The Ties That Bind
Studies of human social networks go high-tech
When John Cacioppo walks around Chicago with his book Loneliness, he hides the cover. “Who wants to go around with a big L on their forehead?” he says. Society, he complains, treats loneliness as a disease.
“People think it’s just neuroticism, or it’s people who can’t form relationships,” Cacioppo says. In 15 years of studying social isolation, the University of Chicago psychologist has found that loneliness is just another emotion. “Everybody has the capacity to be lonely, just as everybody has the capacity to feel pain,” he says.
And yet in one sense, his work shows how loneliness is very much like a disease: It can spread like the common cold.
That’s the conclusion of Cacioppo’s recent work with James Fowler, a University of California, San Diego political scientist, and Nicholas Christakis, a sociologist and physician at Harvard Medical School. Fowler and Christakis have made names for themselves studying how unlikely contagions — such as obesity and happiness — spread in social networks. The trio’s study of how lonely people pass on their pain highlights an emerging area of research, which some call computational social science. The field uses new data collection and analytical techniques to understand how people are connected in networks and how influences move through social links. In some cases, the effects that people have on those they know can be startling.
Some researchers hope that, by illustrating large-scale human interactions, this new field could help reduce depression, contain epidemics and generally make the world a better place — not to mention clarify the basics of human nature. “What we’re trying to do,” says Fowler, “is start a revolution.”
Subscribe to Science News
Get great science journalism, from the most trusted source, delivered to your doorstep.
That revolution is fueled by a fire hose of data coming from cell phones, credit cards and the Internet. People leave traces of their place in the social network every time they do a Google search or swipe a subway fare card. Researchers are beginning to use these traces to see how people influence one another. But not everyone believes that these studies show what they claim to. Limitations of data collection and quality leave other interpretations open. Privacy issues can keep data under wraps. And barriers in the social network of academia itself could stunt the growth of the nascent field.
Not only the lonely
To track how behaviors and feelings travel through a network, Christakis and Fowler painstakingly digitized 30 years’ worth of paper records from a long-term health study in Framingham, Mass. The pair constructed more than 50,000 ties among about 5,000 people in the small town. In a series of studies, the researchers plotted how weight, smoking habits and happiness (as measured on a diagnostic test for depression) changed over time. Each of the three traits appeared to move through the network in clumps, like flocks of birds or schools of fish.
The traits also seemed to spread from person to person as far as three degrees of separation. This finding was widely interpreted to mean that your best friend’s roommate’s brother can make you measurably happier or sadder, fatter or thinner, even if you’ve never met.
While many found these results surprising when they first hit medical journals and newspapers, there are intuitive explanations for each trend, Christakis and Fowler argue in their book, Connected (Little, Brown, 2009). If all your friends quit smoking, there’s social pressure for you to quit, too. If your friends all gain weight, there’s less social pressure for you to stay slim. The adage “smile and the world smiles with you” hints that the idea of contagious happiness has some roots in common sense, too.
But loneliness makes a strange virus. Fowler, Christakis and Cacioppo traced loneliness in 5,124 people in the Framingham network using answers from a common screening test for depression. One question explicitly asks how many days in the previous week the participant felt lonely. Cacioppo’s earlier research confirmed that this question can measure loneliness as distinct from depression.
Using the same statistical methods as in the earlier studies, the researchers plotted instances of loneliness in Framingham from 1983 to 2001. They found that lonely people tended to cluster at the edges of the network, and that loneliness — like happiness, smoking and obesity — seems to spread through the network out to three degrees of separation. Having one lonely friend can leave you 40 to 65 percent more likely to be lonely than if you have no lonely friends. A lonely friend of a friend increases your chances by 14 to 36 percent, and a friend of a friend of a friend contributes between 6 and 26 percent to your loneliness — whether you know each other or not. For anyone further away in the network, the effect disappears.
