The ABCDE of Big Data: Assessing Biases in Call-Detail Records for Development Estimates
结合塞内加尔2013年通话详单记录和人口普查数据，分析基于手机数据估计人口密度等发展指标时的偏差，并提出利用双重差分法减少偏差、监测城市化与迁移等变化的方法。
Abstract This article contributes to improving our understanding of biases in estimates of demographic indicators, in the developing world, based on Call Detail Records (CDRs). CDRs represent an important and largely untapped source of data for the developing world. However, they are not representative of the underlying population. We combine CDRs and census data for Senegal in 2013 to evaluate biases related to estimates of population density. We show that: (i) there are systematic relationships between cell-phone use and socio-economic and geographic characteristics that can be leveraged to improve estimates of population density; (ii) when no ‘ground truth’ data is available, a difference-in-difference approach can be used to reduce bias and infer relative changes over time in population size at the subnational level; (iii) indicators of development, including urbanization and internal, circular, and temporary migration, can be monitored by integrating census data and CDRs. The paper is intended to offer a methodological contribution and examples of applications related to combining new and traditional data sources to improve our ability to monitor development indicators over time and space.