Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data

Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data Wenlan Zhang University of Oxford/ University College London Douglas Leasure University of Oxford Daniel Valdenegro University of Oxford Xiang Ao University of Oxford Melinda C. Mills University of Oxford/ University of Groningen December 2025 Mapineq deliverable D1.7 Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 2 Mapineq – Mapping inequalities through the life course– is a three-year project (20222025) that studies the trends and drivers of intergenerational, educational, labour market, and health inequalities over the life course during recent decades. The research is run by a consortium of eight partners: University of Turku, University of Groningen, National Distance Education University, WZB Berlin Social Science Center, Stockholm University, Tallinn University, Population Europe, and University of Oxford Website: www.mapineq.eu The Mapineq project has received funding from the European Union’s Horizon Europe research and innovation programme under the grant agreement No. 101061645. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union, the European Research Executive Agency, or their affiliated institutions. Neither the European Union nor the granting authority can be held responsible for them. Acknowledgement: This document was reviewed by Prof Gerard J. van den Berg (University of Groningen) as part of Mapineq quality assurance procedures. The content of the document, including opinions expressed and any remaining errors, is the responsibility of the authors. Publication information: This work is licensed under the Creative Commons Attribution-Non-Commercial-Share Alike 4.0 International (CC BY-NC-SA 4.0) license. You are free to share and adapt the material if you include proper attribution (see suggested citation), indicate if changes were made, and do not use or adapt the material in any way that suggests the licensor endorses you or your use. You may not use the material for commercial purposes. Summary history Version Date Comments 1.0 12.11.2025 Initial version 1.1 19.12.2025 Review version 1.2 23.12.2025 Reviewed for submission Suggested citation: Zhang, W., Leasure, D., Valdenegro, D., Ao, X and M.C. Mills. (2025). Policy-induced push migration: Measuring cross-border mobility with Meta data. Turku: INVEST Research Flagship Centre / University of Turku. DOI: 10.5281/zenodo.18029636 Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 3 Executive summary Nations often assume that their own internal national migration policies are the primary levers to control migration. Using novel high-frequency digital mobility data from Meta combined with a causal BACI (Before-After-Control-Impact) design, we demonstrate how migration flows into countries that have not changed migration policy can unintentionally be triggered by migration policy reforms in another country. Introducing this policy-induced push migration framework, we test whether host/transit-country rules in Chile (2021) and Turkey (2021–22) generated measurable onward migration movements along the Chile (CL) to United States (US) and Turkey (TR) to Germany (DE) corridors, respectively. Results at a glance − Using two examples, we provide causal evidence that origin-driven changes in migration policy within sending countries operated as a push factor for migration, meaning that migration was not a pull factor to the destination country. − CL→US: +~240% cumulative mobility increases versus synthetic controls; ~57,450 additional movers in the post period (strong pre-fit; placebo p ≈ 0.048) − TR→DE: +~122% cumulative mobility increases versus synthetic controls; ~53,944 additional movers (strong pre-fit; placebo p ≈ 0.048) − Robustness of results: results hold under donor restrictions and leave-one-out checks. Data and Method Data comes from Meta International Migration Data, anonymised monthly flow data derived from opt-in device locations of users on Meta’s social media platforms. We exploit its timeliness, coverage outside of formal channels, and harmonised comparability, while acknowledging limits and interpreting conservatively with strict diagnostics. Impacts are identified via a Before–After–Control–Impact (BACI) design implemented with Synthetic Control, comparing each treated corridor to a data-driven synthetic counterfactual: − Treated unit: policy corridor (e.g., CL→US, and TR→DE). − Counterfactual: convex combination of similar