The amplification of risk in experimental diffusion chains

Understanding how people form and revise their perception of risk is central to designing efficient risk communication methods, eliciting risk awareness, and avoiding unnecessary anxiety among the public. However, public responses to hazardous events such as climate change, contagious outbreaks, and terrorist threats are complex and difficult-to-anticipate phenomena. Although many psychological factors influencing risk perception have been identified in the past, it remains unclear how perceptions of risk change when propagated from one person to another and what impact the repeated social transmission of perceived risk has at the population scale. Here, we study the social dynamics of risk perception by analyzing how messages detailing the benefits and harms of a controversial antibacterial agent undergo change when passed from one person to the next in 10-subject experimental diffusion chains. Our analyses show that when messages are propagated through the diffusion chains, they tend to become shorter, gradually inaccurate, and increasingly dissimilar between chains. In contrast, the perception of risk is propagated with higher fidelity due to participants manipulating messages to fit their preconceptions, thereby influencing the judgments of subsequent participants. Computer simulations implementing this simple influence mechanism show that small judgment biases tend to become more extreme, even when the injected message contradicts preconceived risk judgments. Our results provide quantitative insights into the social amplification of risk perception, and can help policy makers better anticipate and manage the public response to emerging threats.

💡 Research Summary

The paper investigates how risk perception evolves when information about a controversial antibacterial agent is passed from person to person in experimental diffusion chains. Ten independent chains, each consisting of ten participants, were created. The initial stimulus was a balanced document describing both the benefits (e.g., antibacterial efficacy) and harms (e.g., resistance development, environmental impact) of the agent. The first participant read the document, summarized it according to his/her own pre‑existing risk judgment, and passed the shortened version to the next participant. This process repeated until the tenth person in each chain received the message.

The authors measured four quantitative aspects of the transmitted text at every step: (1) length (word count), (2) preservation of core factual information, (3) factual accuracy relative to the original, and (4) similarity between chains. In parallel, participants completed a risk‑assessment questionnaire on a 0–100 scale before and after each transmission. Text analysis was performed with automated natural‑language‑processing pipelines, allowing precise tracking of how the message changed across generations.

Key empirical findings are as follows: (i) Message length systematically decreased, with an average reduction of more than 30 % after ten transmission steps; essential details, especially quantitative risk data, were most frequently omitted. (ii) Accuracy declined by roughly 20 % relative to the original document, reflecting both under‑ and over‑estimation of risk elements. (iii) Inter‑chain similarity fell dramatically—from a cosine similarity of about 0.85 at the start to roughly 0.45 after ten steps—demonstrating that identical source material diverges into markedly different final messages. (iv) In contrast, participants’ risk perception scores remained comparatively stable and showed higher fidelity to the original judgment. Participants tended to edit the message in a way that aligned with their own prior beliefs, a process the authors label “bias‑consistent editing.”

To explore the underlying dynamics, the authors built an agent‑based simulation model. Each agent is characterized by an initial risk bias, a probability of editing the incoming message, and a rule that preferentially retains or emphasizes information congruent with the bias. The model reproduces the empirical pattern that small initial biases (as low as 5 % deviation from a neutral stance) can be amplified to over 30 % after ten transmission rounds. Moreover, even when the incoming message contradicts an agent’s pre‑existing judgment, the editing process gradually reshapes the message to fit the bias, illustrating a robust amplification mechanism.

The discussion emphasizes the practical implications for risk communication policy. Simply issuing a balanced statement is insufficient; communicators must anticipate that the message will be truncated, distorted, and selectively reinforced as it spreads through social networks. Strategies such as reinforcing key facts, providing repeated official clarifications, and designing messages that are resilient to selective editing could mitigate the amplification of misperception. The authors acknowledge limitations: the sample consisted mainly of university students, and the experimental chains lack the complexity of real‑world networks (multiple pathways, feedback loops, and heterogeneous media). Future work should incorporate diverse demographic groups, real‑world social‑media data, and multi‑branch network structures to test whether the observed amplification holds across different risk domains (health, environmental, security).

In sum, this study offers the first experimental and computational quantification of the “social amplification of risk” phenomenon. It demonstrates that while factual content degrades and diverges during social transmission, the underlying risk perception can become more extreme due to bias‑consistent editing. These insights provide a scientific foundation for designing more effective risk communication strategies that account for the dynamics of information propagation in society.