Costs and Benefits of Connection
That things spread from person to person is taken for granted when it comes to gossip and infections. But years ago, as he was beginning his medical and academic career, an interest in end-of-life care led Nicholas Christakis to study the “widowhood effect”: people’s tendency to die soon after a spouse does. (The surviving partner’s chances of dying in the next three months increase by 30 to 90 percent.) Asking what this phenomenon’s cause could be, Christakis—professor of medicine and medical sociology at Harvard Medical School and professor of sociology in the Faculty of Arts and Sciences—came to view the phenomenon as a special case of network effects, similar to the way spouses influence each other’s opinions or give each other colds.
In 2002, Christakis was introduced by a mutual friend to James H. Fowler ’92, Ph.D. ’03, then a graduate student in the government department who was studying another special case of network effects: how one person’s decision to vote influences others’ behavior. Both men were interested in moving beyond these special cases and studying network effects more broadly; inspired by the rapid pace of discovery in the field, they also determined to write a book about network science for a popular audience. Connected: The Surprising Power of Our Social Networks and How They Shape Our Lives (Little, Brown, 2009) presents a cornucopia of provocative research findings from others and reports on the early results from Christakis and Fowler’s own research collaboration.
Although the authors brainstormed all the things that could possibly spread through human social networks and envisioned a longitudinal study that would track as many of those things as possible, funding challenges drove them to use the existing Framingham Heart Study (see main article for more detail.) Still, the Framingham data proved fertile ground. With funding from the National Institute on Aging, the researchers have published findings on the spread of obesity, smoking and smoking cessation, happiness, loneliness, depression, and alcohol use. Early social-network research documented the ways social relationships can affect health, but focused on isolated relationships—how a father’s alcoholism or a mother’s depression affects their children, for example. Those studies might have included a few dozen people; the new generation of network scientists explores how large groups of people—thousands or millions—influence one another’s health.
In addition to studying contagion—what phenomena spread through networks—they have used the Framingham data to study connection—the effects of the simple fact that a particular network structure exists, considered separately from what spreads through it. They believe that what they have found sheds light on the evolution of humans’ tendency to socialize, and on why there is so much variation—why we have loners and introverts as well as social butterflies.
Addressing the Critics
As christakis and fowler forge ahead with their research, they have sometimes found themselves on the defensive, but they firmly believe they have established causation, not mere correlation, for the phenomena they have studied. For instance, they maintain that obesity spreads from person to person—that what they are observing is not just homophily, the tendency for overweight people to choose friends who are also heavy.
Because the Framingham study tracked subjects for more than 30 years, bringing them in for repeated interviews and observation, the survey enabled charting subjects’ weight gain and loss over time. That meant Christakis and Fowler (now a political scientist at the University of California) could see whose weight gain came first and whose came later—mapping weight gain like the spread of a virus. They found that if a subject’s friend became obese, the probability that the subject himself would become obese increased by 57 percent.
Because these effects unfolded over time—first one person becomes obese, then the other—and because they observed, and accounted for, changes in people’s friend networks, Christakis and Fowler feel confident they were not observing people realigning their friendships to choose friends of similar weight.
As far as the mechanism—how or why one person’s weight gain causes her friends to gain weight—the researchers can only speculate. They allow that behavior changes could be the mode of transmission: a friend’s admission that she loves to eat at Burger King might inspire you to give the restaurant a try—or perhaps you go there together as a social activity. Another possibility is shifting opinion: when a friend gains weight, you change your idea of what an acceptable weight is. (Other researchers’ studies with different groups of subjects are trying to tease out the relative import of these factors.)
Christakis and Fowler note as well that they controlled for many factors that could influence body mass, including age and starting or stopping smoking. Their mathematical models also allowed for the fact that people in adjacent locations in a social network will be subject to some of the same external influences. The most compelling evidence, they say, is the before-and-after nature of their results, and the results’ directionality—as with hypothetical subjects Anna and Barbara. If Anna and Barbara reciprocally name each other as friends, then one woman’s becoming obese has enormous influence, increasing the other’s chances of gaining weight by 171 percent. If Anna names Barbara but Barbara does not name Anna in return, then the statistics show that Barbara’s weight influences Anna’s weight—but the influence does not travel the other way. The logical conclusion is that someone’s weight gain influences you only if you consider that person a close friend, an important figure in your life.
Examining the study subjects’ self-reporting of happiness yielded similar results: happiness, the researchers concluded, can also spread through networks. As with obesity, they began with anecdotal knowledge: we can see that obese people tend to have obese friends, and we know that one person’s mood can rub off on another. The quest to quantify this anecdotal knowledge did not disappoint: subjects in the Framingham cohort were 45 percent more likely to report being generally happy if a friend had become happy (gone from “not happy” to “happy”) within the previous six months. Curiously, when researchers considered study participants who were neighbors, they found that weight gain did not spread between neighbors, but happiness did—and it spread onlymost reliably between next-door neighbors; neighbors who lived further apart on the same block had no effect on one another.
