What Do We Learn from Our Networks?

Some workplaces are happy, productive places; others, not so much. And sometimes, for what seems like no reason at all, your office goes from one to the other.

According to Nicholas A. Christakis, the Sol Goldman Family Professor of Social and Natural Science at Yale, the dynamics that shape workplaces are complex and ancient. We have been forming social networks for thousands of years, and they have a surprisingly strong effect on our ideas and behavior. You may think that it was your own idea to get to work on time or keep your desk neat or speak up in meetings. But it’s likely that your behavior, for better or worse, was influenced by those in your network—and that you’ll influence others in turn.

The good news is that once you understand them, Christakis says, you can harness the power of social networks to shape workplaces for the better. By tweaking the makeup of a team, or by seeding an idea with a particularly influential individual, you can turn an unhappy team into an innovative, collaborative one.

Q: As a researcher, what do you mean by “social network”?

Well, the difference between a group and a network is that a group is just a collection of individuals. But a network, in addition to the constituent individuals, also has ties between people, also has connections—a specific number of ties and a specific pattern of ties. The ties between the individuals are what converts a group to a network.

The interesting thing is that we humans have been making networks for tens of thousands of years. The recent online instantiations of networks are grafted onto this very ancient past. The telecommunications networks that we’ve invented over the last 100 years and our internet networks over the last 30 years or so and our social media networks over the last 10 years—all of these are actually in the service of a rather ancient predilection that all of us humans have to connect with each other, to learn from each other, to influence each other, and so forth. So for me, therefore, a “social network” is a general term. It’s not restricted to online networks. It’s something deep and fundamental about our humanity that’s been there, frankly, since time immemorial.

Q: Are we distinct from other animals in this regard?

Human beings are very unusual as a species in that we form long-term, non-reproductive unions to other members of our species. Namely, we have friends.

It’s not hard to provide an account in evolutionary biology as to why we mate with each other. It’s rather more difficult to provide an account for why we befriend each other. Very few species do this. We do, and so do certain other primates, elephants (with whom our last common ancestor was 65 million years ago and who appear to have evolved the capacity for friendship independently of us), and certain cetaceans. So it’s very rare in the natural world that creatures have what we might think of as friendships (though it’s not unheard of).

My lab has been doing some work on the extent to which elephant networks, for example, look very similar mathematically to human networks. They do, and this suggests that there’s something very deep and fundamental about the mathematical structure of social networks that undergirds our sociality. That is to say, there may only be particular ways of being a social mammal. So we’re not that special after all, even in this regard.

Q: Do you think we underestimate the extent to which our networks affect us, our ideas, and our behaviors?

There was a social critic and longshoreman by the name of Eric Hoffer who said, “When human beings are free to choose anything they want, they typically copy their neighbors.” It’s one of the deep ironies that we think of ourselves as being individuals moving through the world, and as having the capacity for free will, but, in fact, most of us just copy what others are doing. In fact, this too is a very deep and fundamental part of our humanity. This capacity for social learning, this desire to be like others, is actually quite valuable—and inescapable.

Some people have looked at the work we’ve done over the last 15 years or so and said that it’s very dispiriting, because again and again we show that things that are seemingly so personal as your body size or as your emotions or your financial habits are actually very strongly influenced by like traits in other individuals. That is to say, your experience in the world is not only a product of your own desires and actions and thoughts but also a product of the desires and actions and thoughts of other people around you.

Now, they’re not mutually exclusive. You have agency. You can choose what to do. But you’re also affected by what others are doing. Both are true.

Some people have looked at the extent to which we documented the role of social interactions influencing your thoughts, actions, and behaviors and said that we have “delivered a whack to free will,” that we’ve subverted the role of the individual in behavior. That’s true. But even while we are showing the extent to which free will is not as important as you might have thought, in parallel we’re lifting it up, because our work has shown that when you make a positive change in your life—when you quit smoking, when you go vote, when you recycle, when you save for retirement, when you pay your taxes dutifully—other people around you copy you. In many circumstances, when you make a choice in your life, you can affect dozens, hundreds, sometimes thousands of other people based on your actions. So free will is also important.

