High-resolution measures of poverty and vulnerability in Afghanistan: Cost-effective solutions based on mobile phone data

To make informed decisions, policymakers need accurate and timely information on the social and economic state of a nation and its population. Reliable measures of economic activity, population density, physical security, and migration are a few examples of information that play a critical role in guiding public policy. In Afghanistan and many other developing countries, it can be difficult and costly to collect and monitor such population statistics at a national scale. In many regions of the country, traditional survey data collection would be prohibitively difficult to coordinate. Policymakers in Afghanistan are thus forced to make difficult choices without access to many of the relevant sources of information that would be available in wealthy nations.

In recent years, there has been rapid progress developing new methods for measuring national statistics and other population characteristics, based on novel sources of “big data.” Specifically, there have been several test cases demonstrating how the data generated by mobile phone operators can be used in a wide variety of pro-development contexts. These include the measurement of local poverty and wealth; accurate estimation of population density; detailed information on patterns of internal migration; and the tracking of diseases such as Dengue and Malaria. Such applications incorporate protocols that extract aggregate insights while carefully safeguarding the confidentiality and privacy of individual subscribers.

The goal of this study is to develop new methods for generating cost-effective maps of poverty and vulnerability in Afghanistan, based on the analysis of large-scale “digital exhaust” from communications networks. Specifically, we intend to develop techniques that generate high-resolution maps and measurements of population characteristics based on regionally-aggregated patterns of mobile phone use. We will collect an original dataset that captures longitudinal panel data on a broad range of welfare outcomes for a random sample of phone subscribers in Afghanistan. The responses from these face-to-face and phone-based surveys will then be merged with an existing terabyte-scale dataset from Afghanistan's largest mobile phone operator, which contains the complete mobile phone transaction histories of those individuals, as well as 7 million additional Afghan citizens. We will use this merged dataset to better understand how accurately the anonymised phone records can be used infer key development outcomes. Our team has conducted pilot work in the past that demonstrates the feasibility of this approach; the goal of this project is to carefully validate and calibrate this pilot work into a set of methods that can be used to produce rigorous and reliable high-resolution measures of poverty and vulnerability in Afghanistan.