corridors (donors) chosen to minimise pre-policy prediction error. − Key diagnostics: pre/post RMSPE ratio, unit and timing placebos, leave-top-donorout, origin-only and destination-only donor sets, and a share-of-destination test. Findings: Chile → United States − Treatment: Passage of Chilean Law 21.325 affecting Venezuelan residents − Impact: The change led to a large increase in outward movers from Chile to US — about 57,000 more people (~240%) compared to expected trends. − Significance: The effect is statistically meaningful. Interpretation. Law 21.325 reduced legal continuity (more out-of-country applications; tighter renewals), particularly affecting recent Venezuelan/Haitian residents. Despite Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 4 stricter US enforcement in 2020–21, outflows from Chile to the US rose—consistent with a host-country policy push rather than a US pull. Findings: Turkey → Germany − Treatment: Change in Turkish Temporary Protection rules affecting Syrians (confounded with a cost-of-living surge) − Impact: The change led to a large increase in outward movers from Turkey to Germany — about 54,000 more people (~122%) compared to expected trends − Significance: The effect is statistically meaningful. Interpretation. A cost-of-living surge, combined with the December 2021 tightening of Temporary Protection rules affecting Syrians, raised the cost and uncertainty of staying, contributing to onward movement toward Germany along established networks. To be noted, the estimate does not hinge on any one group and likely reflects a mix of Turkish and Syrian residents. Key take away − Substantial, measurable policy-induced push migration. After domestic changes, flows jump versus synthetic controls: CL→US ≈ +240% (~57k excess); TR→DE ≈ +122% (~54k). − Push rather than pull, and likely conservative estimate. Effects survive placebos and donor restrictions; spillovers into donor corridors attenuate gaps, suggesting estimates are conservative. − Measures beyond the borders. Digital trace data (with diagnostics) gives early warning of rerouting; pair it with corridor metrics and coordinated monitoring so reforms do not just export pressure to partnering countries. Policy implications − Measure beyond the border. Evaluate reforms with corridor-level metrics (treated origin-destination gaps, share-of-destination shifts) alongside national targets. − Coordinate regionally. Use findings to brief partners and align surge capacity, humanitarian provision, and protection pathways. − Public–private data partnership. Establish a Data Partnership for privacy-preserving feeds (public administration + vetted private providers) and shared dashboards. Limitations Data limitations include unknown nationality, motive ambiguity (where ≠ why), platform dependence, privacy thresholds that smooth small corridors, residual under-coverage in low-signal corridors (e.g., Haiti), and gaps on Mexico-mediated transit, making Latin America estimates conservative. The methods rely on corridor-level shocks that may coincide with policy changes, which makes it difficult to isolate the effect of the policy. In addition, they allow for spillovers between corridors and do not directly control for corridorspecific, time-varying macroeconomic conditions. Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 5 Abbreviations BACI Before–After–Control–Impact CL Chile DE Germany SCM Synthetic Control Method TR Turkey US United States (R)MSPE (Root) Mean Square Predicted Error Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 6 Content EXECUTIVE SUMMARY 3 ABBREVIATIONS 5 1. INTRODUCTION 9 1.1. THE COORDINATION CHALLENGE FOR MIGRATION POLICYMAKERS ___________________________ 9 1.2. CONCEPTUAL GAPS AND INTERDEPENDENCE IN MIGRATION SYSTEMS _________________________ 9 1.3. THE ROLE OF DIGITALISATION IN MIGRATION RESEARCH _________________________________ 10 1.4. RESEARCH SCOPE: DETECTING POLICY-INDUCED PUSH MIGRATION _________________________ 11 2. ANALYTICAL FRAMEWORK 12 2.1. DATA _____________________________________________________________________ 12 2.2. ANALYTICAL APPROACH ________________________________________________________ 15 2.3. MODEL SPECIFICATION _________________________________________________________ 15 2.3.1. DONOR CONSTRUCTION _______________________________________________________ 16 2.3.2. DATA CONSTRUCTION AND PANEL ________________________________________________ 17 2.3.3. WEIGHT ESTIMATION AND CONSTRAINTS ___________________________________________ 