Nicholas Christakis: How Social Networks Predict Epidemics
In his engaging Ted Talk Professor Nicholas Christakis of Yale university uses clever thinking to show how the social networks can predict epidemics before they happen. As of the filming of the talk, the methodology the CDC used to predict epidemics was to record data based on instances of diseases in doctors’ offices nation wide. Then studying the data allowed them to predict where in the nation an epidemic would occur… once it had already happened. Through his study of social networks Nicholas developed a much more efficient and elegant solution to the problem of early epidemic detection.
His idea began with studying the network structures humans live in. Generally he noticed that in a social network, nodes with a higher degree (the degree of a node is equal to the number of connections it has to other nodes) and nodes that are more central are more likely to be infected with an epidemic early on. However, there is a great problem in applying this knowledge to the real world – social networks are often expensive, difficult, or even impossible to map out, thus predicting epidemics based on networks alone is not feasible. However, combining his knowledge of the structure of networks with the friendship paradox does allow for relatively simple, powerful inferences and early epidemic detections.
The friendship paradox is a commonly known paradox in social networking that your friends have more friends (on average) than you do. This can be explained as a type of sampling bias – people with greater number of friends are more likely to be observed in one’s own friends. Thus, we can infer that if you take a sample of random people, and then ask those people to each nominate a friend, the group of friend nodes would be (on average) more central and have a higher degree than the original, random group.
With this idea in mind Nicholas set out to study the diffusion of H1N1 at Harvard University. He and his research team took a random sample of 1300 students and asked them each to nominate a friend. They then studied the two groups over a period of several months, tracking the spread of the virus. What they found was that friend group began to show signs of an epidemic 16 days before the random group. Even more impressive, the point at which the instances present in the two groups began to noticeably diverge was a full 46 days before the general epidemic. Clearly, Nicholas and his team had discovered a successful, new methodology.
And the best part about his process is it can be applied to study anything that spreads via social interactions, not just bacteria. It could also be used to predict the spread of information, voting habits, smoking habits, vaccination use, product adoption, and much more. Applied to advertising, for example, it could help marketers understand how to more efficiently target audiences. This shows the great importance in understanding network structures to relatively cheaply and much more efficiently gather information on the spread of all types of epidemics.
Ted Talk: http://www.ted.com/talks/nicholas_christakis_the_hidden_influence_of_social_networks