Source-linked AI summary
Social Influence and the Collective Dynamics of Opinion Formation
Mehdi Moussaid, Juliane E. Kaemmer, Pantelis P. Analytis, Hansjoerg Neth
TL;DR
The mechanisms of opinion formation remain poorly understood, and local rules of opinion adaptation had not yet been used to study collective dynamics. The paper combines controlled experiments with quantitative descriptions to measure social influence and examine collective opinion. It reports a critical amount of approximately 15% of experts, while participants showed bias toward their own opinions and typically underestimated contradictory feedback.
Problem
The mechanisms of opinion formation remain poorly understood, and local rules of opinion adaptation had not yet been used to study collective dynamics.
Method
The work uses controlled experiments to measure the effects of social influence and provides experimental measurements with quantitative descriptions.
Results
Approximately 15% of experts constituted a critical amount, while participants exhibited bias toward their own opinions and typically underestimated contradictory feedback.
Takeaways & Limitations
The approach addresses social influence as a key element in the formation of public opinions.
Takeaways & Limitations
The simulations require empirical validation in the future, and the assumption concerning the observed decision tree remains to be verified.
Abstract
from arXiv · showhide
Social influence is the process by which individuals adapt their opinion, revise their beliefs, or change their behavior as a result of social interactions with other people. In our strongly interconnected society, social influence plays a prominent role in many self-organized phenomena such as herding in cultural markets, the spread of ideas and innovations, and the amplification of fears during epidemics. Yet, the mechanisms of opinion formation remain poorly understood, and existing physics-based models lack systematic empirical validation. Here, we report two controlled experiments showing how participants answering factual questions revise their initial judgments after being exposed to the opinion and confidence level of others. Based on the observation of 59 experimental subjects exposed to peer-opinion for 15 different items, we draw an influence map that describes the strength of peer influence during interactions. A simple process model derived from our observations demonstrates how opinions in a group of interacting people can converge or split over repeated interactions. In particular, we identify two major attractors of opinion: (i) the expert effect, induced by the presence of a highly confident individual in the group, and (ii) the majority effect, caused by the presence of a critical mass of laypeople sharing similar opinions. Additional simulations reveal the existence of a tipping point at which one attractor will dominate over the other, driving collective opinion in a given direction. These findings have implications for understanding the mechanisms of public opinion formation and managing conflicting situations in which self-confident and better informed minorities challenge the views of a large uninformed majority.
Introduction
The paper addresses how people adapt opinions during social interactions and how local influence generates collective patterns. It combines controlled experiments with an individual-based model to identify opinion attractors and their competition.
- Research gap: Social influence mechanisms remain poorly understood, including how people adjust judgments, what heuristics guide adaptation, and how local changes produce global patterns.Existing opinion-dynamics models often lack empirical verification of their assumptions.
- Approach: The study uses controlled experiments to measure how individuals revise initial beliefs after exposure to another person’s opinion and confidence.A second session involved 59 participants answering 15 questions, revising answers after peer feedback, and reporting confidence before and after exposure.
- Approach: An individual-based model derived from the observations examines how repeated social interactions generate collective opinion dynamics.The model links experimentally measured micro-level influence mechanisms to group-level outcomes.
- Main findings: Collective opinion is shaped by two attractors: a highly confident individual and clusters of low-confidence individuals sharing a similar opinion.The first corresponds to an expert effect, while the second produces a majority effect through shared lay opinions.
- Main findings: Approximately 15% experts are needed to counteract the attractive effect of a large majority of lay individuals.This result identifies a critical amount at which the expert and majority attractors compete.
- Significance: The findings provide a first step toward understanding how social influence propagates, reinforces, or polarizes ideas and attitudes in modern societies.The model demonstrates that repeated local interactions can produce consensus, polarization, or fragmentation.
Results
Participants’ responses revealed structured social influence: feedback effects depended on opinion and confidence differences, while repeated interactions could produce convergence around majority or highly confident individuals.
- Empirical results: The initial opinion distribution had a lognormal shape, and normality tests supported this pattern for most items.The null hypothesis could not be rejected at the 5% level for 84% of items; the remaining 16% still had p-values above 10^-3.
- Empirical results: Confidence was an imperfect accuracy cue: only Ci=6 reliably indicated good or very good estimates, whereas lower levels were less informative.Ci=6 corresponded to a good or very good estimate 80% of the time; Ci=5 accompanied a bad or very bad estimate 39% of the time, and Ci=4 did so 53% of the time.