17 2.4. FIT DIAGNOSTICS AND VALIDITY CHECKS _____________________________________________ 17 2.4.1. MAIN INFERENCE METRIC _____________________________________________________ 17 2.4.2. PRE-FIT VALIDITY ___________________________________________________________ 17 2.4.3. PLACEBO TESTS ____________________________________________________________ 18 2.4.4. SHARE-OF-DESTINATION INFLOW TEST _____________________________________________ 18 2.5. IMPLEMENTATION ____________________________________________________________ 18 3. CASE STUDY: CHILE - UNITED STATES 19 3.1. POLICY CONTEXT _____________________________________________________________ 19 3.2. RESULTS AND FINDINGS ________________________________________________________ 20 4. CASE STUDY: TURKEY - GERMANY 23 4.1. POLICY CONTEXT _____________________________________________________________ 23 4.2. RESULTS AND FINDINGS ________________________________________________________ 24 5. CROSS-CASE DISCUSSION 28 5.1. CROSS-CASE INSIGHTS _________________________________________________________ 28 5.2. IMPLICATIONS FOR MIGRATION POLICY MAKING ________________________________________ 28 5.3. LIMITATIONS ________________________________________________________________ 29 5.4. FUTURE RESEARCH ___________________________________________________________ 30 6. CONCLUSIONS 31 7. REFERENCES 32 8. APPENDIX A: SUPPLEMENT 37 9. APPENDIX B: BASIC BACI METHOD 41 9.1. ANALYTICAL FRAMEWORK AND METHODOLOGY ________________________________________ 41 9.1.1. ANALYTICAL APPROACH _______________________________________________________ 41 9.1.2. MODEL SPECIFICATION _______________________________________________________ 41 Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 7 9.1.3. DYNAMIC EFFECTS: EVENT-STUDY EXTENSION ______________________________________ 41 9.1.4. CONTROL-GROUP CONSTRUCTION _______________________________________________ 42 9.1.5. IMPLEMENTATION ___________________________________________________________ 42 9.1.6. ROBUSTNESS AND DIAGNOSTIC PROCEDURES _______________________________________ 43 9.2. RESULTS AND FINDINGS ________________________________________________________ 43 9.2.1. MAIN DID ESTIMATES________________________________________________________ 43 9.2.2. DYNAMIC EFFECTS (EVENT-STUDY) ______________________________________________ 44 9.2.3. PLACEBO AND TIMING TESTS ___________________________________________________ 46 Tables TABLE 1 KEY CHARACTERISTICS OF THE META INTERNATIONAL MIGRATION OPEN DATA _________________ 12 TABLE 2. SUMMARY STATISTICS: CL→US ________________________________________________ 22 TABLE 3. SUMMARY STATISTICS: TR→DE ________________________________________________ 27 TABLE A1. POLICY EVENT CODING AND TIMELINE_____________________________________________38 TABLE A2. CL (LOG) TOP DONOR WEIGHTS AND GROUP SHARES__________________________________38 TABLE A3. CL→US FULL RESULT ______________________________________________________ 39 TABLE A4. TR (LOG) TOP DONOR WEIGHTS AND GROUP SHARES__________________________________39 TABLE A5. TR→DE FULL RESULT ______________________________________________________ 40 TABLE B2. EXTENDED ROBUSTNESS_____________________________________________________ 44 TABLE B3. FULL EVENT-STUDY COEFFICIENTS (±K MONTHS)____________________________________ 46 Figures FIGURE 1 GLOBAL MIGRATION MAP _____________________________________________________ 13 FIGURE 2. MIGRATION NETWORKS ______________________________________________________ 14 FIGURE 3. MONTHLY MIGRANT INFLOWS BY DESTINATION CONTINENT _____________________________ 14 FIGURE 4. CL→US MAIN TREATED VS. SYNTHETIC RESULT ____________________________________ 20 FIGURE 5. CL→US PLACEBO TEST _____________________________________________________ 21 FIGURE 6. CL SHARE OF TOTAL US INFLOWS AND PRE-TREND PROJECTION __________________________ 21 FIGURE 7. TR→DE MAIN TREATED VS. SYNTHETIC RESULT ____________________________________ 25 FIGURE 8. TR→DR PLACEBO TEST _____________________________________________________ 25 FIGURE 9. TR SHARE OF TOTAL DE INFLOWS AND PRE-TREND PROJECTION __________________________ 26 FIGURE A1. INTRA-REGIONAL MIGRATION NETWORKS BY COUNTRY_______________________________ 37 FIGURE B1. CL→US DID TREATED VS. SYNTHETIC RESULT____________________________________ 43 FIGURE B2. PAIR-LEVEL SERIES: CL→US AS A HIGH-GROWTH OUTLIER___________________________ 45 FIGURE B3. EVENT-STUDY PLOTS_______________________________________________________ 45 FIGURE B4 PLACEBO