Key Publications

Realtime user ratings as a strategy for combatting misinformation: an experimental study

Jonas Stein, Vincenz Frey & Arnout van de Rijt 

Because fact-checking takes time, verdicts are usually reached after a message has gone viral and interventions can have only limited effect. A new approach recently proposed in scholarship and piloted on online platforms is to harness the wisdom of the crowd by enabling recipients of an online message to attach veracity assessments to it. The intention is to allow poor initial crowd reception to temper belief in and further spread of misinformation. We study this approach by letting 4000 subjects in 80 experimental bipartisan communities sequentially rate the veracity of informational messages. We find that in well-mixed communities, the public display of earlier veracity ratings indeed enhances the correct classification of true and false messages by subsequent users. However, crowd intelligence backfires when false information is sequentially rated in ideologically segregated communities. This happens because early raters’ ideological bias, which is aligned with a message, influences later raters’ assessments away from the truth. These results suggest that network segregation poses an important problem for community misinformation detection systems that must be accounted for in the design of such systems.
Figure 2. Theoretical expectations. We simulate sequences of rating decisions in 30,000 groups of bipartisan agents (25 ideologically aligned and 25 misaligned agents per group). The fraction of correct rating decisions is shown (A) as a function of difficulty and (B) as a function of an agents’ position in the rating sequence. In integrated groups, agents classify messages more often correctly than those in independent groups when average difficulty d < 0.5, irrespective of message veracity. In segregated groups, agents classify messages more often correctly than those in independent groups if true messages are being rated but classify messages less often correctly if false messages are rated. Parameters used in all panels: | dalign—dmis |=b=0.2, v =[−1; 1], s = [0; 2.35]. Panel B: d = 0.45.

Agent-based computational models

Andreas Flache and Carlos de Matos Fernandes

Agent-based computational modeling (ABCM)  plays  a  central  role  in  analytical  sociology.  Our  chapter  focuses  on  what  we  see  as  its most important strength, the ability to dissect with systematic experimentation and model analysis  a  potentially  complex  theoretically  proposed  social  mechanism.  This  is  especially  important because full comprehension of an empirically realistic model can be a formidable challenge.  Heterogeneity  of  an  agent  population,  inherent  randomness,  complex  network  structures, or local and non-linear interdependencies between multiple agents are but a few of the features many real social systems exhibit. The complexity of social reality is why social scientists increasingly resort to formal modeling and ABCM in the first place (Mäs, Chapter 4, this volume). The gain is that an ABCM is simpler and much more amenable for scientific analysis than social reality itself. Yet, a reasonably empirically plausible model can still be a complex system in itself where the relation between model assumptions and outcomes is far from trivial. A full understanding of why, how, and under which conditions a model is capable of generating outcomes resembling a real social phenomenon is thus a precondition that needs to be met before the ABCM can be considered to provide a valid explanation of the empirical phenomenon it targets.
Figure 24.1. Final states of two representative runs for different values of ethnic preference threshold (T). The proportion of Red, Green, Blue, and Yellow households is 60%, 20%, 10% and 10%, respectively. Groups are based on Clark and Fossett (2008): African American = Green; Asian = Blue; Hispanic = Yellow; White = Red. World size = 51x51, 90% density (≈2345 households and ≈255 empty cells)
Figure 9. Development of the Proportion of Correct Answers to a Question Within a Group as Consecutive Subjects Cast Their Vote (Thin Lines) and Development of the Proportion of Correct Answers Among Groups with a Wrong Majority (Dashed Lines) (Data from 100-Person Groups Only)

Social Influence Undermines the Wisdom of the Crowd in Sequential Decision Making

