Everything Is Obvious

by Duncan J. Watts

  Nor is the intuitive appeal of special-people explanations restricted to problems to do with networks. The “great man” view of history explains important historical events in terms of the actions of a few critical leaders. Conspiracy theorists imbue shadowy government agents or secret cabals with near infinite capabilities to meddle with society. Media analysts credit high-profile celebrities with setting fashion trends or selling products. Corporate boards pay exorbitant amounts for a CEO whose decisions will shape the fate of the entire company. Epidemiologists worry that a few “superspreaders” can trigger an epidemic. And marketers extol the power of “influencers” to make or break a brand, change social norms, or otherwise shift public opinion.11 In his book The Tipping Point, for example, Gladwell explains the origins of what he calls social epidemics, meaning everything from fads and fashions to shifts in cultural norms and sudden drops in crime rates, in terms of what he calls the law of the few. Just as superspreaders drive real epidemics and great men drive history, so too the law of the few claims that social epidemics are “driven by the efforts of a handful of exceptional people.” For example, in discussing the mysterious resurgence of Hush Puppies in the mid-1990s, Gladwell explains that

  the great mystery is how those shoes went from something worn by a few fashion-forward downtown Manhattan hipsters to being sold in malls across the country. What was the connection between the East Village and Middle America? The Law of the Few says the answer is that one of these exceptional people found out about the trend, and through social connections and energy and enthusiasm and personality spread the word about Hush Puppies just as people like Gaeten Dugas and Nushawn Williams were able to spread HIV.12

  Gladwell’s law of the few is catnip to marketers and businessmen and community organizers and just about anyone else in the business of shaping or manipulating people. And it’s easy to see why. If you can just find these special people and influence them, their connections and energy and enthusiasm and personality would be put to work for you. It’s a plausible-sounding story, and yet as with so many appealing ideas about human behavior, the law of the few turns out to be more a matter of perception than reality.

  THE INFLUENCERS

  The culprit again is common sense. As marketing consultants Ed Keller and Jon Berry argue, “Some people are better connected, better read, and better informed. You probably know this from your own experience. You don’t turn to just anyone when you’re deciding what neighborhood to live in, how to invest for retirement, or what kind of car or computer to buy.”13 As a description of our perceptions, this statement is probably accurate—when we think about what we’re doing when we seek out information, access, or advice, it does indeed seem that we focus on some people over others. But as I’ve already discussed, our perceptions of how we behave are far from a perfect reflection of reality. A number of studies, for example, have suggested that social influence is mostly subconscious, arising out of subtle cues that we receive from our friends and neighbors, not necessarily by “turning to them” at all.14 Nor is it clear that when we are influenced in these other, less conscious ways, we recognize that we have been influenced. Employees, for example, may well influence their bosses as much as their bosses influence them, but they are not equally likely to name each other as sources of influence—simply because bosses are supposed to be influential, whereas employees are not. In other words, our perceptions of who influences us may say more about social and hierarchical relations than influence per se.

  One of the most confusing aspects of the influencer debate, in fact, is that no one can really agree on who the influencers are in the first place. Originally the term referred to “ordinary” people who just happened to exert extraordinary influence over their friends and neighbors. But in practice all sorts of people are referred to as influencers: media giants like Oprah Winfrey; gatekeepers like Anna Wintour, the editor of Vogue; celebrity actors and socialites; popular bloggers; and so on. All of these people may or may not be influential in their own way, but the kind of influence they exert varies tremendously. Oprah Winfrey’s advocacy of an unknown book, for example, may dramatically improve its chances of appearing on the bestseller lists, but that is mostly because her individual influence is magnified enormously by the media empire that she runs. Likewise, a fashion designer might be well advised to have a famous actress arrive at the Oscars wearing his dress, but that is again because her arrival is being recorded, broadcast, and commented upon by the mass media. And when a popular blogger expresses his enthusiasm for a particular product, potentially thousands of people read his opinion. But is his or her influence analogous to that of an Oprah endorsement, a personal recommendation from a friend, or something else?

  Even if we narrow down the problem to direct, interpersonal influence of the kind that excludes the media, celebrities, and bloggers of the world, measuring influence is a lot more difficult than simply measuring the length of message chains. For example, to demonstrate just one incident of influence between two friends, Anna and Bill, you need to demonstrate that whenever Anna adopts a certain idea or product, Bill is more likely to adopt the same idea or product as well.15 Even keeping track of just one such relationship would not be easy. And as researchers quickly discovered, doing it for many people simultaneously is prohibitively difficult. In place of observing influence directly, therefore, researchers have proposed numerous proxies for influence, such as how many friends an individual has, or how many opinions they voice, or how expert or passionate they are about a topic, or how highly they score on some personality test—things that are easier to measure than influence itself. Unfortunately, while all these measures are plausible substitutes for influence, they all derive from assumptions about how people are influenced, and no one has ever tested these assumptions. In practice, therefore, nobody really knows who is an influencer and who isn’t.16

