Skip to content
Apply
Stories

Researchers use mobile phone data to predict employment shocks

People in this story

North­eastern Uni­ver­sity com­pu­ta­tional social sci­en­tist David Lazer and his inter­dis­ci­pli­nary research team have demon­strated that mobile phone data can be used to quickly and accu­rately detect, track, and pre­dict changes in the economy at mul­tiple levels.

North­eastern Uni­ver­sity com­pu­ta­tional social sci­en­tist David Lazer and his inter­dis­ci­pli­nary research team have demon­strated that mobile phone data can be used to quickly and accu­rately detect, track, and pre­dict changes in the economy at mul­tiple levels.

The find­ings, pub­lished Wednesday in the Journal of the Royal Society Inter­face, high­light the poten­tial of mobile phone data to improve fore­casts of crit­ical eco­nomic indicators—information that is extremely valu­able to pol­i­cy­makers in the public and pri­vate sectors.

In par­tic­ular, the team found that call detail records can be used to pre­dict unem­ploy­ment rates up to four months before the release of offi­cial reports and more accu­rately than using his­tor­ical data alone.

Our find­ings are of great prac­tical impor­tance, poten­tially facil­i­tating the iden­ti­fi­ca­tion of macro­eco­nomic sta­tis­tics faster and with much finer spa­tial gran­u­larity than tra­di­tional methods of tracking the economy,” said Lazer, a Dis­tin­guished Pro­fessor of Polit­ical Sci­ence and Com­puter and Infor­ma­tion Sci­ence.

We are hope­fully just begin­ning to learn what this data can tell us, and the promise of more accu­rate, less expen­sive, and higher-​​resolution mea­sures of crit­ical eco­nomic indi­ca­tors is very exciting,” added lead author Jameson Toole, a doc­toral stu­dent at the Mass­a­chu­setts Insti­tute of Tech­nology. “We hope that our results can be used to help pol­i­cy­makers react more rapidly to future eco­nomic down­turns, giving them a more accu­rate pic­ture of the state of the economy.”

In the paper, Lazer, Toole, and their collaborators—a quartet of experts in eco­nomics, engi­neering, public policy, and infor­ma­tion sci­ence from MIT, Har­vard Uni­ver­sity, the Uni­ver­sity of Pitts­burgh, and the Uni­ver­sity of Cal­i­fornia, Davis—harnessed the power of algo­rithms to ana­lyze call record data from two undis­closed Euro­pean coun­tries. Their first study focused on unem­ploy­ment at the com­mu­nity level, where they exam­ined the behav­ioral traces of a mass layoff at an auto-​​parts man­u­fac­turing plant in 2006.

A schematic view of the rela­tion­ship between job loss and call dynamics. A schematic view of the rela­tion­ship between job loss and call dynamics. Researchers used the calling behavior of indi­vid­uals to infer job loss and mea­sure its effects. They then mea­sured the vari­ables and included them in pre­dic­tions of unem­ploy­ment at the macroscale, sig­nif­i­cantly improving fore­casts. Image from Journal of the Royal Society Interface.

 

Using call record data span­ning a 15-​​month period between 2006 and 2007, they designed a so-​​called struc­tural break model to iden­tify mobile phone users who had been laid off. Then they tracked the mobility and social inter­ac­tions of the affected workers, looking at sev­eral quan­ti­ties related to their social behavior, including total calls, number of incoming calls, number of out­going calls, and calls made to indi­vid­uals phys­i­cally located at the plant.

The find­ings revealed that job loss had a “sys­tem­atic damp­ening effect” on their mobility and social behavior. For example, the researchers found that the total number of calls made by laid-​​off indi­vid­uals dropped 51 per­cent fol­lowing their layoff when com­pared with non-​​laid-​​off res­i­dents while their number of out­going calls decreased 54 per­cent. What’s more, the month-​​to-​​month churn of a laid-​​off person’s social network—that is, the frac­tion of con­tacts called in the pre­vious month that were not called in the cur­rent month—increased approx­i­mately 3.6 per­centage points rel­a­tive to con­trol groups. In terms of mobility, they found that the number of unique mobile phone towers vis­ited by people who had lost their jobs decreased 17 per­cent rel­a­tive to a random sample.

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

The paper’s second study ana­lyzed the call detail records of thou­sands of sub­scribers in a dif­ferent Euro­pean country, one that had expe­ri­enced macro­eco­nomic dis­rup­tions during the period in which the data was available.

This time, the researchers looked for behav­ioral changes that may have been caused by layoffs—fewer out­going calls, for example, or an increase in churn—to deter­mine whether those changes could pre­dict gen­eral unem­ploy­ment statistics.

Indeed, they found that changes in mobility and social behavior pre­dicted unem­ploy­ment rates before the release of offi­cial reports and more accu­rately than tra­di­tional fore­casts. Specif­i­cally, the researchers noted that their novel methods allowed them to pre­dict present unem­ploy­ment rates two-​​to-​​eight weeks prior to the release of tra­di­tional esti­mates and fore­cast future employ­ment rates up to four months ahead of offi­cial reports.

While Lazer praised the rapidity, accu­racy, and cost-​​effectiveness of collecting—and sub­se­quently analyzing—passively gen­er­ated data from dig­ital devices, he cau­tioned against viewing his group’s methods as a sub­sti­tute for survey-​​based approaches to detecting and pre­dicting future unem­ploy­ment rates. “We con­sider mobile phone data a pow­erful yet com­ple­men­tary tool,” he explained. “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.”

-By Jason Kornwitz

More Stories

Photo of the Capitol Building at night

High stakes for politics, SCOTUS in 2018

01.04.2018
Photo of the crashed truck that was used in the October 31st attack in Manhattan.

Weaponizing Language: How the meaning of “allahu akbar” has been distorted

11.08.2017
Northeastern logo

Why I love studying Spanish

05.29.20
Uncategorized