TEST____________________________________________________________47 Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 8 Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data Abstract: We test whether domestic policy and macro conditions in host countries generate outward mobility to receiving countries — what we term policy-induced push migration. Using Meta International Migration Data from 2019 to 2022 and a Before–After–Control– Impact (BACI) design implemented via Synthetic Control, we compare each treated migration corridor with a data-driven counterfactual and assess inference using standard placebo and robustness checks. Two cases illustrate the mechanism: (i) Chile’s 2021 migration reform and Chile (CL)→United States (US) flows; (ii) Turkey’s late-2021 cost-ofliving surge plus administrative tightening (address verification, curtailed registrations/referrals, neighbourhood closures) and Turkey (TR)→ Germany (DE) flows. We find large, origin-driven increases in mobility relative to synthetic controls: CL→US ≈ +240% (≈57k cumulative excess migration) and TR→DE ≈ +122% (≈54k). Treating both the origin and the destination clarifies that effects are not primarily pull-driven. Because spillovers can affect donor corridors, estimates are conservative lower bounds. Limitations include the absence of globally comparable monthly corridor covariates and potential timevarying shocks. Policy implications include attention to considering reforms beyond one’s own national border, internationally coordinated migration policy, and public–private data partnerships to monitor real-time migration flows. Policy-induced migration Host country reforms export pressures along specific corridors rather than suppress overall movement. Origin-driven push effect Post-treatment flows vs. synthetic control: CL→US ≈ +240% (~57k); TR→DE ≈ +122% (~54k). Measures beyond the border Use digital trace data corridor metrics to coordinate regionally so reforms do not merely shift pressure elsewhere. ________ ________ ________ Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 9 1. Introduction 1.1. The Coordination Challenge for Migration Policymakers As migration dynamics become more complex and interconnected, governments face a dual challenge: managing border crossings while anticipating how their own policies shape movements beyond their borders and how policies abroad drive flows to their borders (Tagliacozzo et al., 2024). This increasing complexity reflects the interplay of multiple interacting migration drivers, the growing interconnectedness of countries and policy feedback effects, and improved data on flows, which together create non-linear, interdependent migration dynamics (Chiaramello et al., 2024) . National migration policies often pursue domestic goals, such as border control, labour supply, and humanitarian protection, but these choices rarely operate in isolation or within controllable national borders (Charles-Edwards et al., 2025). Because mobility is inherently transnational, one country’s decisions can inadvertently and rapidly shape the broader international migration context in which other nations must respond. Beyond regular movements for study, work, or family, global mobility is increasingly driven by sudden exogenous shocks such as conflict, political instability, and natural disasters (Solano and Massey, 2022). Such crises can displace large populations within weeks, transforming neighbouring states into hosts of refugees and other displaced people. The Syrian conflict, for instance, reshaped migration across the Middle East, while Venezuela’s collapse produced mass displacement throughout South America. As host countries adapt their legal and administrative systems to manage these inflows, their responses influence not only those they receive, but also the wider geography of movement across regions (Brzozowski and Coniglio, 2021). These cross-border ripple effects illustrate how domestic reforms in one country – although designed to stabilise migration – can generate volatility and new interdependencies across international migration systems. Large-scale displacement typically unfolds in stages, often referred to as step migration or serial migration trajectories (Paul, 2011). In general, people move to immediate neighbouring countries, where entry is feasible, social ties exist, or assistance is accessible (UNHCR, 2019). As conditions evolve, such as tightening registration rules, shrinking work opportunities, expiring permits, or saturation in host/transit communities, some individuals undertake a second step to third countries (Schapendonk, 2012). This onward movement does not necessarily reflect new pull incentives abroad; rather, it is frequently a response to changing constraints in the initial host/transit nation. Policies that narrow legal continuity (e.g., stricter renewals, address verification, locality quotas) can therefore redirect rather than reduce flows, with impacts materialising along corridors connecting the first host to subsequent destinations (Bertoli and Fernández-Huertas Moraga, 2013; Czaika and Hobolth, 2016). 