Simplifying models of human mobility at the daily scale
When modeling human mobility at the daily scale, researchers generate “trip chains” by including a stochastic component to allow for all possible choices about travel mode, route, sequence and trip purpose an individual can make in a given city. These models rely on detailed data about the daily routines of individuals as reported in travel surveys, information that is also hidden in data that is passively collected and stored as digital traces left by mobile phones. But recent studies have shown that despite some degree of change and spontaneity, people’s daily mobility is actually characterized by patterns of deep-rooted regularity. If these underlying patterns of mobility could be extracted from the passive digital data and if identical patterns are found in the travel surveys, these patterns could greatly simplify the process of generating trip chains in the activity-based models. It could also allow researchers to benefit by using the alternative and lower-cost digital data sources.
Professor Marta González and postdoctoral associates Christian Schneider and Vitaly Belik used Paris cellphone data for 500,000 people over 154 days. With these data, they identified 40,000 people whose call frequency was great enough to allow the researchers to track the trajectories of those individuals’ daily trips via the location of the cellphone towers that handled each call or text message. The researchers assumed eight hours of inactive time at night and 16 hours of active travel time. For employed individuals (data obtained from a Paris travel survey), the 16 hours included a fixed activity — work — and flexible activities, which are trips to other destinations with a stay of at least 30 minutes. The repetition of destinations and trajectories over the 154 days revealed daily routines for each person that could be seen as daily networks, with nodes representing destinations and directional lines representing trips. The researchers then discarded information about trip distance, duration and a destination’s latitude and longitude (from which destination type, akin to the purpose of the trip used in activity-based models, can be inferred) and denoted the home node as starting point. They now had simple, unweighted sequential diagrams or trip configurations. For comparison, they made daily trip configurations using travel survey data from Paris and Chicago.
The researchers found that trip configurations created from cellphone data were consistent with those created from travel surveys, indicating that passive cellphone data could be substituted for more expensive travel surveys. Also, the distribution of trip configurations was similar in both cities. And an important pattern emerged: with the addition of each flexible activity, the number of possible trip sequence configurations increased exponentially, but the number of configurations used did not increase much, if at all. In both cities, 90 percent of the employed population made daily trips from home to five or fewer locations using only 17 of the more than 1 million possible trip configurations. Using these findings, the researchers created a model of daily intracity travel from which another pattern emerged: once people engaged in a single flexible activity, they were 10 times more likely to engage in an additional flexible activity on the same trip than were people who went only to work.
The prevalence of the 17 trip configurations indicates that they represent “motifs,” which in network theory are patterns that occur with such frequency that the statistical probability of their random occurrence is negligible. The presence of motifs indicates that the study has uncovered a basic principle that can be used in predictive models for daily intracity travel, allowing trip chains to be based on a relatively simple mathematical formula and considerably reducing the complexity of algorithms that simulate behavior. Such a model could work in any city that has cellphone towers located within 1 kilometer of each other and enough cellphone activity to track the population in 30-minute intervals across the 16 active hours. The next step will be to incorporate the nature of flexible activities using digital maps and a destination’s coordinates to determine the type of facility at that location.
A paper on this research written by Marta González; first author Christian Schneider; Vitaly Belik, now at Max Planck Institute for Dynamics and Self-Organization in Göttingen; and Thomas Couronné and Zbigniew Smoreda of France Telecom appeared online May 8 in the Journal of the Royal Society Interface.
Download this issue as a pdf.