The Connected States of America

Press release

The Connected States of America -- Using mobile communications to redraw community boundaries.

CAMBRIDGE, Mass. -- Researchers at MIT Senseable City Lab, AT&T Labs-Research and IBM Research are revealing new research that redefines regional boundaries in the United States, using patterns of social connectedness across the country derived from anonymous and aggregated cell phone data.

In some cases, connectedness follows traditional demarcations such as state lines -- but in other cases, new patterns are emerging that have little to do with political or administrative boundaries. By looking at billions of instances of aggregated^[*] mobile communication, researchers are able to define communities through the more informal lens of social networks.

"This work proposes a novel, fine-grained approach to understanding cities and human communities in space," says Carlo Ratti, director of the MIT Senseable City Lab.

Cities play an important role in defining community boundaries, as they tend to pull nearby counties into their radius of influence. This radius of influence depends on factors such as size, population density and geography of the city’s surroundings. The researchers explain cities' radius of influence in terms of laws similar to Newton's laws of gravitation: Larger places attract more people and businesses than smaller ones, and the attraction between closer places is greater than that between remote ones. As a first approximation, the likelihood of two people communicating with one another depends on the respective populations of the origin and destination of the call, and drops off according to the distance between them.

The "Connected States of America" provides a more natural delineation of regions that follows relationships between family, friends and business partners. However, "telecom and state partitionings of the US results are very similar, as 90% of counties in the official state partitioning fall within a corresponding (by largest overlap) telecom community" - comments Francesco Calabrese, advisory research staff member at IBM Research-Ireland. "Sister states" emerge, such as Georgia and Alabama, Mississippi and Louisiana, and Tennessee and Kentucky, among others. Metropolitan areas often form pockets of influence that extend into neighboring states or communities; for example, Chattanooga, Tenn., is more closely linked to communities in Georgia and Alabama than to the rest of Tennessee. Pittsburgh, Penn., and West Virginia form a new "state," while St. Louis, Mo., exhibits an expanded reach that splits Illinois into two regions. New Jersey and California also divide into two distinct regions due to large cities. In contrast, Texas remains whole: Despite the potentially splitting influence of cities such as Dallas, Houston, San Antonio and Austin, the researchers found that there is enough inter-city communication to hold the state together.

However, a simple gravitational model does not explain all of the results. For example, distance between places can be measured in several ways: as the crow flies, along transportation routes or by travel time. Mountain ranges and other geographic features influence how people interact, because they contribute to an increased perception of distance and therefore hinder communication. "This phenomenon may explain, for example, why Chattanooga appears cut off from the rest of Tennessee and better connected to parts of Georgia and Alabama," says Dominik Dahlem, a postdoc at the Senseable City Lab.

Interestingly, analyzing boundaries according to aggregate and anonymous records of text messages (SMS) instead of phone calls yields a different map of connectedness. Some sister-state pairings change — for example, instead of Georgia-Alabama and Louisiana-Mississippi, SMS data link Mississippi and Alabama, leaving Louisiana and Georgia as stand-alone states. Oklahoma and Arkansas break apart, while West Virginia and Ohio join together. California splits into three communities instead of two. According to the researchers, these differences can be explained by the fact that SMS is generally favored by a younger population and is less likely to reflect cross-generational communication. Also, the SMS map is more divided overall, indicating that people are less likely to send text messages over large distances than they are to make phone calls.

This data reveals patterns of social and economic activity that the researchers expect will be of interest to social scientists and policymakers. "We are particularly interested in how such rich information can help us gain a better understanding of our society, which in the future, could lead to more democratic, bottom-up structures of governance," Ratti says.

"This example illustrates once again the insights that can be inferred from aggregated communication patterns, as wells as how collaboration across fields of research can benefit for society," said Alexandre Gerber, a researcher at AT&T Labs.

Analogous results for Great Britain were recently published by the same team in the journal PLoS ONE, which analyzed 12 billion anonymized records representing more than 95 percent of Great Britain's residential and business landlines. In that study, communities that emerge out of people's communication habits were found to be more cohesive than administrative boundaries.

The research was done in partnership with AT&T Labs-Research, IBM Research and the National Building Museum in Washington. Support was generously granted by the Rockefeller Foundation, the National Science Foundation, the AT&T Foundation, the MIT SMART program, GE, Audi Volkswagen, SNCF and the members of the MIT Senseable City Lab Consortium. For visualizations and more background information, please visit http://senseable.mit.edu/csa.

^[*] All communication data was aggregated by county, determined by the caller's and recipient’s most frequently used cell tower, which was assumed to be near their residence. In order to determine which counties are connected most closely by communications, researchers analyzed anonymized location data for both ends of the calls and texts. No personal information was used.