Mobile phone communications have provided fertile territory for research into the spatial dimensions of communities. Studies of calling patterns have shed new light on the complex nature of networks. The analysis of billions of calls across a number of countries has led to a surprising conclusion: telephone exchanges are still largely dictated according to administrative boundaries laid down long before the arrival of the mobile handset.
What are the most significant distinctions between mobile phones and their land-based forebears? They are truly personal, as the number corresponds to an individual rather than a household, and costs are no longer determined according to the distance that separates callers within the same country. Most obviously, they are portable and always available no matter the location of the user. It would be safe to assume that evolving technology necessarily leads to changing habits and with the market penetration that has occurred over the last 15 years in the cellular industry surely communications have changed as we are no longer constrained by the limits of geography or sharing the same handset between family members.
Yet assumptions can be dangerous things and the reality is that the more things have changed the more they have remained the same. If we take a step back and analyze the spatial distribution of communications networks it becomes clear that, whether land-based or mobile, they are still for the most part local. A 2008 study demonstrated that the probability of random phone calls declined rapidly in direct relation to the distance between households. Put simply, the greater the distance the less likely we are to call. The phenomenon is hardly surprising and has been backed by rigorous analysis of the relevant data. What is somewhat less understandable is why research clearly demonstrates that most users seem to restrict their calls to residents within the same administrative region and this remains true even for those living on, or close to, frontiers. This curious fact has been confirmed with a high degree of precision across the following three industrialized nations who nevertheless each present their own unique set of circumstances.
The Belgian experience
To extract the community structure of large networks a research team at the Université catholique de Louvain drew from data collected from telephone exchanges between Belgium’s 2.6 million mobile phone subscribers, measuring the frequency of calls as well as their duration in order to a clearer picture of how callers from various municipalities create their own communities. The scale of the project was made possible by a “modularity maximization” algorithm sometimes called the “Louvain Method”. Modularity occurs when groups are sub-divided to enable a comparison of the increased density of links between the members of a given group with that obtained in a random group with the same overall characteristics. The formation of large communities from smaller networks corresponds to the notion of social identity but we should underline that geography played absolutely no role in the pre-defined parameters of the Belgian study and findings were based solely on actual calls made.
Nevertheless, the results were astonishing as there was clear evidence of a clustering effect and groups were made up of municipalities that existed in close proximity to each other. In fact, they are in all cases adjacent and there is not a single exception to this rule.
The groups are also spatially balanced into 17 “telephone areas” each consisting of 15 to 66 municipalities.
Last but not least: there exist frontiers that bear clear traces of the evolution of Belgium and its own unique administrative and geopolitical history.
Figure 1 – Belgian ‘Telephone Areas’.
Each color corresponds to a defined area: (1) regional city (2) major city and (3) provincial borders.
(Source: V. Blondel, G. Krings, I. Thomas, “Regions and borders of mobile telephony in Belgium and in the Brussels metropolitan zone”, 2010)
Across the Channel: The British Experience
A similar study was conducted in Britain by MIT’s SENSEable City Laboratory with one important difference in that the data set was drawn solely from British Telecom’s business and residential landlines.
In Belgium, researchers used the zip code of the mobile subscribers billing address to obtain geographical information whereas the MIT team used the address of the actual landline but placed anonymity safeguards in place to avoid pinpointing the exact address using municipalities as a kind of mask to protect individual identities.
Figure 2 – British ‘telephone areas’.
Working from this somewhat different base (Source: C. Ratti et al., “Redrawing the Map of Great Britain from a Network of Human Interactions”, 2010) researchers reached remarkably similar conclusions to their Belgian counterparts. The results illustrated in Figure 2 as well as video animations demonstrate a clear relationship between telephone interactions and traditional frontiers.
Ingrained habits: a French exception?
Regional linguistic variations offer one logical explanation for the divisions observed in Belgium and to a somewhat lesser degree in Britain but in France no such claim can be made. It would be reasonable to expect therefore some differences to the previous examples but the actual results fly in the face of this assumption. Indeed, if anything, France demonstrates an even starker division along administrative lines and we invite our readers to examine the details of our study to better understand how we have arrived at this startling result.
