Collective communication and behaviour in response to uncertain ‘Danger’ in network experiments

Hirokazu Shirado, Forrest W. Crawford, Nicholas A. ChristakisProceedings of the Royal Society A Mathematical Physical and Engineering Sciences 2020 · DOI

TL;DR

This study uses large-scale network experiments to investigate how communication networks affect group behavior during uncertain emergencies. They found that social networks, while good at preventing unnecessary actions in safe situations, paradoxically hinder necessary evacuations during actual dangers due to the rapid spread of false reassurance.

What Problem Does This Paper Address?

Coordinating collective action during emergencies is notoriously difficult. While social networks are often seen as efficient channels for information, their role in high-stakes, uncertain situations like disasters is complex. This paper addresses the question of whether people connected in networks fare better or worse than isolated individuals when faced with an unpredictable threat. Specifically, it explores how interpersonal communication within these networks influences decisions to evacuate, and whether network structure itself plays a role. The authors highlight that in real-world disasters, human response can sometimes be more damaging than the disaster itself, underscoring the importance of understanding these coordination dynamics.

Methods

The researchers conducted a series of laboratory experiments involving a simulated emergency scenario. They recruited a large number of participants (2480 in the main experiment, with supplementary experiments bringing the total to 4348) who were tasked with deciding whether to "evacuate" from a potential "disaster."

Key methodological aspects include:

  • Networked Experimentation: Participants were organized into groups of varying sizes (10, 20, 40, 60 nodes) and assigned to different network topologies (random-regular and ring-lattice). A control group of isolated individuals was also included.
  • Simulated Disaster: A 75-second game was designed where participants could choose to evacuate at a cost. A disaster would strike at the end of the game with a 50% probability.
  • Information Diffusion: In networked conditions, participants could send "safe" or "danger" signals to their immediate neighbors. One randomly selected participant in each group was designated as an "informant" with prior knowledge of whether the disaster would occur.
  • Payoff Structure: Participants received a small endowment and earned bonuses for correct decisions (evacuating when a disaster struck, staying when it didn't) and for positive externalities (other players making correct decisions). This incentivized accurate communication and decision-making.
  • Behavioral Modeling: Cox proportional hazards models were used to analyze the time to evacuation and the influence of neighbor signals.

Datasets

The study utilized data generated from their own experimental setup. No external datasets were used for training or evaluation. The experiments themselves served as the data-generating mechanism, creating a controlled environment to study collective behavior under uncertainty.

Key Insights and Ideas

  • The "False Reassurance" Effect: The most striking finding is that communication networks, while effective at preventing unnecessary actions (like evacuating when there's no danger), actively suppress necessary evacuations during actual dangers. This is driven by the rapid spread of "safe" signals, which often turn out to be false reassurance.
  • Inconvenient Truths are Ignored: Social networks can be poor conduits for bad news. People are more likely to amplify popular opinions and ignore unpopular warnings, even if those warnings are accurate.
  • Individual Thresholds Matter: Individuals' responses to social information are more sensitive to the negativity of signals (i.e., danger signals) than their actual accuracy.
  • Network Structure's Indirect Role: While network topology (like clustering) can influence the speed of information spread, its ultimate impact on individual behavior is largely mediated by the individual's exposure to signals from their immediate neighbors. Participants were unaware of the global network structure.
  • Social Reinforcement Overrides Individual Judgment: In networked conditions, individuals tend to rely more on the opinions of others than their own inclinations, even when those inclinations might be more accurate.

Key Results

  • Reduced Unnecessary Evacuations: In situations without a disaster, networked subjects were significantly less likely to evacuate than isolated individuals.
  • Increased Necessary Evacuations Suppressed: When a disaster did strike, networked subjects were less likely to evacuate than isolated individuals. This led to greater losses in networked groups.
  • "Safe" Signals Spread Faster: "Safe" signals diffused significantly faster than "danger" signals, regardless of their accuracy.
  • Networked Groups Suffered Greater Losses: In disaster scenarios, networked subjects lost an average of $3.43 per person compared to the maximum possible earnings, with $2.43 of this loss attributed to miscommunication.
  • Informant's Impact Limited: While informants could influence their immediate neighbors, their effect did not propagate strongly enough to significantly alter evacuation behavior in larger or more distant parts of the network.

Limitations and Future Work

The authors acknowledge several limitations and suggest avenues for future research:

  • Artificiality of the Experiment: The study uses a simplified game with small economic stakes, which may not perfectly replicate the high-stakes, complex realities of real-world disasters.
  • Limited Information: Participants were not informed about the global network structure, group size, or the number of informants, which might influence behavior in real-world scenarios.
  • Future Research Directions:
    • Investigating the impact of multiple informants, unequal weighting of information sources, and different types of social ties.
  • Exploring how different payoff structures might alter decision-making.
  • Examining the long-term responses of social systems to repeated exposures to uncertain dangers.
  • Further exploring the role of post-evacuation information availability and structural heterogeneity (e.g., networks with shortcuts).
  • Investigating the influence of individual psychology more deeply, beyond basic risk tolerance and trust.

Summary updated

Abstract

In emergencies, social coordination is especially challenging. People connected with each other may respond better or worse to an uncertain danger than isolated individuals. We performed experiments involving a novel scenario simulating an unpredictable situation faced by a group in which 2480 subjects in 108 groups had to both communicate information and decide whether to 'evacuate'. We manipulated the permissible sorts of interpersonal communication and varied group topology and size. Compared to groups of isolated individuals, we find that communication networks suppress necessary evacuations because of the spontaneous and diffuse emergence of false reassurance; yet, communication networks also restrain unnecessary evacuations in situations without disasters. At the individual level, subjects have thresholds for responding to social information that are sensitive to the negativity, but not the actual accuracy, of the signals being transmitted. Social networks can function poorly as pathways for inconvenient truths that people would rather ignore.
BibTeX
@inproceedings{shirado2020collective,
  author = {Hirokazu Shirado and Forrest W. Crawford and Nicholas A. Christakis},
  title = {{Collective communication and behaviour in response to uncertain ‘Danger’ in network experiments}},
  booktitle = {Proceedings of the Royal Society A Mathematical Physical and Engineering Sciences},
  year = {2020},
  doi = {10.1098/rspa.2019.0685},
}
Figure