While conventional wisdom may say loneliness is a symptom of having few social connections, not a cause, Cacioppo says that lonely people are found at all levels of popularity. It doesn’t matter how many social ties you actually have, only how satisfied you are with them.
The feeling starts as a sense that the world is unfriendly, that those closest to you might not back you up in a crisis. Cacioppo thinks this feeling could have evolutionary roots. If human ancestors faced a greater risk of becoming prey when alone, an emotional early-warning signal that not all was well in their social network could have been lifesaving.
Thinking the world is unfriendly can become a self-fulfilling prophecy. “Lonely individuals expect social threats,” Cacioppo says. “If you think others around you are going to be nasty to you, you see more nasty behavior.” Feeling under fire, lonely people alternately lash out and withdraw in self-defense. This erratic behavior pushes friends away, but it also makes the friends more likely to mistrust the world and start the cycle over again.
This could explain how loneliness might be catching. But the idea that people are not alone in their isolation could have troubling implications for society.
“We think of society like a crocheted sweater,” Fowler says. “If one thread becomes loose, the whole sweater can become unraveled.”
A tangled net
There are two other reasons why people with certain traits might be clumped together in a network. One is that similar people tend to like each other — happy people have happy friends, smokers befriend other smokers. The other is that some outside factor influences everyone in a network equally — say, a fast-food place opens just before everyone in the neighborhood gains weight.
To try to control for people choosing similar friends, Christakis and Fowler’s studies all looked at snapshots of the network at different points in time. If a group of smokers had all quit during a certain period, researchers assumed it probably wasn’t a result of people’s fondness for those like themselves.
To account for outside influences, the team also looked at the direction of the ties. If Alice named Bob as a friend but he didn’t name her back, for example, Bob has more of an influence on Alice than she has on him. The strongest influence occurred between mutual friends — those who named each other. If a change was caused by the environment, such a directional effect shouldn’t be there.
Some scientists doubt that these control measures are enough, though. Jason Fletcher of Yale School of Public Health used Fowler and Christakis’ methods on a network of adolescents and found that acne, height and headaches seem to be contagious, too. “If we can find these effects in height, it seems like we can find it in anything. And I think their findings are, ‘Yes, we find it in everything,’” Fletcher says. “It’s hard to tell if their model is strong enough to detect no social effects when they actually don’t exist.”
Fowler responds that acne, height and headaches can all be influenced by diet, especially in adolescents, and peers can certainly change each others’ eating habits. But he admits that the methods are “still not foolproof.”
Others have complained that the data are incomplete: People in the Framingham study named only a few friends, and some of those friends weren’t in the study. So there could be people influencing the Framingham participants whose effects are invisible to researchers.
That’s also a fair criticism, Fowler says. “I don’t think there’s ever going to be a network study where we can say the results of one study are automatically generalizable to all 6 billion people on the planet,” he says. “Each part of the network is inherently incomplete.”
The Framingham studies offer a glimpse of what social scientists could do if they could only collect enough information. Fortunately, there is a place where scientists can collect near-complete large-scale data on social networks and even run experiments: the Web.
“Historically you couldn’t really do large-scale network studies with data,” says Duncan Watts, a physicist-turned-sociologist who has studied networks for over a decade. “Now we can.”
Watts is perhaps most famous for a network model, codeveloped with mathematician Steven Strogatz of Cornell University, that showed how many real-world networks lie somewhere in between completely ordered and completely random. Watts and Strogatz called them “small world” networks because the average path from any one spot in the network to another was short. Networks with these properties show up everywhere, from the neural networks of nematodes to the power grid of the United States to networks of actors costarring in films.
E-mail let Watts test the small world model on a global scale. In 2003, he and two colleagues replicated the famous experiment that produced the phrase “six degrees of separation.” The team asked more than 60,000 online participants to forward e-mails to a far-away target through networks of acquaintances. The e-mails, the majority of which originated in North America, reached their destinations — including an archival inspector in Estonia and a veterinarian in the Norwegian army — in five to seven steps.