Christakis and Fowler also found that loneliness spreads in networks. (It may seem counterintuitive to include loneliness in a study of social networks, but researchers have long known that it is possible for people to feel lonely despite having friends.) The Framingham study asked subjects how many days in the past week they had felt lonely; Christakis and Fowler found that those subjects directly connected to a person who reported ever feeling lonely (whether rarely or frequently, as distinct from respondents who said they never felt lonely) were 52 percent more likely to report feeling lonely themselves. For friends who lived within a mile of each other, one additional day of loneliness per subject per week correlated with an additional one-third of a day that the subject’s friends spent feeling lonely. Here, too, directionality was relevant, with the results being stronger in mutual than in nonmutual friend pairs.
Like happiness (but unlike obesity), geographic proximity mattered in the case of loneliness, which spread between next-door neighbors, but not between neighbors living farther apart. The type of relationship mattered, too: spouses affected one another less than friends, and siblings did not affect one another at all. Gender made no difference in the spread of happiness, but in the spread of loneliness it did: loneliness is more contagious among women than among men. Such details highlight one key observation: even though certain principles apply broadly across different types of networks, it’s important to understand exactly where and how a given phenomenon spreads before attempting network-based policy interventions. “Not everything spreads in networks, and not everything that does spread spreads in the same way,” Christakis notes. “Therefore, we should be wary of a ‘one-size-fits-all’ approach [to intervening in networks]. The details matter.”
Among Christakis and Fowler’s findings, those on obesity have received by far the most publicity—perhaps because people easily accept that moods spread from one friend to another, but are loath to believe a friend’s behavior could trump, or drive, one’s own diet and exercise choices. Journalists who mention contrary evidence frequently cite the work of two researchers at the Yale School of Public Health who have found what they claim are network effects—similar to those Christakis and Fowler found—for acne, headaches, and height. Christakis disputes their analytical methods, as well as the claim that there is no plausible mechanism of network influence on any of these phenomena. He calls it a “false controversy,” saying, “So often when a paper of ours appears, these guys are quoted” as a counterpoint. “It’s interesting that no one else is ever quoted.” Still, his lab group has undertaken a spate of follow-up studies trying to replicate their own original results. Thus far, he reports, their own and other researchers’ studies in this area have confirmed the initial findings. Separately, they have found evidence for other behaviors’ spread through networks: teenagers who don’t sleep enough influence their friends to skimp on sleep, too; subjects copy others’ altruistic actions in laboratory experiments; mapping which other doctors a doctor knows helps explain her prescribing patterns. In addition, Mark Pachucki, a Christakis doctoral student, is mapping how dietary fads spread. (For a fuller discussion of these criticisms and the response from Christakis and Fowler, see “Is Happiness Catching?” from the New York Times Magazine, September 13, 2009.)
Popularity: A Mixed Blessing
Throughout their research, regardless of the particulars of what spreads to whom, Christakis and Fowler have generally found that being highly connected (i.e., having lots of friends) seems to make people happier. The most consistently happy group of people were those who were centrally located in their networks—people with lots of friends who were themselves well connected. What’s more, being highly connected seemed to guard against the spread of loneliness. Each additional friend a subject had—regardless of whether that friend reported feeling lonely—acted to decrease the time the subject spent feeling lonely by 5 percent. The average person in the study reported feeling lonely 48 days per year; adding two friends to one’s network, then, would decrease the time spent feeling lonely by nearly five days.
It is common knowledge that highly connected people have an easier time finding a job and can benefit in other ways: for example, studies of “public goods games,” in which people demonstrate altruistic behavior in a laboratory setting, have found that popular people tend to end up with more money at the end of the game. So if being connected is good, why is there so much variation in the population? Because being connected can be good or bad, depending on what’s spreading through the network. “If genes influence whether we are in the middle or at the periphery of a social network, they can also affect how rapidly we get a piece of gossip (the center is better) or how likely we are to become infected with an epidemic disease (the center is worse),” Christakis and Fowler write in Connected. “Traits that are always adaptive tend to reach what geneticists call fixation in the population: in the long run, everyone becomes the same. But when there are conflicting pressures—under some circumstances, a trait is beneficial, but under others, it is not—then it is possible to maintain diversity in the population in the face of natural selection.”
Christakis believes it’s important to be aware of the advantages that being highly connected confers. Unconnected people may be modern society’s next disadvantaged group; Christakis doesn’t find it hard to imagine a movement that advocates for the rights of this group, especially considering the partly genetic basis of social behavior (see Networks, from Neolithic to Now). Just as groups try to bring attention to discriminatory hiring practices that disadvantage people who are nonwhite or overweight, so, too, might a campaign argue for a thumb on the scale in favor of people for whom forming numerous ties doesn’t come naturally. Suggests Christakis, “There is a new kind of hierarchy we might think about as policymakers and social scientists. It’s not just whether y ou are rich or poor, black or white, urban or rural. It’s also about where you are located in the network.”
The larger point, he adds, is that “the benefits of a connected life”—whether the connections are in person, by phone, or online—“far outweigh the costs. Even though things like violence and suicide and germs and slander can spread in networks, so too do love and happiness and kindness and altruism and news. And the benefits of those things compensate for the spread of bad things.”