The Economist Highlights Ph.D. Student Emily Aiken’s Research Using Machine-Learning and Mobile-Phone Data

In poor countries, statistics are both undersupplied and underused

In the rich world, people worry that prying governments know too much about them. Popular culture valorises characters who go off the grid, like Jack Reacher (the hero of 25 novels by Lee Child and two films starring Tom Cruise). He drifts around America on Greyhound buses, eschewing a driving licence, credit cards and email...

Given the shortage of conventional statistics, many people are enthusiastic about more novel forms of data, gleaned from mobile phones, social media and satellite imagery. In the early months of the covid-19 pandemic, patterns of mobile-phone use showed who could and could not afford to stay at home in a city like Jakarta, outlining the uneven impact of lockdown measures in many developing countries. That kind of data can help donors better target their aid efforts. Emily Aiken of the University of California, Berkeley, and her colleagues have tested whether a machine-learning algorithm can identify the poorest households in 80 Afghan villages based on mobile-phone data, such as the duration of their calls, their network of contacts, and how often they paid for more minutes of call-time. For the 80% of households that owned a mobile phone, the algorithm worked about as well as more traditional targeting methods, such as counting fridges, clothes irons, and other physical assets.

But, as the study’s authors are careful to note, not everyone owns a mobile phone. And algorithms that work in one place and time may not necessarily travel well or endure for long. Joshua Blumenstock of Berkeley has pointed out that international calls may be a less reliable indicator of prosperity during the Haj pilgrimage season, when many more people travel...

Listen, or read more...