Facebook might be the key to fighting a flu epidemic. And obesity. And crime. And…everything. Nicholas Christakis at Harvard Medical and James Fowler at UC San Diego have found ways in which social networks can predict the spread of trends – from diseases to ideas. By mining data from modern networks we might be able to prevent bad trends (like epidemics) and capitalize on good ones (like happiness). Christakis and Fowler’s book, Connected, explores how important nodes in social networks, the people who have the most connections to other people, can be found and used to predict the rise of trends in the larger group. Watch Christakis explain the benefits of this new prediction method, followed by some eye-opening examples, in the videos below. This is a powerful way of analyzing data and it could revolutionize how we approach healthcare, advertising, education, and more.
In this case, social networks mean all forms of face to face groups: friends, coworkers, towns, etc. We’ve had such social groups since the beginning of time. Modern networks don’t always have the same connectivity (how well do you know all 500 of your Facebook friends?), but they can give insight into the face to face groups. In fact, because they are digital, scientists can collect more data, and more quickly, from modern social groups than they ever could before. Christakis shows how a passive collection of data in a network, mobile phone calls, can rapidly yield a lot of information about social networks (~14:07). What can you do with this new data? In the fall and winter of 2009 Christakis used his social networking methodology to analyze the spread of flu at Harvard College (~8:58). Watching for flu reports among the highly connected people could have predicted the peak of the flu epidemic 16 days ahead of time. Not only that, but watching for when the rate of flu among the highly connected deviated from the norm would have given 46 days advance notice of the epidemic! Think of the money and lives saved if we could use the same prediction methods on a larger scale and for deadlier diseases. Oh wait…I think Christakis and Fowler’s point is that we can.
Predicting the spread of diseases through data mining the internet isn’t a new concept. Many of you may be familiar with Google Flu Trends, which uses search terms to monitor flu levels, and there have been academic papers that have explored the research as well. What makes Christakis and Fowler interesting is how widely they want to apply their social networking analysis. In the same way that Levitt and Dubner revitalized economics by applying it to unexpected areas in Freakonomics, Connected is bringing the ‘important node’ approach of social network analysis to practically everything.
In his presentation above, Christakis suggests applications in information sharing, behavior, politics, product adoption, and vaccination. In the videos below, we see explanations for three easily understood examples: obesity, laughter, and smoking. The first two are cartoonish, while the third is a much drier presentation of a graph.
Are Christakis and Fowler really saying anything new when they tell us that our most popular friends are the ones to watch if we want to predict trends? I mean, we’ve all heard the expression “it’s not what you know, it’s who you know”. Looking to the well connected people and predicting trends seems like a really old idea, every fashion mogul has been doing so for years.
Yet as Fowler tells Stephen Colbert in the video below, it’s not the media figures and celebrities who are establishing the trends in disease, and ideas, it’s the large base of social nodes as a whole. The face to face connections are crucial, and we have to monitor a fairly large number of them to understand how things will spread.
As billions of people come online we’re going to radically increase the number of nodes we can monitor through digital means. More of the world will be easily analyzed through social networks. That means that we have the possibility of really affecting things like the spread of AIDS, hatred and terrorism, and education.
Part of what really struck me in Christakis’ presentation was the description of using social network analysis to maximize the group benefits of vaccination. Giving 96% of the population a vaccine will effectively stop the spread of the disease, even though not everyone is immune. Those who do get the disease have no one to spread it to. If just 30% of the population is vaccinated, the disease can still spread easily through the network. However, if the 30% is targeted to include the most connected social nodes, you can get group protection. Affect the highly connected nodes and you affect the entire group.
Think of how powerful that might be, not just in healthcare but in all manner of applications. Will advertisers pay Facebook to mine data so that they can plan a marketing campaign that targets highly connected nodes? It would get the most advertising bang for their buck. Politicians could do the same thing. So could principles who want to increase good study habits in a school. With the rise of digitalized social networks comes the possibility for quickly and easily exploiting that network. Either for profit or for mutual benefit. Again, we’ve done such targeting for millenia, looking to popular people to establish trends. But now instead of working to influence the most powerful one or two individuals, we can target the broader and more crucial 30% of well connected people. In the future, those who can correctly identify and control those nodes will affect trends and social networks all over the world.