A gravity analysis of refugee mobility using mobile phone data

Michel Beine, Luisito Bertinelli, Rana Comertpay, Anastasia Litina, and Jean-Francois Maystad

Journal of Development Economics, Volume 150 (2021)



This paper examines patterns of refugee mobility across provinces in Turkey using mobile phone data.

Based on measures of mobility calculated from phone data, the authors use a standard gravity model to estimate the determinants of refugee movements. Gravity models are used extensively in the voluntary migration literature to analyze the determinants of migration movements, taking into account differences in employment opportunities and income per capita between areas of origin and destination, as well as geographical and cultural distance as proxies for migration costs. In this paper, the authors derive a gravity equation that includes variables for: (a) differences in income per capita between areas of origin and destination; (b) distance between locations as a proxy for migration costs; (c) other factors shaping the attractiveness of the areas of origin and destination, specifically humanitarian aid and asylum grants; and (d) the occurrence of particular events such as outbreaks of violence or protests that could affect the perceived attractiveness of a location.

The analysis is based on call detail records from the Data for Refugees Turkey (D4R) Challenge, a non-profit initiative that aimed to expand research on the living conditions and social integration of Syrian refugees in Turkey. Data include the location of 100,000 randomly selected mobile phone transactions (involving 50,000 refugees and 50,000 non-refugees) recorded by cell towers across 26 regions in Turkey in 2017. For the levels of income at origin and destination, the authors use data on quarterly GDP per capita from the Turkish Statistical Institute.

Main results:

  • Refugees tend to move more often than non-refugees, but they move shorter distances. Compared to non-refugees, refugees are less likely to travel longer distances, possibly due to the practical difficulties and costs associated with moving across regions.
  • Refugees respond to income differences between regions, but tend to respond to ‘pull’ and ‘push’ factors differently to non-refugees. Compared to non-refugees, refugees are less likely to respond to income levels at destination, possibly because they don’t have access to the same information about the destination. Moreover, while refugees tend to leave relatively poor areas, non-refugees do not show any propensity to leave poor areas, possibly because of a stronger attachment to their current location.
  • The finding that refugees are motivated to move between regions for economic reasons is confirmed when alternative explanations of mobility are taken into account, including the propensity of refugees to cross Turkey from East to West, their propensity to leave refugee camps, and the attraction to agricultural areas during the harvest season.
  • Refugees appear to be sensitive to humanitarian aid and asylum grants. An increase in the provision of these services tends to decrease their probability of moving out of their current location. However, they are not systematically attracted to locations providing higher levels of these services.

In their conclusion, the authors note that the findings on refugees’ mobility can inform a more optimal provision of aid and support across refugee hosting regions. In addition, insights into how refugees respond to differentials in economic attractiveness across locations is important to better understand whether refugee movements contribute to a more efficient allocation of labor across space.


Big Data