This paper examines patterns of spatial segregation of Syrian refugees in Turkey, and how patterns of spatial segregation influence internal mobility decisions of refugees and natives as they move to other regions within the country. The analysis is based on anonymized Call Detail Records (CDRs) from Turk Telekom for their Syrian customers and for a large sample of Turkish customers, covering the period from January 1 to December 31, 2017. The database includes phone activity for a sample of nearly one million customers, out of which approximately 185,000 are tagged as refugees. The granularity of the data permits the analysis of calls made/received by refugees and natives at a geographically disaggregated level (i.e. for each cell phone tower) for each hour in 2017.
Using call volumes for refugees and natives at each cell phone tower as a proxy for the population distribution, the authors construct two indices (widely used in academia):
- A dissimilarity index measuring the share of refugees that would have to move from high to low concentration regions to match their average distribution across the country, i.e. to achieve full integration (an index of 1 implies complete segregation); and
- A normalized isolation index measuring the probability that refugees interact with the native population (an index of 1 implies no interaction).
The authors compare these indices over time, across regions of Turkey that have been differently exposed to the inflow of refugees, and across hours of the day. They then estimate gravity models of internal mobility between provinces to understand the determinants of refugees’ and natives’ mobility decisions during 2017. Using Istanbul as a case study to investigate the factors associated with internal mobility within provinces, the authors also estimate gravity models for internal migration between 41 districts in Istanbul.
- Dissimilarity is generally stable over time, while isolation tends to increase over time. The dissimilarity index is relatively stable over time both at the national and provincial levels while the isolation index gradually increases over time, possibly due to increases in the market share of Turk Telekom or cell phone penetration among Syrian refugees (since the specification of the isolation index is sensitive to the share of Syrian refugees in the sample).
- There is more segregation in provinces with a smaller share of refugees. Provinces with a higher share of refugees tend to have significantly lower levels of dissimilarity and isolation, i.e. these provinces tend to be more integrated and there is more interaction between refugees and natives. Conversely, provinces with a lower share of refugees tend to have higher levels of dissimilarity and isolation, i.e. these provinces tend to be more segregated and there is less interaction between refugees and natives.
- Refugees tend to move to provinces that host a larger number of refugees and lower levels of segregation, while levels of segregation do not influence natives’ mobility decisions. Refugees tend to move to provinces that have large overall populations, high income per capita, larger numbers of refugees, and lower measures of segregation. Similar economic factors influence the mobility decisions of natives, but segregation has no effect.
- Economic factors and diaspora networks are the key determinants of intra-provincial mobility within Istanbul. Refugees tend to move to districts that have larger numbers of refugees, but segregation has no effect.
- Residential segregation appears to be higher than labor market segregation. Both dissimilarity and isolation indices dip during the day, especially in early afternoon and increase in the evening. These patterns imply that there is large variation between where refugees live and work. This finding may explain why segregation has no impact on intra-provincial mobility decisions.