  This ambiguity is confusing, but it’s still not the real source of the problem. If we could invent a perfect instrument for measuring influence, presumably we would find that some people are indeed more influential than others. Yet some people are also taller than others and that is not necessarily something about which marketers should care. So why are they so excited about influencers? Consider, for example, that many studies count someone as an influencer if at least three acquaintances named them as someone to whom they would turn for advice. Now, in a world where the average person influences just one other person, influencing three others makes you 300 percent as influential as average—a big difference. But on its own it doesn’t solve the kinds of problems that marketers care about, like generating a hit product, driving public health awareness, or influencing a political candidate’s election chances. All these problems require influencing millions of individuals. So even if each one of your influencers can influence three other ordinary people, you will still need to find and influence a million of them, which is rather different from what the law of the few promises. As it turns out, there’s a solution to this problem as well, but it requires that we incorporate another related but distinct idea from network theory—that of social contagion.

  THE ACCIDENTAL INFLUENTIALS

  Contagion—the idea that information, and potentially influence, can spread along network ties like an infectious disease—is one of the most intriguing ideas in network science. As we saw in the last chapter, when everyone is being influenced by what other people are doing, surprising things can happen. But contagion also has important implications for influencers—because once you include the effects of contagion, the ultimate importance of an influencer is not just the individuals he or she influences directly but also all those influenced indirectly, via his neighbors, his neighbors’ neighbors, and so on. It is through contagion, in fact, that the law of the few gets its real power. Because if just the right influencers can trigger a social epidemic, then influencing four million people may in fact require only a few of them. That’s not a good deal—that’s a great deal. And because finding and influencing
just a few people is quite different from finding and influencing a million, it qualitatively changes the nature of influence.17

  What it means, though, is that the law of the few is not one, but two hypotheses that have been mashed together: first that some people are more influential than others; and second, that the influence of these people is greatly magnified by some contagion process that generates social epidemics.18 It was therefore this combination of claims that Peter Dodds and I set out to test a few years ago in a series of computer simulations. Because these simulations required us to write down explicit mathematical models of how influence spreads, we had to specify all the assumptions that are typically left unstated in anecdotal descriptions of influencers. How should an influencer be defined? Who influences whom? What kinds of choices are individuals making? And how are these choices influenced by others? As I’ve discussed, no one really knows the answers to any of these questions; thus it’s necessary, as in any modeling exercise, to make a number of assumptions, which could of course be wrong. Nevertheless, to cover our bases as much as possible, we considered two very different classes of models, each of which has been studied for decades by social and marketing scientists.19

  The first was a version of Granovetter’s riot model from the previous chapter. Unlike Granovetter’s model, however, where everyone in the crowd observed everyone else, the interactions among individuals were specified by a network in which each individual got to observe only some relatively small circle of friends or acquaintances. The second model was a version of the “Bass model,” named for Frank Bass, the marketing scientist who first proposed it as a model of product adoption, but closely related to an even older model used by mathematical epidemiologists to study the spread of biological diseases. In other words, whereas Granovetter’s model assumes that individuals adopt something when a certain fraction of their neighbors do, the Bass model assumes that adoption works like an infection process, with “susceptible” and “infected” individuals interacting along network ties.20 The two models sound similar, but they’re actually very different—which was important, because we didn’t want our conclusions about the effect of influencers to depend too much on the assumptions of any one model.

  What we found was that under most conditions, highly influential individuals were indeed more effective than the average person in triggering social epidemics. But their relative importance was much less than what the law of the few would suggest. To illustrate, consider an “influencer” who directly influences three times as many of his peers as the average person. Intuitively, one would expect that, all other things being equal, the influencer would also influence three times as many people indirectly as well. In other words, the influencer would exhibit a “multiplier effect” of three. The law of the few, it bears noting, claims that the effect would be much greater—that the disproportionality should be “extreme”—but what we found was the opposite.21 Typically the multiplier effect for an influencer like this was less than three, and in many cases, they were not any more effective at all. Influencers may exist, in other words, but not the kind of influencers posited by the law of the few.

  The reason is simply that when influence is spread via some contagious process, the outcome depends far more on the overall structure of the network than on the properties of the individuals who trigger it. Just as forest fires require a conspiracy of wind, temperature, low humidity, and combustible fuel to rage out of control over large tracts of land, social epidemics require just the right conditions to be satisfied by the network of influence. And as it turned out, the most important condition had nothing to do with a few highly influential individuals at all. Rather, it depended on the existence of a critical mass of easily influenced people who influence other easy-to-influence people. When this critical mass existed, even an average individual was capable of triggering a large cascade—just as any spark will suffice to trigger a large forest fire when the conditions are primed for it. Conversely, when the critical mass did not exist, not even the most influential individual could trigger any more than a small cascade. The result is that unless one can see where particular individuals fit into the entire network, one cannot say much about how influential they will be—no matter what you can measure about them.