Vincenz Frey and Arnout van de Rijt

Teams, juries, electorates, and committees must often select from various alternative courses of action what they judge to be the best option. The phenomenon that the central tendency of many independent estimates is often quite accurate—“the wisdom of the crowd”—suggests that group decisions based on plurality voting can be surprisingly wise. Recent experimental studies demonstrate that the wisdom of the crowd is further enhanced if individuals have the opportunity to revise their votes in response to the independent votes of others. We argue that this positive effect of social information turns negative if group members do not first contribute an independent vote but instead cast their votes sequentially such that early mistakes can cascade across strings of decision makers. Results from a laboratory experiment confirm that when subjects sequentially state which of two answers they deem correct, majorities are more often wrong when subjects can see how often the two answers have been chosen by previous subjects than when they cannot. As predicted by our theoretical model, this happens even though subjects’ use of social information improves the accuracy of their individual votes. A second experiment conducted over the internet involving larger groups indicates that although early mistakes on easy tasks are eventually corrected in long enough choice sequences, for difficult tasks wrong majorities perpetuate themselves, showing no tendency to self-correct.

Comparing the Slider Measure of Social Value Orientation with Its Main Alternatives

Dieko M. Bakker and Jacob Dijkstra

Social Psychology Quaterly, 2021

DOI: 10.1177/01902725211008938

The Slider Measure of social value orientation (SVO) was introduced as an improvement from existing measures. We conduct an independent assessment of its suitability compared with the Ring Measure and the Triple Dominance Measure. Using a student sample, we assess the measures’ test-retest reliability (N = 88; using a longer time interval than previous studies) and sensitivity to random responses. Analyses pertaining to convergent validity, criterion validity, and the advantages of a continuous over a discrete measure are presented in the online appendix. Compared with alternatives, the Slider Measure has the highest test-retest reliability. However, it classifies random responses in an unbalanced way, assigning the vast majority of random responses to cooperative and individualistic, rather than altruistic and competitive, orientations. For all three measures, we propose improved ways of weeding out inconsistent responses.
Figure 1. Typical opinion dynamics generated by agent-based models of social influence: Evolution of the distribution of opinions in a one-dimensional bounded opinion space in a fully connected population. 

Models of Social Influence: Towards the Next Frontiers

Andreas Flache, Michael Mäs, Thomas Feliciani, Edmund Chattoe-Brown, Guillaume Deffuant, Sylvie Huet and Jan Lorenz

In 1997, Robert Axelrod wondered in a highly influential paper "If people tend to become more alike in their beliefs, attitudes, and behavior when they interact, why do not all such differences eventually disappear?" Axelrod’s question highlighted an ongoing quest for formal theoretical answers joined by researchers from a wide range of disciplines. Numerous models have been developed to understand why and under what conditions diversity in beliefs, attitudes and behavior can co-exist with the fact that very often in interactions, social influence reduces differences between people. Reviewing three prominent approaches, we discuss the theoretical ingredients that researchers added to classic models of social influence as well as their implications. Then, we propose two main frontiers for future research. First, there is urgent need for more theoretical work comparing, relating and integrating alternative models. Second, the field suffers from a strong imbalance between a proliferation of theoretical studies and a dearth of empirical work. More empirical work is needed testing and underpinning micro-level assumptions about social influence as well as macro-level predictions. In conclusion, we discuss major roadblocks that need to be overcome to achieve progress on each frontier. We also propose that a new generation of empirically-based computational social influence models can make unique contributions for understanding key societal challenges, like the possible effects of social media on societal polarization. 

Investments in and returns on network embeddedness: An experiment with trust games

