Nicholas Christakis and James Fowler have released a new paper that looks at the potential predictive power of social networks. They claim that current methods of contagion detection are, at best, contemporaneous with the actual epidemic. What is needed is a true early detection method, one that would actually provide an accurate prediction of a coming epidemic.
Christakis and Fowler claim that social networks can be used as sensors for various types of contagions (whether biological, psychological, informational, etc). In an inventive twist, they leverage what is known as the Friendship Paradox–the idea that, for almost everyone, a person’s friends tend to have more friends than they do. Contagions tend to appear sooner in those individuals that are closer to the center of a social network. The logic goes that if you ask a group of people to name one of their friends, those friends will be closer to the center of the network than the people you asked. Rather than map and monitor an entire social network, simply monitoring these friends should allow researchers to detect the outbreak of, say, H1N1 much earlier.
They tested their theory using Harvard College undergrads, attempting to detect the outbreak of the flu. (You can watch Christakis discuss the paper and research during a recent TED talk in the video embed below). What did they find?
Based on clinical diagnoses, the progression of the epidemic in the friend group occurred 13.9 days (95% C.I. 9.9–16.6) in advance of the randomly chosen group (i.e., the population as a whole). The friend group also showed a significant lead time (p,0.05) on day 16 of the epidemic, a full 46
days before the peak in daily incidence in the population as a whole. This sensor method could provide significant additional time to react to epidemics in small or large populations under surveillance. The amount of lead time will depend on features of the outbreak and the network at hand. The method could in principle be generalized to other biological, psychological, informational, or behavioral contagions that spread in networks.
That is a pretty impressive result. By simply tracking those individuals located closer to the center of the network, Christakis and Fowler were about to detect the progression of the flu a full 2 weeks before the general population. They were also able to derive an early warning signal over a month before the peak of the outbreak in the general population.
If this result can be replicated and validated there are various ways it can be utilized.
Here are a few off the top of my head:
- Product Launches: Particularly in the tech industry–where so often we now see product launches as proto-typing–we could use this method to very quickly gauge the awareness and adoption of a new product and predict the extent to which it will spread throughout the general population. Companies would have better early-warning systems, which would allow for killing dud products or boosting marketing for those products that are poised to explode. I would assume this would be particularly applicable to products that benefit/rely on network effects.
- Political Indicators: One can think of political unrest as a contagion–discontent starting earlier with a core group within a social network and then, over time, spreading to those on the outskirts of the network. Tracking the population as a whole may not give you an early warning of unrest, but rather a snapshot of a problem at a time when it is too late to do much about it. Focusing on those closer to the core of a social network could provide enough lead time to diffuse tensions or intervene in other ways to avoid a full-scale upheaval. Moreover, businesses and investors could also use the early warning as a signal to make adjustments in supply chain and their portfolios to take into account the potential unrest. Finally, citizens within those countries could benefit by having more lead time to evacuate conflict zones, etc.
- Economic Indicators: Investors, businesses, and politicians are always looking for better economic indicators–those signals that are leading indicators of larger economic trends. I wonder if adjusting the sampling frames of various polls to incorporate the Friendship Paradox might give us an even earlier warning for mortgage defaults, consumer confidence and spending, manufacturing activity, etc. Not as sure about this one, but certainly much of economic activity takes place in a networked structure.
Would love to hear other thoughts.