Resolving problems like this one is important because it has implications for how we go about dealing with controversial issues like racial segregation and affirmative action. Settling the matter with data, however, is extremely difficult because disentangling the various cause-and-effect relationships requires one to keep track of individuals, networks, and groups over extended intervals of time.17 And historically, that sort of data just hasn’t been available. Communication technologies like e-mail, however, have the potential to change all that. Because reciprocated e-mails for the most part represent real relationships, it is possible to use e-mail exchanges as a way to observe underlying social networks. And because e-mail servers can easily log the interactions among thousands or even millions of individuals over long periods of time, it is possible to reconstruct the evolution of even very large networks in great detail. Combine this sort of information with other data that is routinely collected by firms, universities, and other organizations about their members, and a rough approximation of the more complete picture starts to emerge.
Recently my former graduate student Gueorgi Kossinets and I used exactly this kind of approach to study the origins of homophily within the students, faculty, and staff of a university community. As with previous studies, we found that acquaintances—meaning people who exchanged e-mail on a regular basis—were considerably more similar on a range of attributes such as age, gender, academic major, and so on than strangers. We also found that similar people who were not acquainted were more likely than dissimilar people to connect to each other over time—just as common sense would contend. Finally, however, we found that individuals who were already “close” to each other, either because they shared mutual friends or belonged to the same groups, were more similar than distant pairs, and that most of the bias toward connecting similar individuals disappeared once we accounted for the effects of proximity. Our conclusion was that although the individuals in our community did exhibit some preference for others who were similar, it was a relatively weak preference that had been amplified over time, by successive “rounds” of choices, to generate the appearance of a much stronger preference in the observed network.18
Another problem to do with homophily that the Internet may help to answer is one that political scientists and sociologists have long worried about—namely, that Americans, whether by choice or by circumstance, are increasingly associating with like-minded neighbors and acquaintances. If true, the trend is thought to be problematic, as homogeneous social circles can also lead to a more balkanized society in which differences of opinion lead to political conflict rather than exchanges of ideas among equals. But is there actually any such trend? Political scientists generally agree that Congress is indeed more polarized now than at almost any point in history, and that the media is not much better. However, studies of polarization among ordinary citizens have tended to reach conflicting conclusions: Some find that it has increased dramatically while others point to levels of agreement that have changed little in decades.19 One possible explanation for these contradictory results is that people think that they agree with their friends much more than they actually do; thus much of the polarization may be perceived rather than real. But testing this hypothesis, although simple in theory, is difficult in practice. The reason is that in order to measure whether friends agree as much as they think they do, one would need to ask, for every issue of interest, and for every pair of friends A and B, what A thinks about the issue, what B thinks about the issue, and what A thinks B thinks about it. Do this for lots of issues and many pairs of individuals, and you have a tremendously laborious survey exercise, especially if you also have to get each respondent to name friends and then go track them down.20
On Facebook, however, it’s relatively straightforward. Everyone has already declared who their friends are, and it’s even possible to differentiate different strengths of friendships, by counting how many mutual friends they share.21 Equally important, in 2007 Facebook launched their third-party developer “platform,” which allowed outside programmers to write their own applications that would then “run” on Facebook’s underlying network. The result was Friend Sense, an application that my Yahoo! colleagues Sharad Goel and Winter Mason built over the course of a few weeks in early 2008 that asked people what they thought about a range of social and economic issues, and also what they thought their friends thought about them. By Facebook standards, Friend Sense was a modest success—about 1,500 people signed up to use it, generating nearly 100,000 responses. But by network survey standards it was quite large. Using traditional interview methods, a study of this scale would have taken a couple of years to plan, fund, and run, and would have cost a couple of hundred thousand dollars (mostly to pay interviewers). On Facebook, we spent a few thousand dollars for ads and had our data in a matter of weeks.
