Collective Dynamics Group
Collective Dynamics Group
How closely connected are two people who live on opposites sides of the world? In what contexts are social ties most likely to be generated? To what extent do people make choices based on the preferences of others? Can we predict the likelihood that two individuals will know each other? ISERP's Collective Dynamics Group (CDG) tackles these and other social questions using the modern mathematical and computational techniques central to physics, applied mathematics, statistics, and computer science.
When Duncan Watts, founding director of CDG, joined the Columbia Sociology Department in 2000, his background in applied mathematics and physics was unusual. Though there are others in the department with atypical backgrounds and broad conceptions of sociology-including Harrison White, who was a pioneer in bringing mathematical models to social research-Watts's appointment was indicative of a new direction for the social sciences at Columbia and of a broadening notion of sociology's disciplinary boundaries.
CDG received its name (naturally a choice arrived at collectively) in 2003 when the group officially joined ISERP, and its infrastructural base moved into ISERP's new research suite. Along with Watts, the group is comprised of graduate students, postdoctoral fellows, a research scientist, a programmer, and several affiliated faculty, with backgrounds spread across sociology, economics, political science, physics, mathematics, engineering, epidemiology, and computer science. The group's research spans three major interconnected areas: mathematical model construction, online large-scale experiments, and data collection and analysis. Areas of current theoretical research for the CDG include collective action, decision making, and problem solving; biological and social contagion; search and robustness in complex networks; and sustainability of cooperation in dynamically evolving networks. The group has been supported in its endeavors by the National Science Foundation, the James F. McDonnell Foundation, the Office of Naval Research, Legg Mason, and the Santa Fe Institute.
With respect to mathematical modeling, the group's general philosophy is to construct simple models of individual behavior and then explore the consequences at a population level. These "simple models," however, are more than just transplants of physics models for, say, how magnetism arises out of many interconnected individual spins. While the group seeks to create parsimonious, "physics-ish" models of individual behavior, its models are informed by research in areas such as sociology, social psychology, behavioral economics, game theory, and market research. When working with graduate students and postdoctoral scholars who have strong technical backgrounds, "it is important to provide an environment in which they are encouraged to avoid the usual pitfall of natural-scientist-as-sociologist, which, in a nutshell, is to assume they already know which questions are interesting and how to formulate them," Watts explains. "The result is methodologically dazzling solutions to theoretically nonsensical or otherwise irrelevant questions." Watts's motivation to create an applied math research program in a social science environment, rather than the other way around, sprung from his conviction that the trick to proposing interesting models was to know something about the phenomena in question. The solutions Watts's team arrives at are central for social science, but social scientists may need to develop new ways of looking at the world to deeply appreciate the significance of its work.
Over the last few years, the group has constructed and successfully run several major online experiments. These experiments are designed to be enjoyable and increase participation while displaying scientific value. CDG's seminal experiment was an updated version of the Small World experiment carried out by Stanley Milgram in the late 1960s. The experiment, currently on its updated second run, gives participants "target persons" to move a message closer to by emailing a current friend or acquaintance. This movement of messages, a kind of social search, results in chains of messages tracing their way towards targets and is dependent on a collective effort of participants. The eighteen designated targets were distributed around the world in thirteen different countries. Approximately 60,000 participants were involved. The experiment received wide media coverage, greatly increasing its exposure and participation level.
Some important findings of the Small World experiment were that message chains that reached their target consisted predominantly of relatively weak ties, did not strongly depend on highly connected individuals, and that the constituent ties were more likely to be workplace based. The group estimated that the lengths of successful message chains had a median of 5 to 7 (after the participation rate was accounted for). The group also found that incentives and perceptions mattered; if individuals could imagine how a sequence of individuals could connect them to the target, then they were more likely to participate-in other words, it's a small world if you think it is. The experiment has continued to evolve and more versions are planned for release soon.
The group is currently running a second major experiment, which explores the degree to which people follow each other in making choices. This experiment, Music Lab, allows participants to listen to, rate, and download songs recorded by unsigned artists. Here the group investigates how individual behaviors aggregate to produce collective outcomes. Collective decisions are generated by a social contagion process, which is augmented by the machinery of mass media. Consequently, understanding interpersonal influence is crucial to understanding the behavior of both individuals and groups.
Alongside this experimental work, the group continues to advance theoretical work on the problem of social search. Watts's early work on small-world networks demonstrated the topological property that no two members of a real-world network are far apart but that the network is nevertheless highly clustered (your friends are likely to know each other). But a major question remained: How is it that individuals actually find each other? It turns out that "navigation" of complex networks in general is a difficult task. The group's insight to this question was simply that individuals have identities and are not just featureless nodes on a graph. This observation led to the use of a basic functional model of identity: where an individual works and lives, their religious and community involvements, their hobbies. All these attributes bring individuals into contact with others who have a matching identity in at least one attribute, and given the description of two individuals, an estimate of how far apart they are can easily be made. Using these ideas, the group was able to construct a sociologically plausible model of social search that succeeded in emulating the performance of real populations.
CDG's current major data analysis project is a study of an evolving, large-scale social network, as reflected by email exchange. The group has a record so far of fifteen months of email interactions between individuals at a large institution. While email logs provide only a partial sample of communication between individuals, email does correlate well with other modes of interactions. What sets this study apart from other similar work is a substantial accompanying set of demographics. The projects being carried out include understanding the mechanisms of formation and dissolution of social ties between individuals, the temporal characteristics of communication (such as response times), and the coevolution of individuals' formal affiliations and their social connections.
The group meets at a weekly session where speakers from both within and outside the group are typically questioned relentlessly for three hours, and there is much roundtable discussion. Anyone interested in receiving emailed announcements of these meetings and other topics can be added to the CDG's mailing list. The group's website provides, among other things, project descriptions, paper downloads, links to online experiments, and the group meeting schedule.
See Also
- Collective Dynamics Group
- Related research grants
- Press release: Peer Influence on Music Choices
- Press release: Analysis of a Network of 14.5 Million E-Mail Messages