- Empirical results: Social influence reduced the reliability of highly confident judgments by making their post-influence errors noisier, more widespread, and less informative about accuracy.This effect was observed for very confident individuals with Ci=5 or 6.
- Process model: Across observations, 53% of revisions kept the initial opinion, 43% were compromises, and 4% adopted the other opinion.The average compromise weight was ω=0.4 (SD=0.24), indicating a bias toward the initial estimate.
- Collective dynamics: Repeated simulations produced a tipping point: a critical mass sharing an opinion increased confidence there, creating reinforcement and majority convergence despite residual fragmentation.The simulations also identified attractors associated with a critical mass of uncertain individuals and one or a few highly confident individuals.
Discussion
The study combines controlled experiments with simulations to quantify social influence and explain how confidence, opinion similarity, experts, majorities, and neutral individuals shape collective opinion dynamics. It identifies mechanisms that can stabilize, amplify, or redirect group opinions, while highlighting assumptions and empirical-validation needs.
- Method: The study combines controlled experiments measuring social influence with computer simulations that scale individual behavior to collective opinion dynamics.The simulations help characterize large-group mechanisms that would be difficult to measure under laboratory conditions.
- Individual influence: Participants preferentially retained their initial opinions, while corroborating feedback increased confidence and contradictory feedback was often underestimated or ignored.The influence map shows “keep initial opinion” as dominant, with stronger weighting of participants’ own initial opinions during compromise.
- Individual influence: Opinion differences constrain influence: people holding completely different beliefs exert very little influence, consistent with bounded-confidence models.Influence diminishes progressively as opinions become more distant, and remote opinions may be ignored entirely.
- Collective dynamics: Confidence acts as system memory: repeated feedback makes individuals less easily influenced and progressively moves their opinions toward stable values.Increasing confidence also helps create basins of attraction by strengthening the influence of people sharing similar opinions.
- Implications: Social influence can make high confidence signal consensus rather than accuracy, while artificial confidence growth may trigger snowball effects that drive groups in erroneous directions.High confidence predicts accuracy before social influence but loses that reliability after exposure to others’ opinions.
- Collective dynamics: The simulations identify two competing attractors: an expert effect from highly confident individuals and a majority effect from a critical mass of low-confidence people sharing opinions.Neutral individuals around these attractors increase unpredictability and reduce the crowd’s vulnerability to either force.
- Implications: The framework may inform management of conflicts in which a small opinionated minority challenges a large uninformed population, but its simulations still require empirical validation.Future work should also test whether the experimentally derived decision tree remains unchanged over repeated interactions and examine emotional or subjective beliefs.
Materials and Methods
The study used two controlled laboratory experiments to compare factual judgments without social influence and after exposure to another participant’s estimate and confidence. It tracked opinion and confidence changes across 52 participants answering 32 questions and 59 participants completing 885 social-influence interactions on 15 questions.
- Experimental design: Two experiments compared factual-question responses without social influence and with feedback from another participant.Participants entered the laboratory individually and answered questions on a computer under otherwise similar conditions.
- Experiment 1: 52 participants answered 32 general-knowledge questions in Experiment 1, which established the initial answer and confidence distributions.The resulting 1,664 data points were also used as a pool of social influence and to define simulation initial conditions.
- Task and measures: Participants gave spontaneous numerical estimates and then rated confidence on a 6-point Likert scale from 1, very unsure, to 6, very sure.Questions covered sports, nature, geography, and society/economy, with correct answers ranging from 100 to 999.
- Analysis and controls: The researchers assessed answer accuracy, confidence, and possible learning effects while withholding correct answers until the experiments ended.Randomized question order and nonsignificant accuracy-order correlations for 90% of participants supported the absence of a learning process over the session.
Supporting Information Legends
The supporting information documents the answer distributions for the first experiment and the question set used in the study. It also illustrates confidence amplification across simulation rounds.
- Figure S1: Figure S1 shows answer distributions for all 32 questions from Experiment 1, using normalized estimates divided by the true value.Dashed lines mark the correct answer at normalized value 1, the mean, and the median; question 27 is used in Figure 1A.
- Figure S2: Figure S2 presents three representative examples of confidence evolution over simulation rounds for the 52 Experiment 1 participants.After a few rounds, average global confidence shows a sharp transition toward high confidence levels.
- Table S1: Table S1 provides the full list of questions used in the study.