Read on Twitter
New paper out!
https://www.nature.com/articles/s41586-020-2909-1
We present a simple model that captures the scales of human mobility, and explain why it was previously thought that human mobility was scale-free.
Tiny summary
.
https://www.nature.com/articles/s41586-020-2909-1
We present a simple model that captures the scales of human mobility, and explain why it was previously thought that human mobility was scale-free.
Tiny summary
.
It's not terrible complicated. Our model computes the probability of going from place A to place B given some hierarchy of containers (think countries, regions, cities, neighborhoods).
Then, we take a long trajectory of mobility (like your Google Maps GPS data for a year) and find the hierarchical partition of space that makes all the trips in your data as likely as possible, as estimated by this equation. Standard MLE to use a fancy term.
This produces two (pretty sweet) things.
(1) It gives you a hierarchical partition of space unique to YOU (person whose data is analyzed). It finds buildings, neighborhoods, cities, etc.
Try it with your own data, @lau_retti has our code up: https://github.com/lalessan/scales_human_mobility
(1) It gives you a hierarchical partition of space unique to YOU (person whose data is analyzed). It finds buildings, neighborhoods, cities, etc.
Try it with your own data, @lau_retti has our code up: https://github.com/lalessan/scales_human_mobility
(2) It assigns a scale to each trip in your data. Was this trip between buildings within a neighborhood or was it between cities in a region? Of course these scales are simply represented by indices, but enriching movement data like this is useful for a myriad of applications!
A final (extra sweet) fruit of this model is that once it has been fitted, it can be flipped on its head to create new synthetic mobility traces. It enables a new and much improved way of simulating human mobility.
You can explore how generating synthetic traces works in this interactive notebook: https://observablehq.com/@ulfaslak/a-model-for-generating-multiscale-mobility-traces
The scale-free result of earlier work emerges because of aggregation. Here's an observable notebook that visually explores how power-laws emerge when one aggregates over scales: https://observablehq.com/@ulfaslak/a-visual-exploration-of-how-a-power-law-can-emerge-from-aggre
Thanks to my amazing co-authors @lau_retti and @suneman. This research was the most fun and meaningful thing I ever worked on in academia!