Q: What should leaders of organizations know about those effects if there are certain behaviors that they want to spread in their organizations?

I’d like to answer that question first by invoking a metaphor and then by coming to two broad kinds of strategies that people might think about.

Most of us in high school chemistry learned that carbon exists in many forms. These are known as the “allotropes” of carbon. The classic ones, of course, are graphite and diamond. A pencil lead and a diamond are different because the carbon atoms are assembled in very different structures; in the graphite, the carbon atoms are assembled in hexagonal sheets and, in the diamond, the carbon atoms are assembled into tetrahedrons. The graphite is soft and dark, and the diamond is hard and clear.

There are two key intellectual ideas there. The first is that softness and darkness and hardness and clearness aren’t properties of the carbon atoms. They’re properties of the collection of carbon atoms. Second, which properties you get depends on how you connect the carbon atoms to each other. Connect them one way, you get one set of properties. Connect them another way, you get a different set of properties.

And, it’s the same with social groups or firms. We’ve shown this experimentally in my lab. You can take a group of people and connect them one way, and they’re healthy, happy, cooperative, innovative, and they coordinate their activities. Or you take the same human beings and assemble them a different way, with a different network topology, and they’re none of those things. They’re unhealthy, unhappy, uncooperative, uninnovative, and can’t coordinate worth a damn. It’s the same human beings, but they have these different properties at the level of groups.

These properties—innovation, for example, or coordination or cooperation or health—might therefore be seen not just, or even, as properties of individuals, but rather as properties of collections of individuals, as properties of groups. Groups of people can have properties that transcend the individuals themselves, just like the carbon example, and that these properties are very situation specific.

If I asked you, “Which is better, graphite or diamond?” you would say, “Well, it depends on what you’re trying to do. If you’re trying to write a message to propose to your wife, you’d want a pencil. If you’re trying to give her a gift on the occasion of your proposal, it’s probably better to give her a diamond.”

I have to give the same answer to executives who ask, “What properties should my workplace have?” The answer is: it depends on the objective function. It depends on what it is that you’re trying to achieve. There is no generic answer. But I can tell you that structure matters.

Now, if you want to intervene in structure, there are two broad ways that you can think about that. You can intervene in connection or in contagion. You can intervene in the structure of the network or in the function of the network.

A structural intervention is an intervention in which you manipulate who’s connected to whom. For example, a very simple, health-related example (and almost all my work is in health) would be Alcoholics Anonymous. You have a group of people who are strangers to each other. You bring them together. You form a group where everyone knows everyone else, and this group now has a collective property, namely: the ability to suppress alcohol consumption.

Or you take a group of men and women who previously were strangers to each other, and you bring them together and you put them in military basic training, put them in a squad, and, by the end, they’re willing to die for each other. Here are these people who couldn’t give a damn about each other before, and now are willing to die for each other based on an intervention that you have implemented where (among other things) you manipulate the structure of the interactions.

These collective properties, the alcohol-suppression property or the willingness-to-die property, are emergent properties that did not arise within the individuals themselves but because of how you connected them.

There are many examples of this: in workplaces, you can think about how you arrange offices, what kind of rules you have for how work teams form and reform, and so on. In the latter regard, we’ve done some interesting experimental work showing that you want not too much fluidity, not too much rigidity, in work teams. A middle range is optimal for a variety of objectives. So: the structure matters.

In practice it’s often very difficult to manipulate structure of networks. You can’t force people to be friends with each other typically, except in the army. You can’t force people to cut off their relationships. So, as an alternative, you accept the structure of the network, but you instead thoughtfully intervene in what we call contagion. Rather than manipulating connection, you manipulate contagion, and you intervene in the function rather than the structure of the networks. You might identify structurally influential individuals, people who are located at a location within the social graph such that when they adopt a practice, others copy them. This is very valuable for the diffusion of innovation.

