Inferring and Modeling Migration Flows Using Mobile Phone Network Data

Abstract

Estimating migration flows and forecasting future trends is important, both to understand the causes and effects of migration and to implement policies directed at supplying particular services. Over the years, less research has been done on modeling migration flows than the efforts allocated to modeling other flow types, for instance, commute. Limited data availability has been one of the major impediments for empirical analyses and for theoretical advances in the modeling of migration flows. As a migration trip takes place much less frequent compared to the commute, it requires a longitudinal set of data for the analysis. This study makes use a massive mobile phone network data to infer migration trips and their distribution. Insightful characteristics of the inferred migration trips are revealed, such as intra/inter-district migration flows, migration distance distribution, and origin-destination (O-D) movements. For migration trip distribution modelling, log-linear model, traditional gravity model, and recently introduced radiation model were examined with different approaches taken in defining parameters for each model. As the result, the gravity and log-linear models with a direct distance (displacement) used as its travel cost and district centroids used as the reference points perform best among the other alternative models. A radiation model that considers district population performs best among the radiation models, but worse than that of the gravity and log-linear models. INDEX TERMS Migration flows, trip distribution modeling, mobile phone network data, gravity model, log-linear model, radiation model.

A general approach to detecting migration events in digital trace data

“…This body of work, and the way the data are used to measure migration, are summarized in S1 Table. In work most closely related to the current study, Blumenstock [7] proposed a rudimentary method for inferring migration from phone data, which defined a migration event as one in which an individual remains within one administrative unit for k consecutive months and then a different administrative unit for k consecutive months. This approach, and its slight variations, have subsequently been used to study migration using phone data [10,21,36] and social media data [8].…”

“…For example, Zagheni et al [8] and Fiorio et al [18] assign users to the county from which they posted the majority of tweets during a specified period of time. Papers using phone data typically assign home locations based on the cell tower or administrative unit with the densest call activities [7,9,[19][20][21]. See S1 Table for a full inventory of the prior work using trace data to study migration.…”

Evaluation of home detection algorithms on mobile phone data using individual-level ground truth

“…The many applications of “big data” analytics to any kind of official statistics depend critically on our ability to identify, with more or less error, where someone lives , i.e., detecting an individual’s home location. This impacts all aspects of the work on statistics with non-traditional data sources such as the estimation of population density [ 10 , 18 , 38 ], commuting and migration flows [ 5 , 15 , 17 , 19 , 28 ], air pollution [ 21 , 37 ], and the estimation of privacy risk [ 8 , 9 , 12 , 32 , 33 ], and is of special importance now to inform epidemic models of COVID-19 transmission [ 34 ]. The knowledge of the home location of individuals forms the crucial link between digital data and census data, making it a key enabler for the integration of these two sources of information.…”

Looking for a better future: modeling migrant mobility

“…Previous research shows that the gravity model performs better than the radiation model in multiple scenarios (Bouchard and Pyers 1965) including migrations (Hankaew et al 2019;Poot et al 2016). The radiation model underestimates the flows across many modes of transportation, specially if destinations have larger populations (Masucci et al 2013).…”