By John Bohannon
ScienceNOW Daily News
17 August 2009
Your telephone may know more about your private life than you do, according to a new study of mobile phone calls. The insight opens the door to mining massive data sets from mobile phone call logs, which should allow researchers to test theories for how relationship networks make or break businesses, shape the flow of information, and even affect the course of epidemics.
A nagging problem for social scientists is the limitation of self-reported survey data. Not only are people expensive to poll, but they are also notoriously error-prone when they try to recall their own behaviors. What researchers would prefer is a record of people’s behaviors that is cheap and accurate. Mobile phone call logs can certainly provide enormous amounts of cheap data. Researchers have used such data to map out people’s social networks, utilizing the duration and frequency of calls between pairs of people as a measure of the intimacy of their relationships. Doing so has revealed patterns of people’s contact with each other both in time and space, which is crucial for modeling everything from gossip to how flu viruses spread across populations.
But how accurately do call patterns reflect the intimacy of relationships? After all, sometimes the closest of friends rarely call each other, while some motor mouths call just about everyone.
To put telephone data to the test, a team led by Nathan Eagle, an engineer at the Massachusetts Institute of Technology in Cambridge, gave mobile phones to 94 MIT students and faculty members. For 9 months, software on the phones kept track of the volunteers’ location and logged all calls made between these phones. Over the same period, the researchers also gathered social data from the subjects in the traditional way, asking them whether the other subjects were friends, acquaintances, or strangers. Finally, the subjects rated their job satisfaction, which has been shown to strongly correlate with the number of workplace friendships.
Just by analyzing the calling patterns, the researchers could accurately label two people as friends or nonfriends more than 95% of the time. But the results, published online today in the Proceedings of the National Academy of Sciences, show that the mobile phone data were better at predicting friendship than the subjects themselves. Thirty-two pairs of subjects switched from calling each other acquaintances to friends in the traditionally gathered survey data. These are most likely new relationships that formed during the course of the study, say the researchers, and they left a clear signal in the mobile phone data. Friends call each other far more often than acquaintances do when they are off-campus and during weekends. The pattern is so distinct that the researchers spotted budding friendships in the phone data months before the people themselves called themselves friends.
Finally, the team compared people’s self-reported job satisfaction with their networks of friendship at their workplaces. Because the mobile phones kept track of people’s proximity to each other, the researchers had a clear measure of people’s daily contact with friends at work, not only through calls but through physical proximity. As predicted, the more contact people had with friends at their workplace, the more highly they rated their job satisfaction. And conversely, the less face-to-face contact people had with friends at work, the less they said they enjoyed it.
The finding that you don’t have to ask people about their relationships–that just looking at the pattern of their phone calls is sufficient–“is very new,” says Brian Uzzi, a social-network scientist at Northwestern University in Evanston, Illinois. The next question is whether the new methods for maintaining contact with friends–such as e-mail and social Web sites–are weakening the need for physical proximity to friends. “It’s a face-off between Facebook and face-to-face contact,” he says, and “it looks like face-to-face contact still matters a lot.”
- SOCIAL SCIENCE: Tracking People’s Electronic Footprints
- John Bohannon (10 November 2006)
Science 314 (5801), 914. [DOI: 10.1126/science.314.5801.914]