Research was conducted using a data set based on cell phone usage over the Orange France network and we ensured a focus on interpersonal transaction by excluding the activities of enterprise customers. Links were detected with the same algorithm as the Belgian study and we had at our disposal several billion actual calls made by approximately 17 million subscribers. The geographical location of users was determined by using the municipality (zip code) of the customer’s billing address which meant that even if a call was made from Paris by a user from Toulouse the data would still be systematically aggregated with that of Toulouse.
Figure 3 – French ‘Telephone Areas’. The video animation gives a dynamic representation of the map that resulted from the study.
Figure 3 demonstrates the structure of telephone communications organized according to the parameters of the study, and while lacking some of the finer details, data produced through application of the modularity maximization algorithm gives definitive proof of the tendency for communities to organize themselves spatially according to pre-defined administrative frontiers. Each community is represented by colors onto which the borders of France’s 22 régions have been overlaid for comparison.
The image speaks louder than words yet we can nevertheless illuminate our discussion through some additional commentary.
To begin, as was the case in the Belgian and British studies, the communities uncovered adhered to distinct spatial logic although no geographical presupposition had been made when data were being collected. Communities have aligned themselves along rather strict borders and their populations are generally uniform (as opposed to being spread across numerous disjointed pockets). Moreover, aside from a small number of outliers, communities are divided according to previously defined administrative boundaries with a surprising level of precision.
It should also be noted that while most regional boundaries contain a single homogenous “community”, others contain multiple groups but this should come as little surprise as French régions are a fairly recent invention and as figure 4 clearly demonstrates local populations are more deeply attached to the more ancient notion of their départment thus adding further weight to the argument that communication patterns tend to follow well-established administrative boundaries.
Figure 4 – In the French région of Languedoc-Roussillon there is a clear frontier between the départements of l’Hérault and l’Aude.
In light of these somewhat surprising results some might be led to question their legitimacy as well as raise doubts over the methodology that was used giving rise to the following questions.
What advantages does the “Louvain” method have over other similar approaches?
Dividing a network in a way that maximizes modularity presents numerous difficulties and the method offers an optimized analysis of the flow of telephone traffic and demonstrates that the number of calls within communities is far superior to those between them: respectively, on average, 85% compared to 15%. The nodes of the network are more densely connected internally than externally and the difference is quantifiable to the point where the differential remains surprisingly constant even for those on the edges of a defined boundary. In the majority of border zones the proportion, although slightly diminished attains 80%/20%, and in a few exceptional cases is reduced to 75%/25% which is still significant.
What patterns are displayed concerning call duration?
As one would expect, the nature of communication changes in relation to the distance that separates callers. The characteristics of short local exchanges can be systematically placed under the broad umbrella of conversations such as: “Where are you now?”; “When are you arriving home?”; or, “I’m running late.”, etc. They act as a complement to frequent face-to-face contact. On the other side of the coin are calls made over longer distances which while less frequent are also of longer duration. Needless to say this creates a danger that we have biased our results by placing too much emphasis on local communication.
Yet we must also recognize that the two types of call (short vs. long distance) both serve the same function as a foundation for defining social relationships. Over greater distances the telephone fills the gap left by an individual who is no longer physically present whereas in local communication calls play an important role in the more banal aspects of daily life by providing a constant link between numerous interrelated actors. Concentrating on the frequency of conversations as opposed to the distance for example provides a far more accurate illustration of the ways in which interpersonal relations are constructed and interact internally. As an indicator they help identify “community activities” and it is on this point that our real interest lies. In any case, the distance variable plays a negligible role in our discovery that communities have been constructed largely along artificially created administrative boundaries as our map clearly illustrates in relation to border dwellers and their social behavior.
So if we resolve that there is no real bias to the methodology could we expand on why your findings are surprising?