By understanding how things spread within networks, and by using a variety of mathematical tools, it’s possible to identify structurally influential individuals. We’ve done many experiments to show this, often in developing-world settings with public-health interventions, but the same ideas can be deployed in firms that are trying to facilitate the diffusion of innovation, or hospitals that are trying to affect drug-prescribing behavior, or factories that are trying to enhance safety practices by creating artificial tipping points. You say, “Okay, there are 100 or 1,000 people in this village. Who are the 10 or 50 people that I need to persuade to change their behavior so that if they do it, the rest of the village will copy them?”

So I would sum up by saying to managers that it’s very important to understand it’s not just about the human capital in your organization. It’s about the social capital, about the network structure and function in your organization. But there isn’t a specific, generic recommendation of what to do that I can give you. If you ask me, “Which is better, graphite or diamond?” I would say, “What do you want to do?”

Q: How do you go about identifying those people in an organization who are likely to be influential?

It’s possible to map organizational networks using existing data like email traffic, for example, or LinkedIn data. Or you can use some de novo mapping technologies—for example, in my lab we have some software called Trellis that allows you to map villages or organizations. We’ve developed it for public health applications, but it can be used elsewhere (it’s available for commercial and non-commercial use at trellis.yale.edu).

You would also need to define the kinds of interactions you’re interested in. In a firm, for example, you might ask people about many different sorts of ties: “Who do you need to rely on in order to do your job effectively?” or “Who would you trust for a referral for a local babysitter?” or “Who do you hang out with after work?” Those are all different questions and might lead to different structures of the network. Then you could begin to think about, “Okay, which individuals are structurally influential for this or that purpose?” Having done that, there are mathematical rules of thumb you can use to identify structurally influential individuals and then focus your efforts on them.

For example, you might have a group of doctors in a city who are connected to each other through the referral of patients. I share 100 patients with you and 1,000 patients with her and 1 patient with him, and then there are a whole bunch of people that I, as a doctor, am not connected to at all.

Now, say I’m an insurance company and I’m trying to get the physicians to stop ordering needless tests. Well, having mapped the network of physicians, you can identify who are the doctors on whom you should focus your energies to get them to change their test-ordering behavior—because when you get me to stop wasteful testing it affects the doctors to whom I’m connected. You get a ripple effect, a cascade effect, through the network.

But the insurance company wanting to get me to reduce needless testing is a completely different application than the movie production company wanting to get me to encourage my friends to watch a new science fiction movie. I might be very powerful in influencing my friends with prescribing behavior and have no effect at all in my community on what movies are seen as good. Still, in both cases, there are people who can be found who are influential. So once again, there isn’t a one-size-fits-all strategy.

Q: How has how technology affected social networks? Is it just a reflection of existing networks or is it actually forming them and shaping them?

I struggle with this. What role do modern social media technologies play in human social interactions? I think the answer is that online networks are “the same but different” than offline, face-to-face networks.

In the most deep and profound ways, they are the same. We humans still have a deep urge to connect to each other. We humans still are interested in influencing each other and being influenced by each other. As I said before, these modern telecommunications technologies are grafted on to this very ancient set of predilections.

I can illustrate that with just a simple anecdote. If you could talk to my grandmother, who was born about 100 years ago in a little village in Greece and is no longer alive, and ask her how many friends she had when she was a 10-year old girl, she would say, “I had one or two best friends and there were four or five of us girls. We were thick as thieves. We hung out together all the time.” And 100 years later, if you asked my daughter Lena the same question, she would give you the same answer. Even though she has an iPhone in her pocket.

So there’s a way in which the technology hasn’t changed this very fundamental aspect of our sociality at all, any more than the invention of the telephone took us from being a monogamous species to a polygamous species.

But there are things that technology has changed, which we talk about in our book Connected. There are four ways, I would say, in which online networks differ from offline networks. One is simply the scale. The size of the networks is what we call “enormity.” It’s possible to interact with an enormous number of other individuals. Another is the way in which online networks facilitate large-scale cooperation. We would call that “communality.” For example, Wikipedia would be a large-scale communal effort that would be very difficult without online tools. The third thing would be what I would call “specificity.” So, 20 years ago, if you wanted to find a veterinarian in the Norwegian Army, this was a very difficult thing to do. But now, with the internet, you can specifically find such a person very quickly. In fact, as we also discuss in Connected, there are all kinds of online communities that now make it possible for people with esoteric interests and tastes to find each other in a way that previously was not possible. And the fourth thing I would say is “virtuality.” It was always possible for people to pretend they were someone who they were not. For example, cross-dressing was possible 100 years ago, but it is much easier now for a man to have a female avatar and a woman to have a male avatar, or a disabled person to have an able-bodied avatar, than it was in the past.

