SEATTLE — In the age of the internet and telecommunications, mobile phones can do much more than connecting people through calls and text messages. University of Washington (UW) researchers are now using mobile phone metadata for a greater cause – fighting worldwide poverty.
Poverty mapping has always been a difficult task, given the high cost and long timespan needed to perform ground-based surveys, especially in impoverished and war-ravaged countries. Without sufficient and up-to-date data, NGOs and policymakers are unable to distribute aid efficiently or carry out poverty reduction projects in the regions that are the most in need.
Led by Joshua Blumenstock, an assistant professor in the UW Information School and adjunct professor in computer science and engineering (now an assistant professor at the University of California, Berkeley), researchers at UW found a way to access the financial profiles of the poorer regions of the world through analysis based on mobile phone metadata.
How It Works
The research focused on Rwanda, an East African country with a population of 12 million. Beginning with 1,000 random phone interviews done by students at the Kigali Institute of Science and Technology – a project supervised by Blumenstock in 2009 – researchers recorded hints of wealth among a large amount of metadata generated by texts and calls.
Blumenstock explained in an interview with UW News that the information gathered by the phone interviews is an important reference, as it shows the phone-using habits of relatively well-to-do individuals.
Researchers then correlated that information with a larger sample of metadata from a Rwandan telephone company, attempting to sort out “socioeconomic hallmarks” about the phone owners.
With the help of a highly sophisticated machine learning algorithm, researchers discovered simple patterns in the ways people with different financial situations use the phone.
Patterns in the Data
The number of calls made, the frequency of calls made every day, the amount of money people pre-pay for phone time, and the ratio of calls made to received are all indicators of the phone owner’s wealth.
More calls made usually means the phone owner may be wealthier. A high frequency of calls during business hours often suggests a stable office job. More money put onto the phone can suggest a better income. Because Rwandan telephone companies charge more for making calls than receiving calls, the number of calls made or received might say something about the owner’s wealth as well.
Once a simple pattern is confirmed to be an accurate representation of the wealth of the sampled population, researchers can extrapolate to the mobile phone metadata of the entire population, constructing an estimated map of poverty distribution.
Why Mobile Phone Metadata Matters
Not only is it much faster, the entire poverty mapping process with mobile phone metadata costs only a fraction of the money that would have been spent on ground-based surveys, greatly reducing the cost while providing a much more reliable geographical poverty estimation.
After the research paper on the project was published in the journal Science in 2015, many other scholars are trying to build on Blumenstock’s result and come up with more accurate and sufficient poverty mapping techniques ready for field application.
Researchers at the University of Southampton made use of both mobile phone metadata and satellite imagery to estimate the poverty distribution in Bangladesh. Their progress showed that, even when taking into consideration that people in the poorest regions may not be able to afford cell phones, the wealth-estimating method still works quite well.
For a long time, “small area estimation” based on available survey data has been the way to construct the poverty profile of a region. Due to the large number of variables and lack of data, it has long been criticized but never abandoned.
The new method that utilizes phone metadata not only appears to be a promising next-generation poverty mapping system, but also opens many doors for frontier technologies like machine learning to take part in solving social problems. With the help of these ever-evolving technologies, the end of global poverty could come sooner rather than later.
– Chaorong Wang