How to lose friends and predict epidemics

October 19th, 2010 by

With The Social Network out in cinemas, everyone is talking about Mark Zuckerberg and his 500 million friends. Seeing as we have recently discussed Mark, today I would rather talk about his nemesis, Nicholas Christakis.

While Mark is busy building friendships, Nicholas is destroying them: according to him, if your friend gets fat, the chance of you getting fat at the same time increases by 57%. Because of Nicholas, overweight people all over the world dine alone tonight (crowd goes awwww).

The social network makes us all stars
To be fair to Christakis, his point is not to make you ditch your fattening beer-buddies, but rather show how real social networks influence our health, norms and behavior. To be clear, we are not talking about virtual networks like Facebook, but those networks humans have formed for thousands of years, well before the Internet became our one and only master.

Counting our family, friends and work relationships, we are connected to probably hundreds of people. They in turn are connected to hundreds of people, some of whom we are connected to and many of whom we are not. We are all embedded into that vast fabric of humanity, mutually affecting each other. As a picture, the people in a social network look a bit like stars in the night sky, all connected to their neighbors with little lines.

Christakis and his colleague Fowler focus on how the structure of social networks can help us to predict the spread of epidemics. Their ideas apply to any type of social behavior spread by people, including consumer adoption of a product or diffusion of abstract ideas such as political views.

Everything you know is old
Today, the best way to predict epidemics is with labs (or analysis nodes), which report the incidence of certain conditions to a central database. One or two weeks later we find out at what stage the epidemic was the day the information was collected.

According to Christakis and Fowler there is a better way. Epidemics don’t spread randomly amongst a population. If we want to get an early warning of an epidemic or even forecast it, the key is to figure out how it travels through the structure of social networks.

In such networks, different people have different numbers of connections. One person might have three friends, while another might have 100. Curiously, however, two people with the same number of connections may not be equally important in the spread of an epidemic. Those individuals who are the stars at the center of the social galaxy are the key to early detection. Because of their centrality, when these individuals pick up a piece of information, a germ or a specific behavior, it will quickly spread through the rest of the population. If you want an early warning of an epidemic, it is much more effective to monitor these central people than it is to monitor random people in a population.

Your friends have more friends than you do
Mapping a network, however, is a hard task: it can be expensive, unethical, or technically unfeasible, not least because these networks are changing all the time. So how do you find out who are the central people in a network? Christakis and Fowler came up with a cunning insight: the friends of randomly chosen people have more connections, and are more central, than the random people themselves. As the saying goes, your friends have more friends than you do.

The two researchers tested this theory by observing the behavior and emails both from random subjects and their friends. They found that monitoring the friends allowed them to forecast an epidemic 60 days before it hit. The length of this advance warning depends on factors such as the nature of what is spreading and the structure of the network, but the main point remains: we can predict future events with amazing accuracy by simply understanding the ties between people.

Apart from the fact that it is fascinating, the reason I mention all these ideas is because of how they can be applied to the concept of distributed work.

An obvious example is how a crowdsourced group of related volunteers could help us to understand trends faster. More interesting is how these ideas might improve the efficiency of distributed work. Would adding a social layer to existing crowdsourcing services (where people are generally unrelated) help forecast the diffusion of good practices, data and other information throughout a network of workers?

If you have any thoughts on this subject, please spread them. We would love to become infected.