Q: Has your research shown anything about how inequality affects networks?

We had a paper last year in Nature on economic inequality. We have a software platform we call Breadboard that allows us to do experiments where we create temporary virtual societies of real people online and then we can manipulate the rules of interaction among these individuals to test ideas experimentally about how social systems work. This software is available at http://breadboard.yale.edu (and it can be used for both research and commercial applications—in fact, we are doing some exploratory work with Tata Consulting in the latter regard).

There’s been a lot of discussion about whether inequality is good or bad for our society. People don’t think it’s good, necessarily, but they may think that inequality is a kind of unavoidable side effect of rising wealth, for example. Other people, myself included, think that inequality can be quite corrosive, that it can adversely affect the health of a community, the economic productivity of a community, and so forth.

But it’s very difficult with observational data to be sure about this. So we decided to do an experiment in which we created societies of real people—small-scale societies in which we manipulated the inequality, as measured by the Gini coefficient. We made some of these societies totally equal—the Gini was zero. Some were slightly unequal, like Scandinavia, which is 0.2, and some quite unequal, like the United States, which is 0.4.

Then we picked people at random and dropped them into these little networks with different levels of inequality. We assigned some to be rich and some to be poor, also at random. Furthermore, we manipulated what we call the visibility of the situation, whether you could see the wealth of the other people around you.

So we had six different kinds of worlds: equality, low-level inequality, or high inequality, along with visibility or no visibility. Then we let the people interact with each other to test how bad inequality is for the ability of these societies to produce wealth, and for the cooperativity of these societies, and the friendliness of these societies.

We found some things that were very surprising to me. We actually found that inequality in our experiment was not corrosive to wealth production or (very much) to cooperation or to friendliness. But visible inequality was bad. What really was bad in these societies was the visibility of wealth, not the distribution of the wealth, not the inequality of the wealth.

Now this is a highly stylized experimental situation. The real world might be different. These were small financial stakes, brief interaction over the course of an hour—all very simplified. But nevertheless, these are some very suggestive results.

I think part of the reason we may not have appreciated this as readily before is that, when you go out into the real world and look for inequality, it’s necessarily visible. There aren’t many circumstances in which inequality is invisible. So I think, in the past, people may have been conflating the impact of inequality of wealth with the impact of visibility of wealth.

I also think that our findings have some implications that are important for firms. We found that there was an interaction between inequality and visibility. When equality was high, visibility was not harmful. But when inequality was high, visibility was harmful. What do I mean by that? Imagine you’re thinking about pay transparency in a firm. Basically, if the wage gap between the CEO and the line worker is, let’s say, on the order of five times, then you should publicize everyone’s pay. Everyone should know what everyone else is making. Everyone then becomes happier, more cooperative, and friendlier, and you generate more wealth in our experiments.

Conversely, if inequality is very high and the CEO makes, let’s say, 500 times what the line worker makes, do not publicize the inequality. That’s very corrosive. So the impact of flipping on the visibility switch varies depending on the baseline level of inequality. Our experiments were able to tease that apart.

There are other examples like that. Think about policies regarding uniforms in schools or in workplaces. Basically, those are invisibility cloaks. By implementing uniform policies, you prevent the public display of wealth. Those too can be thought of as tools to modulate the impact of inequality.

Those are just a couple of illustrations of this little experiment that we did, but we increasingly have become interested not just in the structure of networks, not just in the function of networks, but also in how you distribute people with certain attributes within the fabric of a social structure. That, too, can be highly relevant to how social networks affect our lives.

Interviewed and edited by Ben Mattison.