Following the French Revolution, the former division of the national territory into provinces was replaced by the creation of départements in 1790, most of whom nevertheless retained a strong local identity. Régions are a more recent creation (1919) and in most cases play no more than a symbolic role, and are for the most part administrative bodies. The results come as a surprise because aside from some notable exceptions, Corsica in particular, there has never been a sentimental attachment between local populations and the artificially created régions.
What conclusions can be drawn from this turn of events?
Analysis of telecommunications flows (cellular networks and landlines) reveals that 80% of all calls cover distances of no more than 50km. It takes no great stretch of the imagination to understand why as social networks tend to be constructed around geography and the main points of interaction such as schools, the workplace, neighborhoods, etc. School districts are determined by administrative borders and form the foundation for interpersonal relations, not only among students but their parents as well due to the fact that administrative boundaries determine where children will receive their education (crèches, schools, etc.) Further reinforcement of the trend is provided by the fact that in France, grown children are unlikely to stray too far from the proverbial nest upon reaching adulthood. Additionally, employment patterns play a significant role in the delicate daily balancing act between professional and family obligations and tend to be determined more by administrative boundaries than proximity. The conclusion is that interpersonal relationships are driven as much by administrative boundaries as geographical proximity.
Administration in France tends to operate according to a highly centralized bureaucracy and as départmental boundaries were being drawn each seat of local government (known as the préfecture or chef-lieu) was generally located at the département’s geographical core, in a city or town of some importance, within one day’s horse ride distance of any other part of the départment. As towns and cities are where populations tend to congregate the logical conclusion is that these are also the spaces with the highest concentration of interpersonal communications. The tendency for urban areas to suck in their surrounding neighbors only adds to the phenomenon and the dynamic interaction between patterns of habitation, education, and employment bears the indelible imprint of administrative boundaries which act as a thread running through all aspects of daily life.
For the first time, by studying the structure of telecommunications we have the ability to measure with a startling degree of accuracy levels of regional coherence with interpersonal ties.
Through the combined forces outlined above the régions, which have until now been viewed as no more than a collection of culturally distinct départements would seem to be evolving towards a more deeply rooted sense of shared identity. This is, in any case what the studies have suggested.
Regional identity: a new social reality?
While society trumpets the virtues of ever increasing mobility our results demonstrate that social relationships remain anchored to a well-defined geography. Despite technological advances daily interactions continue to revolve primarily around local concerns. The sepia-toned image of the travelling salesman alone in his hotel, hit by the sudden urge to phone home, was thought to have vanished along with other remnants of a bygone age. In the global village does his modern homologue suffer the same longing for hearth and home? The answer is clearly in the affirmative and it would seem that the era of the globetrotting nomad has yet to arrive and remains far from the norm.
Moreover, results demonstrate that in France, somewhat more than in neighboring Belgium or Britain, the technocrat’s dream of rational boundaries based purely on efficient administration may be close to becoming a reality. Could it be that the grey-suits are cracking a rare smile from the grave?
- “Fast Unfolding of Communities in Large Networks,” (V. Blondel, J.-L. Guillaume, R. Lambiotte, E. Lefebvre, Journal of Statistical Mechanics: Theory and Experiment, 2008)
- “Regions and Borders of Mobile Telephony in Belgium and in the Brussels Metropolitan Zone” (V. Blondel, G. Krings, I. Thomas, Brussel Studies, 42, 2010)
- “Geographical Dispersal of Mobile Communication Networks (R. Lambiotte, V. Blondel, C. de Kerchove, E. Huens, C. Prieur, Z. Smoreda, P. Van Dooren, Physica A, 387 (21), 2008)
- Geographic Constraints on Social Network Groups (J.-P. Onnela, S. Arbesman, M.C. González, A.L. Barabási, N.A. Christakis, PLoS ONE, 6 (4), 2011)
- Redrawing the Map of Great-Britain from a Network of Human Interactions (C. Ratti, S. Sobolevsky, F. Calabrese, C. Andris, J. Reades, M. Martino, R. Claxton, S.H. Strogatz, PLoS ONE, 5 (12), 2010)
More on paristech review
By the author
- Voice on the border: do cellphones redraw the maps?on November 15th, 2011