Predicting Refugee Movements
Google searches for the names of Turkish provinces are significantly correlated with the number of Syrian refugees in those areas, and may be useful in estimating refugee movements. This is the main finding of the study "Search for a New Home: Refugee Stock and Google Search" by Ebru Şanlıtürk of the Max Planck Institute for Demographic Research in Rostock, Germany, and Francesco Billari of the Bocconi Department of Social and Political Sciences in which they use online search data to predict the distribution of Syrian refugees under temporary protection in Turkey. The authors concluded that Google searches are good predictors for estimating the intentions of refugee populations, especially when official data are not available.
The main objective of this study is to assess the ability of Google search data to predict Syrian refugees’ movements in Turkey from 2016 to 2019. The authors use an innovative approach that exploits the difference between the Turkish and Arabic alphabets to distinguish between host and migrant populations in online searches. This method allows them to track interest in different Turkish provinces in real time and correlate this interest with Syrian refugee movements.
Google Trends data on searches for Turkish province names, split between Arabic and Turkish languages, were compared with official refugee data provided by Turkey's Directorate General of Migration Management. A unique dataset covering four years was then created, including weekly updates on search popularity and refugee populations. Arabic-language searches were considered an indicator of refugee displacement intentions, while Turkish-lanuage searches reflected the interest by local population.
The statistical model employed considered province-specific factors and explored the relationship between the search popularity ratio (SPR) in Arabic and Turkish and the number of refugees in Turkish provinces. The authors posited that online searches may precede actual movements, providing a time advantage for predictions over official data that are often available much later.
The results actually showed that an increase in the popularity ratio of Arabic searches over Turkish searches correlated with an increase in the number of Syrian refugees in the corresponding Turkish provinces. This relationship proved particularly useful when examined with time lags of one, two, or more weeks, suggesting that online searches actually precede physical movements. For example, an increase of 1 in the SPR report one week earlier is associated with a 1.6 percent increase in the number of refugees in a certain province.
The authors point out that "the association between online searches and refugee movements is significant and can offer a valuable opportunity for predicting migration patterns in contexts of forced displacement." This is particularly relevant in situations where official data are limited or out of date, offering a useful alternative for policy makers and humanitarian organizations.
The study thus highlights the potential of online search data as a forecasting tool for forced migration. With some caveats, such as those related to data representativeness and the digital divide, the results suggest that integrating online search data into forecasting strategies can significantly improve the ability to respond to migration flows in real time.
