By analyzing anonymized information about billions of telecommunications records, the researchers assembled a map depicting the strength of the connections between different parts of Great Britain, based on the amount of information they exchanged. They then developed an algorithm that partitioned the map into ever-smaller regions, with the requirement that the connections within regions be stronger than those between regions. They found that, to a large extent, their partitioning agreed with existing administrative, geographical and historical boundaries. But there were a few notable exceptions: Some parts of Wales, for instance, had much stronger connections to cities in western England than they did to the rest of Wales, suggesting that in some ways, the historical distinction between England and Wales may be obsolete.
“The problem of defining regions in space is an old and important one,” says Carlo Ratti, director of MIT’s Senseable City Lab and lead author on the new paper. “It allows us to understand the interplay between geographical and social institutions that we’re building, and it helps us better capture the way our cities or our countries work. It could also have a lot of practical consequences: If you can identify regions, you can build better governance.”
Ties that bind
In network science, partitioning algorithms are generally indifferent to geography. For many telecom applications, it would be perfectly acceptable to lump, say, New York City and Los Angeles together, because of the amount of data they exchange, and partition them from Oklahoma City. Since that kind of clustering obviously wouldn’t lead to practical administrative regions, the researchers — including a half-dozen members of the Senseable City Lab, Cornell mathematician Steven Strogatz, and Jon Reades of University College London — planned to modify their algorithm so that its partitions would be geographically contiguous. The researchers found, however, that this additional restriction was unnecessary: At every level of resolution, the highest volumes of data exchange were almost always between adjacent regions.
It’s not surprising that the result of the partitioning algorithm largely mirrored the existing political boundaries in Great Britain, Ratti says. After all, if communities have been lumped together culturally and politically for centuries, they have good reason to exchange information with each other. “The interesting thing is when they don’t correspond,” Ratti says. “Then you can go and look and find out why not.”
“They produced some interesting and actually quite plausible divisions, which are similar but a little different from those that people have agreed upon in the past,” says Michael Batty, Bartlett Professor of Planning at University College London, who wasn’t involved in the research. “They don’t impose the criterion for them to be physically adjacent, but the results they get do show that the groupings they get are physically adjacent. Which is extremely encouraging, really, because it gives a lot of strength to the argument that these are unique and very well-defined partitions.”
It’s unlikely, of course, that mid-Wales and the English midlands will be fused into a single administrative region, so the discovery that several Welsh cities seem to depend heavily on the exchange of information with nearby English cities, and vice versa, probably won’t have major political consequences. But, Ratti says, the value of the new work is that it shows that analyzing information flow could be a useful tool in the drawing of political boundaries. And, he says, it will only become more useful as data from other networks — the Internet, Internet telephone networks, instant-messaging networks — become available for analysis.
Moreover, he says, the researchers are now applying their algorithm to telephone data from countries in which the re-drawing of the political map is a very real possibility. In such cases, Ratti says, analysis of communication patterns could help define boundaries whose imposition would minimize the disruption to the lives of ordinary people.
The work was conducted with support from the National Science Foundation, the AT&T Foundation, the MIT SMART program, GE, Audi Volkswagen, SNCF, ENEL and the members of the MIT Senseable City Lab Consortium.