts sometimes fail to be more convincing than laypeople.

Methods: The Anecdote Game

The authors formalize this interaction as a three-round sender-receiver game.

Timeline of the sender-receiver anecdote game.
  1. Round 0 (Nature): Nature chooses a state $\theta$ and generates $n$ independent anecdotes $\vec{x}$ and $y$ from a distribution centered at $\theta$.
  2. Round 1 (Sender): The sender observes all $n$ anecdotes but can only transmit a single anecdote $\tilde{x}$ to the receiver. The sender cannot lie; $\tilde{x}$ must be one of the observed data points.
  3. Round 2 (Receiver): The receiver observes $\tilde{x}$ and takes an action $a$. Both players have quadratic loss functions, but their ideal actions are shifted by constants $M_S$ and $M_R$. The misalignment $\Delta = M_R - M_S$ drives the strategic behavior.

The analysis focuses on Perfect Bayesian Equilibria, specifically "translation invariant" strategies where the receiver's response is a fixed shift from the observed anecdote.

Key Insights: Targeting and the Curse of Informedness

In equilibrium, the sender does not choose the most representative anecdote (the one closest to the mean). Instead, they employ a "targeting" strategy, selecting the anecdote closest to a specific target offset designed to pull the receiver's action toward the sender's ideal.

The targeting equilibrium: the sender selects an anecdote closest to a target offset to influence the receiver.

Informational Homophily

The model provides a rational basis for echo chambers. Receivers prefer senders with similar biases (low $\Delta$) because these senders face less "persuasion temptation." Consequently, like-minded senders choose more representative anecdotes, leading to higher-precision communication.

The Curse of Informedness

Counter-intuitively, more information can degrade credibility. As a sender becomes an "expert" (larger $n$), they have a higher probability of finding extreme outliers in the tails of the distribution. If the underlying data is heavy-tailed (e.g., Pareto), the variance of these chosen outliers grows faster than the benefit of having more samples. In such cases, a receiver actually learns less from an expert than from a less-informed sender because the expert's ability to cherry-pick extreme evidence is too great.

Results and Welfare

The authors prove that a translation-invariant equilibrium always exists if the anecdote distribution has a bounded mean. While more data eventually improves communication for light-tailed distributions (like Gaussian), the precision of the signal is permanently degraded by the sender's selective reporting. Misalignment always results in a loss of "communication fidelity," as the receiver must rationally "debias" the anecdote, which increases the noise of the final estimate.

Conclusion and Future Directions

The "single anecdote" constraint captures the core tension of selective reporting. While sharing multiple anecdotes might reduce information loss, the fundamental drive toward polarization remains. Future research could explore cognitive limitations in memory storage (treating anecdotes as internal representations) or the role of "framing," where senders add spin to facts rather than just selecting them.

Summary updated

Abstract

We study a communication game between a sender and a receiver. The sender chooses one of her signals about the state of the world (i.e., anecdotes) and communicates to the receiver who takes an action affecting both players. The sender and the receiver both care about the state of the world but are also influenced by personal preferences, so their ideal actions can differ. We characterize perfect Bayesian equilibria. The sender faces a temptation to persuade: she wants to select a biased anecdote to influence the receiver's action. Anecdotes are still informative to the receiver (who will debias at equilibrium) but the attempt to persuade comes at a cost to precision. This gives rise to informational homophily where the receiver prefers to listen to like-minded senders because they provide higher-precision signals. Communication becomes polarized when the sender is an expert with access to many signals, with the sender choosing extreme outlier anecdotes at equilibrium (unless preferences are perfectly aligned). This polarization dissipates all gains from communication with an increasingly well-informed sender when the anecdote distribution is heavy-tailed. Experts can therefore face a curse of informedness: receivers will prefer to listen to less-informed senders who cannot pick biased signals as easily.
BibTeX
@misc{haghtalab2022communicating,
  author = {Nika Haghtalab and Nicole Immorlica and Brendan Lucier and Markus Mobius and Divyarthi Mohan},
  title = {{Communicating with Anecdotes}},
  year = {2022},
  eprint = {2205.13461},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2205.13461},
}
Figure