Source-linked AI summary
The amplification of risk in experimental diffusion chains
Mehdi Moussaid, Henry Brighton, Wolfgang Gaissmaier
TL;DR
The paper examines how risk perception changes as people transmit risk information, a process that may shape population-level patterns. Using experimental transmission of Triclosan risk information, it finds that distorted messages can strengthen existing group biases even when the message initially supports the opposite view.
Problem
How social transmission changes risk perception and shapes population-level dynamics remains an open question, despite research largely focused on individuals.
Source: Introduction · pp. 3–4Method
The authors experimentally examine how people transmit risk information about Triclosan and how human subjects influence its propagation.
Source: Introduction · p. 4, Discussion · p. 10Results
After a few rounds of distortion, a message can strengthen a group’s existing bias even when it initially supported the opposite view.
Source: Discussion · p. 11Takeaways & Limitations
Human transmission of risk information can shape the propagation of that information and the development of collective risk-perception patterns.
Source: Discussion · p. 10Variation within and between chains made a systematically controlled analysis difficult.
Source: Results · p. 8
Abstract
from arXiv · showhide
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.
Significance Statement
Risk judgments are socially transmitted, but messages about risk lose fidelity as they pass between people. Experimental diffusion chains show that participants’ risk perceptions can shape what they communicate and influence subsequent judgments.
Significance Statement
- The study addresses how social transmission shapes risk perception and the population-level patterns it may produce, a question largely overlooked by individual-focused research.The authors use diffusion chains to examine how message content and signal propagate.Source: Introduction · pp. 3–4
- Participants communicated information about Triclosan’s benefits and harms through open conversations in 15 experimental chains of up to 10 people each.The first participant received media-article information, and risk perceptions were assessed before and after participation.Source: Introduction · p. 4
- The authors use an experimental diffusion-chain paradigm to examine how risk information and people’s risk perceptions change during social transmission.They describe this as the first experimental study of these changes under social transmission.Source: Introduction · p. 5
- Risk signals propagate with high fidelity and are typically amplified, while message content becomes shorter, gradually inaccurate, and increasingly dissimilar between chains.Participants bias the signal according to their subjective risk perceptions, influencing receivers’ judgments.Source: Introduction · p. 5
Results
The results examine how messages and risk judgments change as they pass through diffusion chains, alongside simulations of repeated transmission and its effects on risk signals and judgments.
Message content
- Most information units disappear as chains unfold, while those transmitted persist with low fidelity and become increasingly distorted.In one illustrative chain, 13 of 30 units reached the second position and only 3 reached the final position; all three were distorted.Source: Results · p. 5
- Information-unit death is high at the first two positions, then stabilizes near 0.2; birth and distortion probabilities remain largely constant across positions and coding categories.Source: Results · p. 6
- 0.53 at position 1 rose to 0.87 at position 10 in average message differentiation, indicating increasing dissimilarity between chains.Source: Results · p. 6
Message signal
- Negative statements decay more slowly than positive ones, so their relative share increases as messages propagate, at the expense of positive statements.Source: Results · p. 7
- Participants tended overall to amplify negative statements and attenuate positive ones, with filtering patterns varying by individual and corresponding partly to reported risk perception.Higher-risk-perception participants tended to filter out positive statements and amplify Triclosan’s dangerous aspects; lower-risk-perception participants tended toward the opposite pattern.Source: Results · p. 7
- 0.19 (STD=0.17) was the absolute average change in participants’ risk perception after the experiment.Source: Results · p. 8
Collective patterns
- Risk-perception change correlated with the difference between incoming message signal and prior assessment (c=0.55, p-value<0.001), while judgments shaped the outgoing signal and amplified risk.The authors note that fluctuations within and between chains made systematic experimental analysis of amplification difficult.Source: Results · p. 8
Results
- The experimental influence estimate was s=0.45, consistent with prior empirical estimates of 0.3–0.5 (37, 38).Source: Results · pp. 8–9
- In simulations with varied initial perceptions, chain-position-10 distributions corresponded to experimental risk-signal and risk-perception distributions (KS tests p=0.57 and p=0.08), while final perceptions remained variable.The model’s gradual signal amplification coexisted with variability attributable to diverse initial judgments.Source: Results · p. 9
- With a neutral message entering a concerned population, it rapidly mutated toward the population’s initial view and had no long-run impact on individuals’ opinions.Source: Results · p. 9
- In the model’s parameter-space exploration, social amplification of the risk signal was visible after 10 transmissions and continued to increase as the message propagated further.Source: Results · p. 9
- Even a neutral message in a neutral population could polarize judgments through amplified random mutations, yielding high-or-low outcomes despite mean signal and perception values of 0.5.The high standard deviations indicated a U-shaped outcome distribution.Source: Results · p. 10
Discussion
In these experimental diffusion chains, message content degraded and diverged, while participants’ subjective signal shaped risk judgments and could amplify group biases. The study also identifies open questions about how complex networks and different information sources affect risk propagation.
