The United States Department of Agriculture Economic Research Service (USDA ERS) builds commuting zones across administrative boundaries in the United States to define local labor markets and economies, as well as to understand areas in which people tend to spend much of their time and interact. However, a consistent version of commuting zones does not exist for the entire globe, and the USDA approach (more recently generated and maintained by Penn State) is limited to using 10-year census data to construct new zones, which means they can be out of date.
One standard version of commuting zones does not exist for the whole world. Facebook is able to build a version of commuting zones for the world based on aggregations from inputs like cities that people put on their facebook profile and location information people choose to share with Facebook. We can build our commuting zones consistently across all regions where we have sufficient numbers of users to protect privacy. We are able to update our commuting zones estimates more often than census-based methods to see how they evolve over time or are altered rapidly by a crisis.
Why are Commuting Zones useful?
Commuting zones represent areas where we spend the majority of our time. These areas of economic integration are independent from political boundaries and can illustrate how economic communities and commute patterns transcend regional boundaries. Furthermore, although the definition of regional boundaries can also vary substantially across countries, we create the commuting zones in a consistent way across the world. USDA ERS commuting zones have been used in a number of economic studies. By providing a representation of areas that are closely connected across the world, our commuting zones may be helpful for these studies as well as epidemiological research aimed at understanding where diseases may be transmitted.
Our commuting zones are rebuilt every three months using aggregated estimates of home and work locations. Only users who opt-in to sharing their location data with Facebook are included. For each continent, we build a graph that connects population centers within the continent and cluster these locations.
In these graphs, population centers represent the nodes. The edges (connections) between these nodes have a weight. We construct this weight using weekday home and work locations over the previous few weeks using the following formula:
With this formula, the weights that connect nodes always range from 0 to 1. When we construct this graph, there can be connections between very distant places on opposite ends of a continent. Since our current algorithm does not formally take distance into account, including very distant locations adds noise to our graph. To mitigate this noise, we remove connections that are in the 95th percentile or above in the distribution of distances.
To build commuting zones:
- We extract community structure from the graphs constructed above using the Louvain algorithm.
- We impose a hierarchical structure by taking the resulting clusters and partitioning them once more with the Louvain algorithm. For each location, we build Voronoi shapes around their centroids.
- We then create overall shapes on a map by merging all spatial areas associated with the same commuting zone.
This is an active area of research and we are still developing and improving on our methodology to generate these zones.
This map illustrates commuting zones for the United States.