Watts now works at Yahoo! Research in New York City, where he uses specially designed websites and applications on social networking websites such as Facebook to ask specific questions about group behavior. His recent work, presented at meetings and appearing in Social Psychology Quarterly and Management Science, has shown that people will like unpopular songs more if others seem to like them (but not as much as they like songs that are actually popular), that the choice to cooperate with or mooch off an opponent in an online game can cascade through a network, and that people think they and their friends are more similar than they really are.
Other studies examine blogs as indicators of political climate or national mood. Lada Adamic of the University of Michigan in Ann Arbor plotted online links among political blogs in the months leading up to the 2004 presidential election. Her results, which she presented in 2005, showed a sharp split between the liberal and conservative “blogospheres.”
Peter Dodds and Christopher Danforth of the University of Vermont in Burlington tracked emotionally charged words in blog posts. In a 2009 paper in the Journal of Happiness Studies, the two suggest that the method could be used as an emotional barometer, giving researchers a bird’s-eye view of how moods clump and spread. By this measure, the happiest day in the past four years was Election Day, 2008; among the saddest: the day Michael Jackson died.
Using mobile phones to track people’s movements and activities is another example of how technology is being used to better understand social networks. David Lazer, now at Harvard University, and his colleagues gave students and faculty mobile phones that tracked the participants’ calls and proximity to other phones in the study. Lazer’s team compared phone data with the volunteers’ own reports of their friends and habits. The researchers were able to identify 95 percent of friendships based on the phone data alone, the team reported in the Sept. 8 Proceedings of the National Academy of Sciences.
This predictive power could be useful in tracking phenomena from political leanings to the spread of the flu, Lazer suggests. Knowing where and when students congregate could help university officials decide whether it would help to cancel class in the case of an H1N1 outbreak, for example. And pinpointing a person’s place in the network could help in designing customized health programs.
“Until recently you could only think of these things theoretically,” Watts says. “What’s so tantalizing about the Web is that you can really do this for real. You can get inside the black box of society and see how it really works.”
But opening the black box isn’t easy. Social data are hard to obtain without putting personal information at risk.
“It doesn’t take too many signals to uniquely identify someone,” Lazer says. “The trick there is, how do you de-identify the data and not completely destroy the insights that they provide?”
The data are also noisy and hard to work with. “I think we will need new and better methods,” Fletcher says. “We’re almost saturated in data, but we’re not saturated in methods yet.”
Computational social science, as a cross-disciplinary field, could overcome weaknesses in its progenitor fields. Traditional social science research is usually limited to snapshots of small groups and often relies on questionable self-reported data. Physicists and mathematicians borrowed methods from statistical mechanics and computational biology, which produce nice visualizations of how a group can act as a unit. But to reliably reflect human behavior, the models need to be grounded in how people really behave.
And yet the new field’s scholars may be culturally and, on some campuses, physically distant. “Sociologists felt miffed when you had these physicists come in the 1990s repackaging ideas that had been around a long time,” Fowler says. While there’s more cross-talk now than a decade ago, researchers still sometimes step on each others’ toes or talk over each others’ heads, “and everyone just winces.”
One of social network research’s own tools demonstrates the divide. A plot of Twitter usage from a recent conference shows attendees lumped neatly into two groups: physicists and math types on one side, social scientists on the other.
How to bridge this gap? “The same way we’ve always done it: We get them in the same room together,” Fowler says. “We promote real-world social relationships, and then utilize the technology to stay in touch after the conference is over.”
Similar strategies could reach out to people on society’s fringe, too. A November 2009 Pew Internet Survey found that, contrary to popular suspicion, people who spend a lot of time on mobile phones and Facebook socialize more in person than people who don’t. “If you use the Internet to promote face-to-face relationships, it makes them richer, higher quality,” Cacioppo says. Virtual networks may be more than just treasure troves of data. They can also thread lonely people back into the social crocheted sweater.