What do we really know?
The phenomenon of social contagion—that information, ideas, and even behaviors can spread through networks of people the way that infectious diseases do—is both intuitively appealing and potentially powerful.
It appeals to our intuition for two reasons. First, it is obviously true that people are influenced by one another. Reflecting on our individual experience of life, it is easy to recall any number of instances in which we have been influenced, whether by our parents, our teachers, our coworkers, or our friends, and corresponding instances when we have influenced them. And second, once you accept that one person can influence another, it follows logically that that person can influence yet another person, who can in turn influence another person, and so on. Influence, that is, can spread.
Its potential power arises mostly from this second idea. We know that in the world of infectious disease, global pandemics, infecting millions of people — the Spanish Flu, HIV, and maybe one day Avian Influenza—can be triggered by a single individual, a “patient zero,” from which all subsequent infections are derived. If social influence can spread like a disease, then it is only natural to suspect that “social epidemics” can take place as well, and that they too have their patient zeros who trigger them.
In the 19th century, writers like Charles Mackay and Gustave Le Bon viewed social contagion with alarm, seeing it as the cause of collective madness, whether in financial markets or mob violence. By the arrival of the 21st century, however, the prevailing view of contagion had become far more positive—particularly in marketing and related fields. If only an enterprising marketer (or some other change agent) could create the conditions for a social epidemic, the reasoning goes, and if only they could find the right people to trigger it, awesome change could be unleashed for relatively little cost.
One belief that hasn’t changed over time, however, is that social epidemics are responsible for dramatic, possibly sudden social change. But is this assumption really true? And if not, then what exactly can social contagion accomplish?
The honest answer is that nobody knows, but the real picture is probably less frightening, and also less awe-inspiring, than we have been led to believe.
The first wrinkle is that much of what we attribute to social influence probably springs from other sources. When we look at patterns of ordinary social life—accents, fashions, social norms, etc.—we see striking correlations in space and time, and whenever we see these patterns, it’s tempting to attribute them to influence. As the statistician and complexity theorist Cosma Shalizi is fond of saying, there’s a reason that people sound different in Boston than in Pittsburgh, and it isn’t because the waters of the Charles River are somehow different from those of the Allegheny.
But as Shalizi also argues, the fingerprints of influence are rarely this easy to identify. In most real-world settings, the tendency of acquaintances to behave in similar ways can easily arise in the absence of influence. People who were similar before they met were also more likely to choose to live in the same place or to work in the same occupation, which in turn dramatically increased the chances of them meeting. Or they may behave in similar ways not because they’re influencing one another, but rather because they are exposed to similar outside influences. As the great sociologist Max Weber wrote, if you see a crowd of people all put up their umbrellas at the same time, you don’t assume that social influence is responsible.
No one doubts that influence is an important cause of correlated behavior, but as recent debates in the academic community have highlighted, it’s surprisingly hard to prove it. And even when social influence is indisputably present, it doesn’t necessarily spread in anything like the way that infectious diseases do—it may not even spread at all.
In a recent study of online diffusion, my colleagues Sharad Goel and Dan Goldstein at Yahoo! Research and I found that over several different domains—including every video and news story posted on Twitter over a month-long period—roughly 90% of all contagion terminated within one degree of the originator. We did not find evidence of the kind of social epidemics that get marketers so excited.
In another experiment where subjects arrayed in a network played a game of cooperation, my collaborator Sid Suri and I found even less evidence of contagion. Although individuals increased their contributions in response to generous neighbors (really artificial agents played by us), the positive effect did not propagate any further.
As is always the case in science, social influence research has to strike a balance between the questions we would ideally like to ask and those we are in a position to answer.
What’s going on? Again, we don’t know for sure, but we suspect that the analogy with biological disease is badly flawed. For example, whereas it is probably true that most people are susceptible to HIV, our susceptibility to any particular idea, product, musical artists, etc. varies tremendously, depending on our tastes, backgrounds, and circumstances. Unlike for influenza, to which you’re either exposed or not exposed, even the ideas you do encounter have to compete for attention with everything else that you’re exposed to. And unlike models of disease, which assume that disease spreads exclusively from person to person, information can be disseminated by the media and advertising as well as by word of mouth.
All of these differences, along with many others, could dramatically alter the prospects for social epidemics, as well as introduce other mechanisms entirely by which social change can come about, yet models of social influence reflect very little of this added complexity (I know, because I am responsible some of these models).
That’s the bad news. The good news is our understanding of social contagion is likely to improve dramatically in the near future, thanks to an avalanche of new data, much of it generated by social media platforms.
Historically, iit has been very hard to study social contagion empirically. What you need is evidence that individual A has “infected” individual B, but typically all you have is aggregated data at the level of the population: how many iPods have been sold in US, or how many times a YouTube video has been viewed.
Thanks to the technological revolution of the Internet, however, this situation is undergoing rapid transformation. Researchers have recently used social platforms such as Facebook and Twitter to track the diffusion of individual pieces of content over interpersonal networks on a massive scale. Even more recently, Facebook’s data science team has conducted a “field experiment” that, through random manipulations to their newsfeed algorithm, shed new light on the mechanics of interpersonal influence with respect to user actions such as “likes.”
One criticism of these studies is that retweets and likes are relatively trivial actions, and thus the results tell us little about “real” social influence. Whether or not this criticism is fair will only be known when we can execute studies of this type for more consequential behaviors like shopping or voting or losing weight. As is always the case in science, social influence research has to strike a balance between the questions we would ideally like to ask and those we are in a position to answer. Two outcomes seem likely however: first, the science of social influence will progress in leaps and bounds over the next few years, and second, many of our intuitions about it will prove to be wrong.