215 — Limits of Predictability in Human Mobility

Song et al (10.1126/science.1177170)

Read on 23 March 2018
#mobility  #ecology  #population  #travel  #mobile-phone  #location  #prediction  #entropy  #social-dynamics  #society 

Do you know where you’ll be this time tomorrow? How about this time next week? Next month? Year?

It’s likely that even if you don’t know exactly where you’ll be, it can be predicted: Humans don’t vary their mobility that much, and when they do, it’s often in very predictable patterns (such as a vacation).

These researchers took 50,000 (anonymous) random mobile phone users and tracked their positions (using mobile-phone tower location, not GPS) over the course of three months.

From these data, the researchers assigned three measures of entropy to users’ mobility, which combined describe the probability of a user’s position in space and time.

The conclusion? There is 93% potential predictability of all of the users’ mobility. In other words, of all of the randomly sampled users’ jobs, lifestyles, travel patterns, and more, less than ten percent was considered systemic entropy (and was thus unpredictable).

Because these data were recorded prior to 2010, the resolution of the spatial component of the data are not great. If this study were performed again today, I imagine there would be much greater spatial resolution due to the amount of GPS data available. I’m curious to see if these conclusions still hold — I imagine they will.

In light of that… I’m curious to see what my mobility predictability is. And now I want an app that rewards you for confusing the predictor.