1.2. Conceptual Gaps and Interdependence in Migration Systems Recent scholarship has paid growing attention to how migration policy reforms shape broader movement patterns (Czaika and De Haas, 2013). Studies of the COVID-19 pandemic, the United Kingdom’s withdrawal from the European Union, and changing Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 16 potential outcome without treatment is 𝐹𝑖𝑡 𝑁, and the treatment effect for the treated unit is 𝜏𝑡= 𝐹1𝑡 −𝐹1𝑡 𝑁 The synthetic control approximates the untreated counterfactual for the treated unit as a convex combination of donor units 𝑗=2,...,𝐽+1: 𝐹 1𝑡 𝑁=∑𝑤𝑗𝐹𝑗𝑡 𝐽+1 𝑗=2 , 𝑤ℎ𝑒𝑟𝑒 𝑤𝑗≥0,∑𝑤𝑗=1 𝐽+1 𝑗=2 Weights are chosen to minimize pre-intervention mean squared prediction error (MSPE): 𝑊∗=𝑎𝑟𝑔𝑚𝑖𝑛𝑊≥0, ∑𝑤𝑗=1, ∑(𝐹1𝑡 −∑𝑤𝑗𝐹𝑗𝑡 𝐽+1 𝑗=2 )2 𝑡<𝑇𝑜 Estimated effects are: − Point effect (any 𝑡): 𝜏𝑡= 𝐹1𝑡 − 𝐹 1𝑡 𝑁 − Cumulative post-treatment effect (levels): 𝐶𝑈𝑀 = ∑(𝐹1𝑡 −𝐹 1𝑡 𝑁) 𝑡≥𝑇0 − Percent change over the post period (levels): %𝛥 =∑(𝐹1𝑡 −𝐹 1𝑡 𝑁) 𝑡≥𝑇0 ∑𝐹 1𝑡 𝑁 𝑡≥𝑇0 Estimation can be done on raw counts or on the log-transform 𝑍𝑖𝑡 =𝑙𝑜𝑔(1+𝐹𝑖𝑡). When fitted on 𝑍𝑖𝑡 (our primary specification), effects are reported in levels by back-transforming via 𝑒𝑥𝑝(𝑧)−1 before computing CUM (total cumulated effect) and %𝛥 (percentage change). 2.3.1. Donor construction For the treated pair (TREATED_FROM→TREATED_TO), we form two donor sets: − Origin-only donors (push benchmark): TREATED_FROM→𝑑 for destinations 𝑑≠TREATED_TO. − Destination-only donors (pull benchmark): 𝑜→TREATED_TO for origins 𝑜≠TREATED_FROM. Within each set we keep the Top-K pairs (default K=10) whose total flow over the full sample is closest to the treated pair’s total (L1 distance in totals). Some origins may be excluded if they appear in the pool (e.g., {RU, UA, AF, HT}). This is to avoid bias from countries experiencing major domestic crises, such as wars, conflicts, or natural disasters since these could independently drive large displacement patterns unrelated to the policy of interest. The combined donor pool is the union of the two sets. Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 17 2.3.2. Data construction and panel − Monthly bilateral counts are reshaped to a unit–time panel. − To obtain a balanced series, missing months are filled by bidirectional interpolation on levels (no forward-fill of zeros). − Donor units must have at least 12 pre-treatment months. − The primary working scale is 𝑙𝑜𝑔(1+𝑦); raw-count runs are used as sensitivity checks. 2.3.3. Weight estimation and constraints Weights are estimated by solving the convex problem above with nonnegativity and a simplex (sum-to-one) constraint using OSQP (Operator Splitting Quadratic Program, first choice), with SCS (Splitting Conic Solver) as a fallback when needed. The OSQP finds the weights that minimise the distance between the treated and synthetic unit step and optimises the fit. SCS is a flexible alternative given it is more flexible and robust on complex problems and can take on more types of constraints. Group balance (combined models only). To ensure representation from both information sources, we impose group-share lower bounds in the combined and combined-leave-topdonor models: 𝛴𝑗∈𝑜𝑟𝑖𝑔𝑖𝑛 𝑤𝑗≥𝑔, 𝛴𝑗∈𝑑𝑒𝑠𝑡𝑖𝑛𝑎𝑡𝑖𝑜𝑛 𝑤𝑗≥𝑔 2.4. Fit diagnostics and validity checks 2.4.1. Main inference metric Post/pre MSPE ratio and a permutation p-value. Preand post-treatment Root Mean Squared Prediction Error (RMSPE) is computed as: 𝑅𝑀𝑆𝑃𝐸𝑝𝑟𝑒 =√1 𝑇0−1 ∑(𝐹1𝑡 −𝐹 1𝑡 𝑁)2 𝑡<𝑇0, 𝑅𝑀𝑆𝑃𝐸𝑝𝑜𝑠𝑡 =√1 𝑇−𝑇0+1 ∑(𝐹1𝑡 −𝐹 1𝑡 𝑁)2 𝑡≥𝑇0 Their ratio provides a scale-free measure of fit deterioration: 𝑅=𝑀𝑆𝑃𝐸𝑝𝑜𝑠𝑡 𝑀𝑆𝑃𝐸𝑝𝑟𝑒 2.4.2. Pre-fit validity − Pre-RMSPE and MSPE. − 𝑅2 and correlation between treated and synthetic. − MAPE in levels after back-transform (robust to working scale). − Pre-trend test on the pre-period gap (OLS of the treated–synthetic gap), reporting slope and p-value. Pre-treatment RMSPE assesses the model’s ability to reproduce the observed outcome trajectory prior to the intervention. A low value indicates a credible counterfactual fit. Posttreatment RMSPE captures the discrepancy between predicted and observed outcomes after the intervention. Substantively meaningful effects are suggested when a close pretreatment fit is accompanied by a pronounced increase in post-treatment RMSPE. Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 18 2.4.3. Placebo Tests − Unit placebos (in space). Each donor unit is, in turn, treated as “pseudo-treated” and fitted using the remaining donors (no group constraints or caps in placebos). We compare the treated unit’s post/pre MSPE ratio to the placebo distribution and report the finite-sample p-value 𝑃=#{𝑅𝑝𝑙𝑎𝑐𝑒𝑏𝑜 𝑟𝑎𝑡𝑖𝑜 ≥𝑅𝑡𝑟𝑒𝑎𝑡𝑒𝑑 𝑟𝑎𝑡𝑖𝑜}+1 𝑁𝑝𝑙𝑎𝑐𝑒𝑏𝑜𝑠 +1 − Month placebos (in time). We shift the treatment date to the last 𝑁 pre-treatment months (default 𝑁=12), keeping the post-window length equal to the actual post period. For each pseudo-date we refit the synthetic and compute cumulative ATT (Average Treatment Effect on the Treated). We present the distribution of placebo cumulative effects and mark the realised effect. The ATT is the average causal effect experienced by the individuals that actually received the treatment, not the entire population. Thus, in our case it shows on average how much the migration-related intervention/policy changed the outcomes for those who were exposed to it. 2.4.4. Share-of-destination inflow test We compute the monthly share of US inflows accounted for by Chile, 𝑠ℎ𝑎𝑟𝑒𝑡=𝐶𝐿→𝑈𝑆𝑡 𝛴𝑜0→𝑈𝑆𝑡 and estimate 𝑠ℎ𝑎𝑟𝑒𝑡=𝛼+𝛽⋅𝑡+𝛾⋅𝐷𝑡+𝑢𝑡 by OLS with HAC (Heteroskedasticityand Autocorrelation-Consistent, Newey–West) standard errors (6 lags for monthly data). A significant 𝛾>0 indicates Chile’s inflow rose relative to other origins or in other words, evidence of a push component. 2.5. Implementation To demonstrate this effect, two case studies were examined (with details provided in Appendix Table A1 and described in the next section): − Chile (2019): introduction of a consular visa for Venezuelan nationals and tightening of residence renewals. − Turkey (2022): deactivation of temporary protection IDs, intensified enforcement (Feb–Jul 2022), and concurrent surges in inflation that increased economic stress. To ensure a full understanding of migration dynamics and to distinguish push effects from possible pull effects, the analysis was conducted from both perspectives. Each event was tested once with the policy country as the treated origin country (i.e., Chile and Turkey), capturing whether domestic reforms generated outward “push” pressures, and again with the main destination country (i.e., the United States, and Germany) as the treated unit, to check whether any observed rise reflected a corresponding “pull” effect in receiving contexts. Comparing results across these two specifications helps to confirm that observed Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 19 surges primarily originate from policy-induced changes within the host or transit country, rather than demand shifts in the destination. Appendix Table A1 summarises the timeline of key policy milestones for both treated corridors. For each case we run two perspectives: push (treated origin = policy country) and pull (treated destination = main receiver). We set 𝑇0 to the policy month used in the Results. Donors come from: − Combined: same origin to other destinations, and other origins to the same destination − Origin-only: same origin to other destinations; − Destination-only: other origins to the same destination Within the latter two, we keep the Top-10 flows closest in total volume to the treated flow. Data are monthly; we build a balanced panel by bidirectional interpolation and require ≥12 pre-months per donor. Primary specification. Estimation on 𝑙𝑜𝑔 (1 + 𝑦); combined donor pool = origin ∪ destination; group-share lower bounds 𝑔≥0.40 for each group; no per-donor cap. Effects are reported in levels after being back-transformed. 