Mobile phone network data are potentially interesting for producing official statistics, in particular on present population, mobility, and tourism. One of the most important parts of the statistical inference on mobile phone network data is the estimation of geographic location. The spatial information that is contained in such data are the locations of the network’s radio cells the mobile devices are connected to over time.
In the present paper this spatial information is refined by estimating a distribution of probable geolocations of connected devices around each radio cell of the network. The authors propose to make these estimates via a Bayesian framework which is modular, in the sense that its various methodological components are interchangeable. The framework allows to incorporate any prior knowledge, such as land use, about where devices are expected to be, and any module which estimates the likelihood of connection to a cell, given a geolocation.
The paper develops concrete implementations of these modules, including a connection module which takes signal strength and overlapping coverage areas of cells into account. The resulting location probability distributions enrich the mobile phone network data, which can be further used for statistical inference.