[Paper Review] Migration patterns, friendship networks, and the diaspora: the potential of Facebook Social Connectedness Index to anticipate displacement patterns induced by Russia invasion of Ukraine in the European Union

Umberto Minora, M. Belmonte+5 more|arXiv (Cornell University)|Sep 5, 2022

Migration, Refugees, and Integration4 citations

TL;DR

This study demonstrates that Facebook's Social Connectedness Index (SCI) strongly predicts Ukrainian displacement flows within the EU during the Russia-Ukraine war, with a 0.94 Pearson correlation between SCI and refugee registrations. The SCI, combined with population and geographic data, enables high-resolution (NUTS-3) forecasting of diaspora-driven migration, offering timely, spatially detailed insights to complement official statistics for crisis response.

ABSTRACT

The conflict in Ukraine is causing large-scale displacement in Europe and in the World. Based on the United Nations High Commissioner for Refugees (UNHCR) estimates, more than 7 million people fled the country as of 5 September 2022. In this context, it is extremely important to anticipate where these people are moving so that national to local authorities can better manage challenges related to their reception and integration. This work shows how innovative data from social media can provide useful insights on conflict-induced migration flows. In particular, we explore the potential of Facebook's Social Connectedness Index (SCI) for predicting migration flows in the context of the war in Ukraine, building on previous research findings that the presence of a diaspora network is one of the major migration drivers. To do so, we first evaluate the relationship between the Ukrainian diaspora and the number of refugees from Ukraine registered for Temporary Protection or similar national schemes as a proxy of migratory flows into the EU. We find a very strong correlation between the two (Pearson's r=0.94, p<0.0001), which indicates that the diaspora is attracting the people fleeing the war, who tend to reach their compatriots, in particular in the countries where the Ukrainian immigration was more a recent phenomenon. Second, we compare Facebook's SCI with available official data on diaspora at regional level in Europe. Our results suggest that the index, along with other readily available covariates, is a strong predictor of the Ukrainian diaspora at regional scale. Finally, we discuss the potential of Facebook's SCI to provide timely and spatially detailed information on human diaspora for those countries where this information might be missing or outdated, and to complement official statistics for fast policy response during conflicts.

Motivation & Objective

• To assess the predictive power of the Ukrainian diaspora in driving migration flows within the EU following Russia’s invasion of Ukraine.
• To evaluate whether Facebook’s Social Connectedness Index (SCI) can serve as a reliable proxy for diaspora presence at regional (NUTS-3) scale.
• To provide timely, spatially granular data on diaspora networks where official statistics are missing or outdated, supporting rapid policy and humanitarian response.
• To explore the integration of innovative social media data with traditional migration indicators for improved forecasting during conflict-induced displacement.

Proposed method

• Used Pearson correlation to assess the relationship between Ukrainian diaspora (measured as Ukrainian citizens in EU countries) and refugee registrations under Temporary Protection schemes.
• Applied min-max normalization to standardize Facebook SCI and total population data, then computed weighted scores via matrix multiplication: $ W = ilde{P} imes ilde{S} $.
• Built a multivariate log-linear model: $ ext{log}(ukr ext{ extunderscore}stock) = eta_0 + eta_1 ext{log}(scaled ext{ extunderscore}sci) + eta_2 ext{log}(pop ext{ extunderscore}tot) + eta_3 ext{latitude} + ext{error} $, trained on 70% of 594 regional observations.
• Validated model performance using the remaining 30% of data, reporting explained variance (R²) on the validation set.
• Used logarithmic scaling for both observed and predicted diaspora values in visualization and model evaluation.
• Colored scatter plots by country to assess model performance across EU member states.

Experimental results

Research questions

• RQ1 To what extent does the presence of the Ukrainian diaspora in EU regions correlate with actual refugee inflows under Temporary Protection schemes?
• RQ2 Can Facebook’s Social Connectedness Index (SCI) serve as a reliable proxy for regional-level diaspora presence where official data are unavailable or outdated?
• RQ3 How well does a model combining SCI, population size, and geographic position predict the distribution of Ukrainian diaspora across NUTS-3 regions in the EU?
• RQ4 Can innovative social media data like SCI improve the timeliness and granularity of migration forecasting during conflict-induced displacement?

Key findings

• A strong and statistically significant correlation (Pearson’s $ r = 0.94 $, $ p < 0.0001 $) exists between the size of the Ukrainian diaspora and the number of refugees registered under Temporary Protection in EU countries.
• The model using SCI, population, and latitude achieved an explained variance (R²) of 0.92 on the validation set, indicating high predictive power.
• The predictive model showed highly significant p-values ($ < 0.01 $) for all predictors, confirming their statistical relevance.
• The SCI effectively captures regional diaspora dynamics at NUTS-3 level, offering spatially detailed insights where official data are often lacking.
• Excluding older migration countries improved the correlation, suggesting that recently established diaspora networks are stronger drivers of conflict-induced migration.
• Facebook’s SCI can provide high-frequency, real-time monitoring of diaspora networks, enabling faster policy response than traditional statistical reporting cycles.