Vincenz Frey, Vincent Buskens and Rense Corten

Social Networks 56, 81-92, 2019

Trust problems are ubiquitous in social and economic exchange. They are known to be mitigated if exchange partners are embedded in social structures that disseminate information on past behavior. If such “network embeddedness” makes exchanges possible that would not be possible otherwise, it is also expected that actors are willing to exert effort to establish embeddedness. Theory suggests that the degree to which network embeddedness facilitates trust depends on the size of the trust problem, and there are reasons to expect that embeddedness facilitates trust more strongly if it is established endogenously rather than imposed exogenously. We tested these predictions in a laboratory experiment in which 342 participants played repeated trust games with exogenous or endogenous embeddedness under varying sizes of the trust problem. The results confirm that embeddedness promotes trustfulness and trustworthiness. The results also show that endogenously chosen embeddedness promotes trustfulness more strongly than exogenously imposed embeddedness. However, we find no systematic variation in investments in embeddedness or effects of embeddedness in the size of the trust problem. 
Figure 1. Timeline of the Repeated Triad Trust Game (RTTG). 
Figure 1. Illustration of the intuition that one-to-many communication fosters isolation. Nodes have three characteristics (color, shape, and letter) that are open to influence. The number of traits shared by two nodes and, thus, the probability that a sender exerts influence on the receiver is shown by the number of lines connecting the nodes. Panel (a) shows the initial setup before the top-left agent communicated his shape trait either under the one-to-one communication regime (Panel b1) or the one-to-many regime (Panel b2) 

Communication in Online Social Networks Fosters Cultural Isolation

Marijn A. Keijzer, Michael Mäs and Andreas Flache

Complexity Article ID 9502872, 2018

Online social networks play an increasingly important role in communication between friends, colleagues, business partners, and family members. This development sparked public and scholarly debate about how these new platforms affect dynamics of cultural diversity. Formal models of cultural dissemination are powerful tools to study dynamics of cultural diversity but they are based on assumptions that represent traditional dyadic, face-to-face communication, rather than communication in online social networks. Unlike in models of face-to-face communication, where actors update their cultural traits after being influenced by one of their network contacts, communication in online social networks is often characterized by a one-to-many structure, in that users emit messages directly to a large number of network contacts. Using analytical tools and agent-based simulation, we show that this seemingly subtle difference can have profound implications for emergent dynamics of cultural dissemination. In particular, we show that within the framework of our model online communication fosters cultural diversity to a larger degree than offline communication and it increases chances that individuals and subgroups become culturally isolated from their network contacts. 

Relative power: Material and contextual elements of efficacy in social dilemmas

Jacob Dijkstra and Dieko M. Bakker 

Social Science Research 62, 255-271, 2017

In Step-Level Public Good (SPG) situations, groups of individuals can produce a public good if a sufficient number of them contribute. In SPG situations it is thus only rational for any group member to contribute if according to the beliefs of that group member her contribution is essential to the production of the public good. An individual's estimate of the impact of their contribution on the likelihood of public good production is known as their efficacy. The classic efficacy – cooperation hypothesis holds that individuals will be more likely to contribute if they estimate their contributions to be more necessary. Based on a game theoretical analysis of the SPG game, we contribute to the literature by identifying two distinct components of efficacy, viz. material efficacy and contextual efficacy. The former is based on objective characteristics of group members (such as resources, power, or skill) and the latter on beliefs about the material efficacy of other group members and expectations concerning their behavior. We present evidence from three experimental studies, showing how information on the distribution of material efficacy in the group can break the monotone material efficacy – cooperation relation. In addition, contrary to what one would expect based on both the efficacy – cooperation hypothesis and game theory, our results show that the effects of material efficacy are not mediated by contextual efficacy, both forms of efficacy having significant effects on behavior. 
Figure 1. Predicted probabilities by shares in Study 1; (a) Incomplete Information (b) Complete Information. 
Figure 2. Understanding shared ownership in relation to project process and outcome dimensions (Community Energy Strategy , 2013). 

