Researchers often regard these dense clusters of connections to be redundant when it comes to spreading information; networks featuring such clusters are considered less efficient than networks with a greater proportion of long ties. But getting people to change ingrained habits, Centola found, requires the extra reinforcement that comes from those redundancies. In other words, people need to hear a new idea multiple times before making a change.
“For about 35 years, wisdom in the social sciences has been that the more long ties there are in a network, the faster a thing will spread,” says Centola. “It’s startling to see that this is not always the case.” Centola’s paper on the subject, “The Spread of Behavior in an Online Social Network Experiment,” is published in the Sept. 3 issue of the journal Science.
The buddy system
To see what difference the form of a social network makes, Centola ran a series of experiments using an Internet-based health community he developed. The 1,528 people in the study had anonymous online profiles and a series of health interests; they were matched with other participants sharing the same interests — “health buddies,” as Centola calls them in the paper. Participants received e-mail updates notifying them about the activities of their health buddies.
Centola placed participants into one of two distinct kinds of networks — those oriented around long ties, and those featuring larger clusters of people — and ran six separate trials over a period of a few weeks to see which groups were more likely to register for an online health forum website offering ratings of health resources.
Overall, 54 percent of the people in clustered networks registered for the health forum, compared to 38 percent in the networks oriented around longer ties; the rate of adoption in the clustered networks was also four times as fast. Moreover, people were more likely to participate regularly in the health forum if they had more health buddies who registered for it. Only 15 percent of forum participants with one friend in the forum returned to it, but more than 30 percent of subjects with two friends returned to it, and over 40 percent with three friends in the forum made repeat visits.
“Social reinforcement from multiple health buddies made participants much more willing to adopt the behavior,” notes Centola in the paper. Significantly, he writes, this effect on individuals “translates into a system-level phenomenon whereby large-scale diffusion can reach more people, and spread more quickly, in clustered networks than in random networks.”
Centola thinks the existence of this effect has important implications for health officials. A “simple contagion,” in network theory, can spread with a single contact; a “complex contagion” requires multiple exposures for transmission. A disease, Centola suggests, can spread as a simple contagion, but behavior that can prevent the disease — such as going to a clinic for a vaccination — might spread only as a complex contagion, thus needing to be spurred by reinforcement from multiple neighbors in a social network.
“If there is a significant difference between simple and complex contagions, that actually matters for our policy interventions,” says Centola. The public promotion of screenings and other forms of disease prevention might best be aimed at communities and groups that act as closely clustered networks.
Studying communities, online and off
Colleagues in the field find the study to be of both theoretical and practical interest. “It’s interesting work because it shows that for the diffusion of certain kinds of things, you really need reinforcing,” says David Lazer, an associate professor of public policy at Harvard’s Kennedy School of Government. “You need wide bridges to transmit complex information like health data, and that is different from the traditional picture of how things spread in a network.”
As Centola acknowledges, the study has limitations. Joining an online health forum has little cost in time or money, unlike many other kinds of health behavior, from vaccinations to changing one’s diet or adopting an exercise routine. “Getting a colonoscopy is hard,” Centola says. “Just hearing about it is probably not going to convince you to do it.” The rate of adoption would likely vary widely for many forms of health behavior, and be relatively low when notable costs are involved.
On the other hand, Centola notes, the existence of those costs implies that social reinforcement, such as having multiple friends and relatives who get colonoscopies, may be especially important. “These redundant signals are necessary to make people adopt the behavior,” Centola says.
Further fieldwork may help determine how resistant people are to changing particular forms of health behavior. “One thing this study begs, in a good way, is more research in natural settings,” says Lazer. To see the effectiveness of public-health measures, he suggests, “You might try to target two neighborhoods in different ways, and then assess which is more effective.”
For his part, Centola thinks there is also further work to be done evaluating the effects of online social networks on behavior. “There is a natural implication in terms of what this means for designing online communities,” says Centola. His new research, building on his current paper, aims to find new designs for online communities, in order to promote good health practices.