Emotions spread through Facebook

thaodo_facebookrain
Photo by Thao Do.

Professor of Social and Natural Science Nicholas Christakis ’84 researches the ways in which networks impact behavior, health, and longevity. In a recent study released this month in the journal PLOS ONE, Christakis, along with researchers from the University of California and Facebook Inc., discovered that emotions spread through the social networking site just like other phenomena such as disease spread through real networks of people. After analyzing over a billion Facebook statuses, the researchers discovered that positive posts, on average, resulted in 1.75 more positive posts from friends, and negative posts generated 1.29 further negative posts. At Yale, Christakis co-directs the Yale Institute for Network Science (YINS), which opened last summer. The News spoke with Christakis on Monday about the network of emotions on Facebook and how studying network interactions can improve public policy.

Q. I understand that your research focuses on social networks and the ways in which social factors can affect health and longevity. Why did you decide to look at online social networks like Facebook?

A. We humans do this amazing thing. Each of us inherits our relatives, and chooses our friends and our co-workers. All of the people to whom we are connected, in turn, make similar choices. As a result, we fabricate this incredibly ornate structure known as a social network, and we proceed to live out our lives embedded in this network. This fact means, among other things, that we are susceptible to spreading processes in networks. All kinds of things spread in networks — ideas, germs, behaviors, norms and even emotions.

Q. When did you first become interested specifically in how social networking sites like Facebook can influence emotions?

A. The topic of emotional contagion is an old one in psychology. It’s been understood for a long time that there’s a spread of emotions [between people]. For example, if you’re on a subway, if a person across the car smiles at you, it’s normal human behavior to smile back. But we became interested in whether emotions could spread more broadly through enormous networks. We published a paper five years ago, which featured the first network map of human emotions. That was a sample of about 5,000 people, and we looked at face-to-face emotional contagion in networks. But for years we had been meaning to do a bigger project.

Q. In your Facebook study, you found that rain impacts not only the emotional content of those experiencing the weather first hand, but also the emotional content of other statuses in the network. Can you elaborate on how you went about studying this pattern?

A. We used Facebook data — millions of people and tens of millions of connections between them and hundreds of millions of Facebook posts. In collaboration with Facebook, we did natural language processing of their posts and saw whether or not the posts were happy or sad. We didn’t read anyone’s posts.

There are lexicons that have been validated to be either happy or sad. For example, if you said, “I feel great,” then that’s a happy post. We used these validated algorithms to assign posts a rating of whether they were happy or sad. We then linked the posting data for a 1,000 days to weather data for the whole nation for three years, and we found, as others had previously found, that if it rains, you are more likely to be sad and make sad posts. That’s not too shocking, [but] it’s good to validate that.

Then we did a natural experiment and what we found was that, if it rains in your city, your friends in sunny cities are more likely to become sad. Rain in New York City made the people in New York sad, but that sadness spread to people in other cities that were otherwise sunny, who are more likely now to express sad posts as a result of their friends being sad. Thus, we were able to find evidence for emotional contagion on this massive scale.

Q. What were some challenges you encountered in carrying out a study that included data from tens of millions of Facebook users?

A. It is challenging to cope with data on this scale, but there are many people at Yale who are working on using such “big data.” At YINS, people like [YINS Co-Director] Dan Spielman ’92 and professor Sekhar Tatikonda are working on mathematical techniques to cope with such enormous data sets. In my lab, we’re using such techniques to analyze real world problems.

It’s also important to note that this type of work was a natural experiment. What we would love to do is conduct a real experiment, where we take a group of people and experimentally make them happy or sad. Psychologists can do this — they can show you a Charlie Chaplin movie or give you an unexpected candy bar, and it makes you transiently happy. We would like to test whether that has an effect on your friends or your roommates. For example, if you get a good grade on a test, especially if it’s unexpected, does it make your roommates or friends happy? But we couldn’t experimentally manipulate people’s emotional state on this scale.

Q. How is the research team planning to expand on this research? 

A. We’ve been doing a couple of things. We are trying to see whether we can take advantage of the deep understanding of these effects to intervene in the world to make it better. For example, can we induce cascades of cooperation, innovation and vaccine adoption; or, in the developing world, can we take villages and make them more likely to change their practices with respect to clean water? Can we make the world better by taking advantage of social contagion? And we believe the answer is yes, and we have a number of experiments that are showing that is the case.

Q. How can this research be applied in other fields, such as in shaping public policy?

A. First, we need to understand that, when we make public policies that improve people’s well-being, we have to take into account not just the direct effect of the public policy but also the indirect effects. If I have a welfare-improving policy, up until now, the conventional perspective has focused just on the welfare of the individual. But what if I told you that if I took better care of college kids who are depressed, their roommates become happier? Or, what if I have a food stamp program and the food stamp program improves not just the health of the people getting food stamps but also their neighbors? So, there are all these additional benefits that you need to take into account if you take network effects seriously.

Comments