  When we hear about a large forest fire, of course, we don’t think that there must have been anything special about the spark that started it. Indeed, such an idea would be laughable. Yet when we see something special happen in the social world, we are instantly drawn to the idea that whoever started it must have been special also. And of course, whenever a large cascade did take place in our simulations, it was necessarily the case that someone had to have started it. However unexceptional that person might have seemed in advance, in retrospect they would seem to fit exactly the description of the law of the few: the “tiny percentage of people who do the majority of the work.” What we knew from our simulations, however, was that there really was nothing special about these individuals—because we had created them that way. The majority of the work was being done not by a tiny percentage of people who acted as the triggers, but rather by the much larger critical mass of easily influenced people. What we concluded, therefore, is that the kind of influential person whose energy and connections can turn your book into a bestseller or your product into a hit is most likely an accident of timing and circumstances. An “accidental influential” as it were.22

  “ORDINARY INFLUENCERS” ON TWITTER

  As many people immediately pointed out, this conclusion was based entirely on computer simulations. And as I’ve already mentioned, these simulations were highly simplified versions of reality, and made a large number of assumptions, any of which could have been wrong. Computer simulations are useful tools that can generate great insight. But in the end they are more like thought experiments than real experiments, and as such are better suited to provoking new questions than to answering them. So if we really want to know whether particular individuals are capable of stimulating the diffusion of ideas, information, and influence—and if these influencers exist, which attributes distinguish them from ordinary people—then we need to run experiments in the real world. But studying the relationship between individual influence and large-scale impact in the real world is easier said than done.

  The main problem is that you need an enormous amount of data, and most of it is very hard to collect. Just demonstrating that one person has influenced another is difficult enough. And if you wanted to make the connection to how they influence larger populations, you need to gather similar information for whole chains of influence, in which one person influences another who in turn influences another, and so on. Pretty soon, you’re talking about thousands or even millions of relationships, just to track how a single piece of information was spread. And ideally you would want to study many such cases. It’s an over-whelming amount of data to test what seems to be a relatively straightforward claim—that some people matter more than others—but there’s no getting around it. It also helps explain why diffusion research, as it is known, has remained such a myth-laden business for so long: when it’s impossible to prove anything, everyone is free to propose whatever plausible story they like. There’s no way to decide who is right.

  As with experiments like Music Lab, however, the Internet is starting to change this picture in important ways. A handful of recent studies have begun to explore diffusion in social networks on a scale that would have been unimaginable just a decade ago. Blog postings diffuse among networks of bloggers. Fan pages diffuse among networks of friends on Facebook. Special capabilities called “gestures” diffuse among players on the online game Second Life. And premium voice services have been shown to diffuse among networks of IM buddies.23 Inspired by these studies, my Yahoo! colleagues Jake Hofman and Winter Mason and I, along with Eytan Bakshy, a talented graduate student at the University of Michigan, decided to look for the diffusion of information in the largest communication network we could get our hands on: Twitter. In the
process, we would look for influencers.24

  In many respects, Twitter is ideally suited to this objective. Unlike Facebook, say, where people connect to one another for a multitude of reasons, the whole point of Twitter is to broadcast information to other people—your “followers”—who have explicitly indicated that they want to hear from you. Getting people to pay attention to you—influencing them, in other words—is what Twitter is all about. Second, Twitter is remarkably diverse. Many users are regular people whose followers are mostly friends interested in hearing from them. But many of the most followed users on Twitter are public figures, including bloggers, journalists, celebrities (Ashton Kutcher, Shaquille O’Neal, Oprah), media organizations such as CNN, and even government agencies and nonprofits (the Obama administration, No. 10 Downing Street, the World Economic Forum). This diversity is helpful because it allowed us to compare the influence of all manner of would-be influencers—ordinary people all the way up to Oprah and Ashton—in a consistent way.

  Finally, although many tweets are mundane updates (“Having coffee at Starbucks on Broadway! It’s a beautiful day!!”), many of them refer either to other online content, like breaking news stories and funny videos, or to other things in the world, like books, movies, and so on, about which Twitter users wish to express their opinions. And because the format of Twitter forces users to keep every message to no more than 140 characters, users often make use of “URL shorteners,” such as bit.ly, to replace the long, messy URL of the original website with something like http://bit.ly/beRKJo. The nice thing about these shortened URLs is that they effectively assign a unique code to every piece of content broadcast on Twitter. Thus when a user wishes to “retweet” something, it’s possible to see whom it came from originally, and thereby trace chains of diffusion across the follower graph.