Phone data predicts unemployment rates months earlier than official statistics

US data researchers have shown how mobile-phone data can be used to “quickly and accurately detect, track and predict changes in the economy at multiple levels”. Their most exciting finding related to predicting job losses. The team used algorithms to examine call-record data from two undisclosed European countries. With this information they could measure unemployment rates up to four months before the release of official reports.

The research was carried out by a team of experts in engineering, public policy, economics, and information science from MIT, Harvard University, the University of Pittsburgh and the University of California. The team's findings were published in the Royal Society Interface Journal.

In this instance, the team had 15 months worth of call-record data from 2006-2007. They created a “structural break model” to figure out which mobile-phone users had been made redundant in that time period.

From that they could follow and measure data relating to those affected, looking at everything from total calls, number of incoming and outgoing calls, as well as calls made to individuals physically located at a place of work.

The data showed the total number of calls made by individuals who had recently lost their jobs dropped by 51 per cent following their layoffs when compared with those still gainfully employed. Their number of outgoing calls decreased by 54 per cent.

Clever approach

“It’s a clever approach,” says Dr Brian MacNamee from the UCD School of Computer Science and Informatics. “In one case, they used data from a particular town, a reasonable sized town with a big factory, where lots of local people were employed. Then that factory closed down. So what they knew was that lots of people had lost their jobs. From that dataset they tried to learn what the behavioural difference was in the way people used their phones before and after they became unemployed.”

They were also able to use GPS to analyse some people’s movements. “You’re likely to be less mobile if you have just lost your job,” says MacNamee. From this and the other phone data collected they generated an unemployment index, which was pretty accurate compared to the official unemployment index. More importantly, they could do so much faster than official reports.

According to the authors, the results show that “a user’s social interactions see significant decline and that their networks become less stable following job loss. This loss of social connections may amplify the negative consequence associated with job loss observed in other studies”.

Big-data zeitgeist

Welcome to the current big-data zeitgeist: how to take a signal from something that people do in their everyday digital lives and extrapolate from it.

“This is very interesting research which falls into an ever-growing category that looks at what information we may extract from our digital footprint,” says MacNamee. “Students of mine did a similar study with M50 traffic volume data. They tracked the unemployment rate in Dublin based on fluctuating levels of traffic on the motorway. There was a pretty strong correlation.”

The question is, can we do anything useful with this kind of information? Not everyone is convinced.

"I'd be fairly sceptical as to some of its conclusions," says Kenny Doyle of the Telecommunications Software and Systems Group at Waterford Institute of Technology.

“They used mobile-phone data to show that people’s social networks of interaction contract after they lose a job. There’s already a long history of social research dating from the early 1930s to the present day saying exactly the same thing without needing expensive big-data analysis.”

But they were able to predict unemployment rates months in advance of official reports. “The official unemployment indicators are important but they are notoriously slow,” says MacNamee. “I’m not convinced government or anyone else could do anything different though, even if they had three or four extra months notice of a spike in unemployment. It’s not long enough to change the outcome.”

Dangers

Big data analysis has its dangers. “There are undoubtedly a lot of areas where big data can be useful,” says Doyle. “Doing mass profiling like this, however, can have negative consequences. Even if an algorithm is 99 per cent accurate, that one per cent missing can equate to large amounts of people.”

The authors of the study were under no illusions as to the weaknesses in big-data analysis as a sole predictor of significant social change.

David Lazer, Harvard director of the programme for networked governance and associate professor of public policy, co-authored the study. He warns against using the team's approach as a substitute for traditional survey-based methods for unemployment or anything else.

“We consider mobile phone data a powerful yet complementary tool,” says Lazer. “Big-data approaches are fast and inexpensive, but the norms governing phone use are constantly changing, forcing us to constantly calibrate how we use them in connection with other methodologies.”