Partnership or placation? The role of trust and justice in the shared ownership of renewable energy projects

Fleur Goedkoop and Patrick Devine-Wright

Governments in several European countries have developed policies that encourage companies to share ownership of renewable energy projects with local communities. Shared ownership presumes that company and community actors have common goals, can form effective partnerships and negotiate fair outcomes. But there is a lack of research on shared ownership, in particular, how it is constructed by different actors, and the role of trust in shaping practice. This study addressed this gap, drawing on qualitative data from in-depth interviews with 19 UK stakeholders from industry, community and advisory backgrounds. Thematic analysis revealed strong support for shared ownership in principle, but significant challenges in practice. Actors held different rationales and contrasting views on whether the policy should be discretionary or mandatory. A lack of trust was prevalent, with developers expressing skepticism regarding the capacities and representativeness of community actors; and community actors viewing developers as solely motivated by profit, instrumentally using communities to gain planning consent. We conclude that for shared ownership to become conventional practice, it will be necessary to provide mechanisms that facilitate partner identification at an early stage, which can help to build relations of trust between actors, within a more stable and supportive policy context. 

Individualization as Driving Force of Clustering Phenomena in Humans 

Michael Mäs, Andreas Flache and Dirk Helbing

PLoS Computational Biology 6 (10), e1000959, 2010

One of the most intriguing dynamics in biological systems is the emergence of clustering, in the sense that individuals self-organize into separate agglomerations in physical or behavioral space. Several theories have been developed to explain clustering in, for instance, multi-cellular organisms, ant colonies, bee hives, flocks of birds, schools of fish, and animal herds. A persistent puzzle, however, is the clustering of opinions in human populations, particularly when opinions vary continuously, such as the degree to which citizens are in favor of or against a vaccination program. Existing continuous opinion formation models predict “monoculture” in the long run, unless subsets of the population are perfectly separated from each other. Yet, social diversity is a robust empirical phenomenon, although perfect separation is hardly possible in an increasingly connected world. Considering randomness has not overcome the theoretical shortcomings so far. Small perturbations of individual opinions trigger social influence cascades that inevitably lead to monoculture, while larger noise disrupts opinion clusters and results in rampant individualism without any social structure. Our solution to the puzzle builds on recent empirical research, combining the integrative tendencies of social influence with the disintegrative effects of individualization. A key element of the new computational model is an adaptive kind of noise. We conduct computer simulation experiments demonstrating that with this kind of noise a third phase besides individualism and monoculture becomes possible, characterized by the formation of metastable clusters with diversity between and consensus within clusters. When clusters are small, individualization tendencies are too weak to prohibit a fusion of clusters. When clusters grow too large, however, individualization increases in strength, which promotes their splitting. In summary, the new model can explain cultural clustering in human societies. Strikingly, model predictions are not only robust to “noise”—randomness is actually the central mechanism that sustains pluralism and clustering. 
Figure 1. Opinion dynamics produced by the bounded confidence (BC) model [33] with and without noise. Populations consist of N~100 agents. Opinions vary between 2250 and 250. Initial opinions are uniformly distributed. For visualization, the opinion scale is divided into 50 bins of equal size. Color coding indicates the relative frequency of agents in each bin. (A) Dynamics of the BC-model without noise [33] over 10 iterations (Each iteration consists of N simulation events t.). At each simulation event, one agent’s opinion is replaced by the average opinion of those other agents who hold opinions oj(t) within the focal agent’s confidence interval (oi(t){Eƒoj(t)ƒoi(t)zE). For E~0:05, one finds several homogeneous clusters, which stabilize when the distance between all clusters exceeds the confidence threshold E. (B) Computer simulation of the same BC-model, but considering interaction noise. Agents that would otherwise not have been influential, now influence the focal agent’s opinion with a probability of p~0:01. This small noise is sufficient to eventually generate monoculture. (C) Simulation of the BC-model with opinion noise. After each opinion update, a random value drawn from a normal distribution with an average of zero and a standard deviation of h (abbreviated by N(0,h)) is added to the opinion. For weak opinion noise (h~5), one cluster is formed, which carries out a random walk on the opinion scale. When the opinion noise is significantly increased (h~18), there is still one big cluster, but many separated agents exist as well (cf. Fig. 4). With even stronger opinion noise (h~20), the opinion distribution becomes completely random.