What we found was that friends are indeed more similar than strangers, and that close friends and friends who say they talk about politics are more similar than casual acquaintances—just as the homophily principle would predict. But friends, whether close or not, also consistently believe themselves to be more similar than they actually are. In particular, our respondents were very bad at guessing when one of their friends—even a close friend with whom they discussed politics—disagreed with them. Here, the numbers were borne out by a series of anecdotal reports we received from people who had participated in Friend Sense, and who were frequently dismayed by how their friends and loved ones perceived them: “How could they think that I thought that?” was a frequent refrain. Many of our participants also reported having the experience of being asked a question about someone they thought they knew well, only to realize that they didn’t know the answer—even though it seemed like a subject that educated, politically engaged friends ought to be talking about.22
So if talking about politics only slightly improves our ability to detect when our friends disagree with us, then what exactly are we talking about? More to the point, in the absence of specific information about what our friends really think about particular issues, what information are we using to guess what they think? By conducting a series of additional analyses, we concluded that when in doubt, which is more often than we’d like to admit, we guess at our friends’ views in part by using simple stereotypes—for example, “my friends are mostly left-wing liberal types, and so probably espouse typical left-wing liberal views”—and in part by “projecting” our own views onto them.23 This last finding is potentially important, because, going back to Lazarsfeld, social and marketing scientists alike have long thought that changes in political opinions are determined more by word-of-mouth influence than by what people hear or see in the mass media. But if it turns out that when thinking about their friends’ beliefs people are really just seeing their own beliefs reflected back at them, one has to wonder how much they can really be influenced. Friend Sense, unfortunately, wasn’t designed to answer questions about social influence, but other researchers have already begun to run experimental studies of influence on Facebook, so hopefully we will have better answers soon.24
Of course, there are all sorts of problems associated with using electronic records, whether they are derived from e-mail exchanges or Facebook friends, as substitutes for “real” social connections. How do we know, for example, what kind of relationship is implied by a connection on Facebook, or how much of all the communication between two people is captured by their e-mail exchanges? I may e-mail my coworkers many times a day and my mother only once or twice a week, but this observation alone says little about the nature and relative importance of these relationships. I may use e-mail to conduct some interactions while preferring text messages, Facebook, or face-to-face meetings to conduct others. And even where I do use the same medium to communicate with different people, some acts of communication may simply be more meaningful than others. From communication frequency alone, therefore, there is only so much one can infer about a given relationship. Nor is it even clear exactly which relations
hips we ought to care about in the first place. For some purposes, like understanding the efficiency of work-related teams, ties between coworkers might be the only ones that matter, whereas for other purposes, like understanding religious or political beliefs, work relationships might be far less relevant. To discover how a fast-moving rumor spreads, it may matter only with whom you have communicated in the past few days, whereas for information that can spread only through networks of trust, the ties that matter may be those that have persisted for years.25
There are many unresolved issues like these that currently limit our ability to draw meaningful sociological inferences from electronic data, no matter how much of it we can acquire. Sheer quantity on its own is certainly not a panacea. Nevertheless, the rapidly increasing availability of observational data, along with the ability to conduct experiments on a previously unimaginable scale, is allowing social scientists to imagine a world where at least some forms of collective human behavior can be measured and understood, possibly even predicted, in the way in which scientists in other fields have long been accustomed.
MESSY MATTERS
It isn’t clear where these new capabilities will lead social science, but it probably won’t be to the kind of simple universal laws that social theorists like Comte and Parsons dreamt of. Nor should it, for the simple reason that the social world probably isn’t governed by any such laws. Unlike gravity, which works the same way at all times and in all places, homophily originates partly out of psychological preferences and partly out of structural constraints. Unlike mass and acceleration, which are defined unambiguously, influence is sometimes concentrated and sometimes distributed, while success derives from a complicated mix of individual choices, social constraints, and random chance. Unlike physical forces, which can be neatly summed to determine their action on a mass, performance is driven by some complicated interaction between extrinsic incentives and intrinsic motivation. And unlike physical reality, which operates with or without us, there is no dissociating social “reality” from our perception of it, where perceptions are driven as much by our own psychological biases as by externally observable attributes.
The social world, in other words, is far messier than the physical world, and the more we learn about it, the messier it is likely to seem. The result is that we will probably never have a science of sociology that will resemble physics. But that’s OK. Just because physics has experienced such great success on the strength of a small number of very general laws doesn’t mean that that’s the only way for science to proceed. Biology doesn’t really have universal laws either, and yet biologists still manage to make progress. Surely the real nature of science is not to exhibit any particular form at all, but rather to follow scientific procedures—of theory, observation, and experiment—that incrementally and iteratively chip away at the mysteries of the world. And surely the point of these procedures is not to discover laws of any particular kind, but rather to figure things out—to solve problems. So the less we worry about looking for general laws in social science, and the more we worry about solving actual problems, the more progress we are likely to make.
