Flu season is rapidly approaching. On that worrisome note - I dread the flu with an irrational passion - there's a excellent new paper in PLoS ONE by Nicholas Christakis and James Fowler. The scientists employ the techniques of social network analysis to better understand the spread of the virus. But they weren't content with theoretical models. Instead, they tracked the spread of the virus in winter 2009 among the Harvard undergrad population, watching as one person infected another. As you'll see below, I think this new paper has some pretty fascinating implications for how we think about contagion. (To summarize: If your friends have lots of friends, then please get a flu vaccine.) I've written about the groundbreaking research of Christakis and Fowler before (and they have an excellent summary of their work in Connected), but the scientists were kind enough to answer a few of my questions about the new paper.
LEHRER: I think most people assume that being at the center of a social network is a good thing. That's where the popular kids are, right? And you've previously shown that having more friends comes with all sorts of benefits. But this paper suggests that being extra-connected also comes with a cost. Could you explain why at the center of a network makes us more likely to catch a contagious illness?
FOWLER AND CHRISTAKIS: That's right. People often ask us 'what is the best type of network to have?' and we often find ourselves giving that very unsatisfying answer, "It depends." Because it does depend -- for example, on what is flowing through the network.
If a deadly germ is flowing through the network, it would be better to be on the edge of the network and/or to have few or no friends. A hermit cannot get a communicable disease.
On the other hand, if a valuable piece of information is flowing through the network (e.g., about how to find a job), then it is better to be in the middle of the network, and/or to have more friends. People at the center are a shorter number of steps from everyone else, so on average they tend to find things out sooner.
And while people who have more friends are at higher risk, it also depends on how many friends their friends have. For example, in the first figure of our new paper on contagious outbreaks, we illustrate intuitively how two different people who both have six friends each might fare very differently in terms of risk of disease, depending on whether their friends are themselves central or peripheral in the network.
Indeed, we argue in our book Connected that there are deep reasons that we do not all of us have the same kind of social network locations. People vary in their locations in social networks in part, we think, because there is no one location that is best, for us as individuals or for us as a species.
Over the course of history it sometimes makes sense to be disconnected (think Black Plague) and sometimes makes sense to be connected (think of our modern economy), and over the long timespan of human evolution this back and forth might have helped to preserve diversity in our networks. Indeed, in another paper we showed that genetic variation can explain as much as half of the variation in how connected we are to one another in our modern day social networks.
LEHRER: It's important to note that this isn't just theoretical. You actually confirmed this model of outbreaks in a study of Harvard students, right?
FOWLER AND CHRISTAKIS: Yes, that is correct. We have been thinking for some time about ways in which insights into how social networks function might actually be put to use. And we have a number of ideas in this regard. When we started reading about the H1N1 epidemic last summer, we wondered if we might be able to do something useful in this area, and we hit upon the idea of using social network theory to help develop a way of anticipating epidemics.
Our thinking in terms of intervening into networks up until then had been focused on the idea of inserting information INTO networks. But then we began to realize that we could turn the problem on its head and extract information FROM networks. That is basically what this project did.
We tracked 740 students daily for 120 days, in a network that we mapped (for a movie of the epidemic click here. But the key here is that we didn't need the network. We compared the group of people who were named as friends to the group of people who were not. Consistent with what social network researchers call the 'friendship paradox', the people who were named as friends were significantly more central in the network than the people who weren't. This is because people with more friends are more likely to be named as a friend.
And we found that the group of named friends got the flu about 14 days ahead of everyone else. That means that we could predict the course of the epidemic in the whole network two weeks in advance just by monitoring the people who were named as friends.
The really cool thing here is that, unlike current methods that focus on giving better information about what's happening today, our method gives better information about what will happen in the future. In other words, this is early detection, not just rapid warning.
LEHRER: What are some of the implications of this research? I imagine, for instance, that this work has direct implications for the distribution of vaccines. How should this paper change the way we fight the next flu epidemic?
FOWLER AND CHRISTAKIS: Absolutely! If we could detect epidemics in advance of their striking the general population, it would provide a sort of generalized advance warning to pubic health experts, such that broad interventions might be implemented. In fact, some of the work that we cite in the paper shows that advance warning might make a huge difference in preventing epidemics in a metropolis like New York City, since it would allow vaccination efforts to be fielded in time. Plus, we could track different parts of the country and get a sense of where the epidemic is peaking sooner.
Incidentally, our method could also be implemented to track non-pathogenic outbreaks, such as outbreaks of behaviors (what kinds of exercise are catching on?), fashions (what new colors are people wearing?), fads (which toy will be this year's Christmas favorite?), ideas (how many people still believe discredited stories about the link between vaccines and autism?), and even political beliefs (will the Tea Party continue to grow?).
LEHRER: We spend so much time treating the individual. But this research suggests that we need to also treat the social network. How might we do this? If you were the head of the CDC, what might you takeaway from this paper?
FOWLER AND CHRISTAKIS: We are thinking of this topic broadly at present. There two basic ways to alter a network, connection and contagion.
To manipulate connection, you have to rewire social networks (cutting and forming ties), and that is potentially quite tricky. How do we affect who is friends with whom? How do we reconnect all those people on the edges of society?
To manipulate contagion, you have to alter what is flowing through the network, for example, by choosing who to target with new information or a behavioral intervention. Many people have been trying this for a long time, but we still don't have a complete understanding of what things 'go viral' and why.
So far there has only been limited research in manipulating connection and contagion. Our current sense is that manipulating contagion is more promising, but we don't really know. The sensor network project, of course, is about extracting information from the network, rather than seeding it, but it is more closely related to the idea of contagion.
The main point we want to make, though, is that modern medicine needs to do a better job of incorporating the web of connections that each of us has. If we can use friends to predict the future, then perhaps we can use them to change it.