Discussion
- The study experimentally examined social transmission of risk information and supports treating risk-perception patterns, at least partly, as social outcomes.The authors relate this contribution to opinion dynamics and empirical support for models of collective risk perception.Source: Discussion · pp. 10–11
- Transmitted messages contained fewer information units, became less accurate, and led different chains to focus on different Triclosan issues.Examples included Greenpeace protests, breast-feeding, and environmental damage.Source: Discussion · p. 10
- Participants altered message signals toward their initial judgments, and receivers’ risk-perception changes depended more on the signal than on message content.The signal reflected whether the message was positive or negative.Source: Discussion · p. 10
- Social transmission amplified or attenuated the signal until the original message had negligible influence, sometimes strengthening group bias despite initially opposing it.The simulations also indicate stronger amplification in chains of like-minded individuals.Source: Discussion · pp. 10–11
- How highly connected individuals and private, public, and social information shape risk-perception dynamics remains open.Source: Discussion · p. 11
Material and Methods
- Across 15 sessions, participants discussed Triclosan sequentially after the first participant read six articles; chains averaged 10.1 participants.Discussions were open-ended, recorded, transcribed, and translated from German.Source: Material and Methods · pp. 11–12
- Risk perception was measured indirectly because participants were unfamiliar with Triclosan, while transcripts were coded for information content, distortion, and positive, negative, or neutral signal.Two independent coders labeled signal sentences, and their results were averaged.Source: Material and Methods · p. 12
Figures
The figures trace how message content changes across experimental diffusion chains and how risk signals and judgments evolve in simulations. They show increasing differences between chains and risk-signal amplification under social influence.
Figure 1
- Figure 1 maps information units through a chain: three units present at the start survive to the end, and seven are introduced, of which two survive.Color indicates cumulated distortion; information IDs on the y-axis are arbitrary.Source: Figure 1 · p. 18
Figure 2
- Message differentiation shifts toward higher values as information propagates, while the figures track information loss, distortion, and appearance across chain positions.Figure 2C compares message pairs at positions 1, 3, and 10; Figure 2B shows the probabilities of disappearance, distortion, and appearance.Source: Figure 2 · p. 19
Figure 3
- Figure 3 tracks normalized positive, negative, and total statement counts across the chain, as well as the proportion of positive statements and filtering-coefficient patterns.The figure includes coefficient distributions and individual profiles.Source: Figure 3 · p. 20, Figures · p. 20
Figure 4
- Participants tend to raise risk judgments when the received message is negative relative to their initial perception and lower them when it is more positive.The fitted relationship is y=0.45x−0.14.Source: Figure 4 · p. 21
Figure 5
- Simulations show risk signals becoming increasingly extreme across 50 transmissions, while individuals with initially extreme judgments move back toward their initial views.Figure 5 uses a social influence parameter of s=0.5 and averages results over 500 simulations.Source: Figure 5 · p. 22
Figure S2
- The supplementary hazard-function comparison finds no specific difference among four information categories in the probabilities of disappearance, appearance, or distortion.The figure suggests these events can occur regardless of the nature of the information communicated.Source: Figure S2 · p. 25
Figure S3
- Simulations contrast heterogeneous and like-minded chains with spontaneous amplification from a neutral message in a neutral chain.The like-minded-chain case has exacerbated amplification; the spontaneous case is attributed to amplification of random signal mutations.Source: Figure S3 · p. 26
Figure S4
- Varying social influence from s=0 to s=1 changes the simulated system from no influence to full adherence to the message, with the transition zone becoming sharper as s increases.At s=1, messages below a signal of 0.5 converge to 0 and those above 0.5 converge to 1.Source: Figure S4 · p. 27