But what kinds of problems can we hope to solve? More to the point—to return to the question that I raised in the Preface—what can social scientists hope to discover that an ordinary intelligent person couldn’t figure out on his or her own? Surely any thoughtful person could figure out just by introspection that we are all influenced by the opinions of our family and friends, that context matters, and that all things are relative. Surely such a person could know without the aid of social science that perceptions matter, or that people care about more than just money. Likewise, a moment’s introspection would suggest that success is at least partly luck, that prophecies can be self-fulfilling, and that even the best-laid plans tend to suffer from the law of unintended consequences. Any thoughtful person knows, of course, that the future is unpredictable, and that past performance is no guarantee of future returns. He or she would also know that humans are biased and sometimes irrational, that political systems are rife with inefficiencies and contradictions, that spin sometimes trumps substance, and that simple stories can obscure complicated truths. He or she may even know—or at least have heard it enough to believe it—that everyone is connected to everyone else by just “six degrees of separation.” When the subject is human behavior, in other words, it is actually hard to imagine anything that social scientists could possibly discover that wouldn’t sound obvious to a thoughtful person, no matter how difficult it might have been to figure it out.
What isn’t obvious, however, is how all these “obvious” things fit together. We know, for example, that people influence each other, and we know that hit movies, books, and songs are many times more successful than the average. But what we don’t know is how—or even if—the forces of social influence operating at the level of the individual drive inequality and unpredictability at the scale of entire markets. Likewise, we know that people in social networks tend to cluster together in relatively homogeneous groups. But what we can’t infer from our own observations of the world is whether these patterns are driven by psychological preferences or structural constraints. Nor is it obvious that it is because of this local clustering, rather than in spite of it, that individuals can navigate through very large networks to reach distant strangers in only a small number of steps. At some level, we accept that the future is unpredictable, but we do not know how much of that unpredictability could be eliminated simply by thinking through the possibilities more carefully, and how much is inherently random in the way that a roll of the dice is random. Even less clear to us is how this balance between predictability and unpredictability ought to change the kinds of strategies we deploy to prepare for future contingencies, or the kinds of explanations we come up with for the outcomes we observed.
It is in resolving these sorts of puzzles that social science can hope to advance well beyond where we can get on the strength of common sense and intuition alone. Better yet, as more such puzzles get resolved, it may turn out that similar sorts of mechanisms come into play in many of them, leading us, perhaps, to the kind of “middle-range” theories that Robert Merton had in mind back in the 1960s. What can we learn by studying social influence in cultural markets that can also tell us something useful about the relationship between financial incentives and individual performance? How, for example, can we connect our findings about the difference between real and perceived similarity in political attitudes with our findings about the origins of similarity in social networks? What can these findings in turn tell us about social influence and collective behavior? And how can we connect network search and social influence, decision making, incentives and performance, perceptions and polarization with the “big” questions of social science—like inequality, social justice, and economic policy?
It isn’t clear that we can. Almost certainly, some of the problems that sociologists and other social scientists find interesting will lie forever beyond the reach of precise measurement. No matter how much the Internet and other new technologies affect their field, therefore, the traditional tools of social scientists—archival research, fieldwork, theoretical models, and deep introspection—will continue to play important roles. Nor is it necessarily the case that the most complicated and pressing real-world problems—such as achieving consensus around matters of social justice or designing institutions that cope with uncertainty—can ever be “solved” in an engineering sense, no matter how much basic science we acquire. For problems like these, we can still discover that some solutions work better than others—using, for example the kind of bootstrapping and experimental approaches outlined in Chapter 8 or the deliberative approach to democracy that political philosophers such as John Rawls and Michael Sandel have long advocated. But the exact cause-and-effect mechanisms may remain forever elusive.
Ultimately, we will probably need to pursue all these approaches simultaneously, attempting to converge on an understanding of how people b
ehave and how the world works both from above and from below, bringing to bear every method and resource that we have at our disposal. It sounds like a lot of work, and it will be. But as Merton noted four decades ago, we have done this kind of thing before, first in physics and then in biology and then again in medical science. Most recently, the genomics revolution that began more than fifty years ago with the discovery of DNA has long promised more in the way of medical treatments than it has been able to deliver; yet that hasn’t stopped us from devoting enormous resources to the pursuit of science.26 Why should the science required to understand social problems such as urban poverty or economic development or public education deserve less attention? It should not. Nor can we claim anymore that the necessary instruments don’t exist. Rather, just as the invention of the telescope revolutionized the study of the heavens, so too by rendering the unmeasurable measurable, the technological revolution in mobile, Web, and Internet communications has the potential to revolutionize our understanding of ourselves and how we interact. Merton was right: Social science has still not found its Kepler. But three hundred years after Alexander Pope argued that the proper study of mankind should lie not in the heavens but in ourselves, we have finally found our telescope.27 Let the revolution begin.…
Everything Is Obvious Page 25