Theoretical models of critical mass have shown how minority groups can initiate social change dynamics in the emergence of new social conventions. Here, we study an artificial system of social conventions in which human subjects interact to establish a new coordination equilibrium. The findings provide direct empirical demonstration of the existence of a tipping point in the dynamics of changing social conventions. When minority groups reached the critical mass-that is, the critical group size for initiating social change-they were consistently able to overturn the established behavior. The size of the required critical mass is expected to vary based on theoretically identifiable features of a social setting. Our results show that the theoretically predicted dynamics of critical mass do in fact emerge as expected within an empirical system of social coordination.

A longstanding problem in the social, biological, and computational sciences is to determine how groups of distributed individuals can form intelligent collective judgments. Since Galton's discovery of the "wisdom of crowds" [Galton F (1907 social networks | collective intelligence | social learning | wisdom of crowds | experimental social science S ince Galton's discovery of the "wisdom of crowds" over 100 years ago (1), results on crowdsourcing (1, 2), prediction markets (3), and financial forecasting (4, 5) have shown that the aggregated judgment of many individuals can be more accurate than the judgments of individual experts (2, 4, 6-8). Statistical explanations for this phenomenon argue that group accuracy relies on estimates taken from groups where individuals' errors are either uncorrelated or negatively correlated, thereby preserving the diversity of opinions in a population (9). Thus, although individuals may have estimates both far above and far below the true value, in aggregate these errors cancel out, leaving an accurate group judgment (2, 9, 10).Recent experimental evidence has suggested that the wisdom of crowds may be undermined by processes of social influence, in which people exchange information about their estimates and revise their judgments to align with one another (11-13). When social influence leads to correlated errors, both independence and diversity are reduced, which has been argued to compromise the reliability of the group judgment (9,(11)(12)(13)(14)(15)(16)(17)(18). In direct contrast with these results, however, theoretical models of social learning (19-21) have suggested that the effects of social influence on collective decisions vary based on the structure of the interaction network, predicting that, under the right conditions, social learning can lead a group's median judgment to improve (20)(21)(22)(23)(24).This prediction derives from the assumption that, when people learn about the beliefs of others, they revise their own beliefs to become more similar to their social referents (11, 12, 25, 26). Following the DeGroot model of social learning, this theory suggests that each individual's revisions are based on a weighted average of their own belief and the beliefs of their social referents (19). Thus, an individual's revision is determined in part by the amount of weight they place on their own belief relative to social information. When this "self-weight" is independently and identically distributed (i.i.d.) throughout a population, and the population is embedded in a decentralized social network (i.e., one in which everyone is equally connected), this model predicts that belief distributions will converge on the statistical mean of the initial, independent beliefs (SI Appendix). Thus, if the initial group mean is accurate, exposure to social influence will lead individuals in the group to become more accurate, improving the accuracy of the group's median, even as the group mean remains fixed (SI Appendix).We build on the DeGroot model to generate theoretical predictions ...

Vital scientific communications are frequently misinterpreted by the lay public as a result of motivated reasoning, where people misconstrue data to fit their political and psychological biases. In the case of climate change, some people have been found to systematically misinterpret climate data in ways that conflict with the intended message of climate scientists. While prior studies have attempted to reduce motivated reasoning through bipartisan communication networks, these networks have also been found to exacerbate bias. Popular theories hold that bipartisan networks amplify bias by exposing people to opposing beliefs. These theories are in tension with collective intelligence research, which shows that exchanging beliefs in social networks can facilitate social learning, thereby improving individual and group judgments. However, prior experiments in collective intelligence have relied almost exclusively on neutral questions that do not engage motivated reasoning. Using Amazon's Mechanical Turk, we conducted an online experiment to test how bipartisan social networks can influence subjects' interpretation of climate communications from NASA. Here, we show that exposure to opposing beliefs in structured bipartisan social networks substantially improved the accuracy of judgments among both conservatives and liberals, eliminating belief polarization. However, we also find that social learning can be reduced, and belief polarization maintained, as a result of partisan priming. We find that increasing the salience of partisanship during communication, both through exposure to the logos of political parties and through exposure to the political identities of network peers, can significantly reduce social learning.

Since the publication of "Complex Contagions and the Weakness of Long Ties" in 2007, complex contagions have been studied across an enormous variety of social domains. In reviewing this decade of research, we discuss recent advancements in applied studies of complex contagions, particularly in the domains of health, innovation diffusion, social media, and politics. We also discuss how these empirical studies have spurred complementary advancements in the theoretical modeling of contagions, which concern the effects of network topology on diffusion, as well as the effects of individual-level attributes and thresholds. In synthesizing these developments, we suggest three main directions for future research. The first concerns the study of how multiple contagions interact within the same network and across networks, in what may be called an ecology of contagions. The second concerns the study of how the structure of thresholds and their behavioral consequences can vary by individual and social context. The third area concerns the roles of diversity and homophily in the dynamics of complex contagion, including both diversity of demographic profiles among local peers, and the broader notion of structural diversity within a network. Throughout this discussion, we make an effort to highlight the theoretical and empirical opportunities that lie ahead.

To identify what features of online social networks can increase physical activity, we conducted a 4-arm randomized controlled trial in 2014 in Philadelphia, PA. Students (n = 790, mean age = 25.2) at an university were randomly assigned to one of four conditions composed of either supportive or competitive relationships and either with individual or team incentives for attending exercise classes. The social comparison condition placed participants into 6-person competitive networks with individual incentives. The social support condition placed participants into 6-person teams with team incentives. The combined condition with both supportive and competitive relationships placed participants into 6-person teams, where participants could compare their team's performance to 5 other teams' performances. The control condition only allowed participants to attend classes with individual incentives. Rewards were based on the total number of classes attended by an individual, or the average number of classes attended by the members of a team. The outcome was the number of classes that participants attended. Data were analyzed using multilevel models in 2014. The mean attendance numbers per week were 35.7, 38.5, 20.3, and 16.8 in the social comparison, the combined, the control, and the social support conditions. Attendance numbers were 90% higher in the social comparison and the combined conditions (mean = 1.9, SE = 0.2) in contrast to the two conditions without comparison (mean = 1.0, SE = 0.2) (p = 0.003). Social comparison was more effective for increasing physical activity than social support and its effects did not depend on individual or team incentives.