3. Case Study: Chile - United States 3.1. Policy Context Chile has long served as a regional migration hub, being one of South America’s most economically advanced and politically stable countries. For decades, migration to Chile was governed by the 1975 Migration Decree Law (DL 1,094), which allowed relatively flexible entry and residence for foreign nationals (Doña-Reveco, 2022). Combined with sustained economic growth and political stability, these conditions made Chile one of the region’s most welcoming destinations. By 2020, migrants accounted for approximately 9% of the national population, with Venezuelan nationals representing 30.7% and Haitians making up 12.5% of the migrant population (Instituto Nacional de Estadísticas, 2025). Following the COVID-19 pandemic, Chile introduced a series of immigration reforms aimed at formalising temporary stays and tightening residence renewal procedures—most notably Law 21 325, the New Migration and Foreigners Law, enacted on 11 April 2021 and later fully implemented in February 2022 (Doña-Reveco, 2022). The law replaced Chile’s 1975 immigration framework with a comprehensive system intended to modernise visa management, strengthen border control, and promote orderly and regular migration. Although formally applied to all foreign nationals, its implementation had uneven effects. By requiring most visa applications to be lodged from abroad and restricting in-country renewals, it disproportionately constrained migrants already living in Chile under temporary or irregular status. This particularly affected recently arrived Venezuelans and Haitians, who faced additional exclusion due to socioeconomic vulnerability and existing visa barriers. As legal pathways narrowed, many residents began seeking alternative destinations, most visibly to the United States. Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 20 Chile’s reform offers a sharp, data-rich test of policy-induced push migration. We set the intervention date to 1 January 2021 to incorporate a three-month anticipation window ahead of the April 2021 announcement of Law 21.325, with subsequent rollout through 2021–2022 reinforcing the shock. The episode lies squarely within the 2019–2022 Meta mobility window. Month and origin–destination pair fixed effects absorb regional COVID19 recovery dynamics and contemporaneous visa-policy shifts elsewhere, isolating the reform’s push effect. In this sense, the Chilean case shows how a host-country rule change can redistribute cross-border movements rather than reduce overall migration. 3.2. Results and Findings Pre-treatment fit The synthetic control reproduces the pre-intervention evolution of CL→US migration with high accuracy. In the pre-period, as shown in Figure 4, Pre-RMSPE = 0.084, and the treated and synthetic series move together closely (R² = 0.979, corr = 0.991, MAPE ≈ 7.4% in levels). A linear pre-trend test on the pre-treatment gap is flat (slope ≈ 0.002; p = 0.435), supporting design validity. Figure 4. CL→US Main Treated vs. Synthetic Result: (a) Time series with vertical line at intervention date; (b) Gap plot with shaded post-period; (c) Pre-treatment fit correlation Main effect Immediately after the intervention, the treated series diverges sharply upward. PostRMSPE = 1.188, implying a Post/Pre MSPE ratio of 202.25. In unit placebos (Figure 5a), only about 1 of 20 placebos matches or exceeds this ratio (permutation p ≈ 0.048). In Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 21 timing placebos (Figure 5b), the realized cumulative ATT lies far to the right of the placebo distribution. Together with the tight pre-fit, these placebos indicate the post-intervention surge is unlikely to reflect chance or common shocks. Magnitude Back-transforming from the log specification, the cumulative post-period ATT ≈ 57,450 migrants (+239.87% relative to the synthetic). A share-of-US inflow test in Figure 6 shows Chile’s share of total US arrivals increases by ≈ 0.0057 (≈ 0.57 points) post-intervention (p = 0.029) relative to a pre-trend projection, while the trend itself is not significant (p = 0.446). Figure 5. CL→US Placebo test (a) Unit-placebo MSPE ratios; (b) Timing-placebo distribution of cumulative ATT Figure 6. CL share of total US inflows and pre-trend projection Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 22 Robustness to donor composition The combined model allocates ≈60% weight to origin-side donors and ≈40% to destinationside donors (the 0.40 destination-share floor binds). Dropping the top donor (CL→CO) leaves the result essentially unchanged (Pre-RMSPE = 0.098; Post/Pre = 130.69; p ≈ 0.050; cumulative ATT ≈ 55,520; +214.53%). Donor-set restrictions To verify that the effect does not hinge on cross-node mixing or contemporaneous policies at either end, we re-estimate the model using restricted donor subsets (see Table 2). With origin-only donors, the post-period increase remains large (cumulative ATT ≈ 65,918; +425.76%; Post/Pre = 109.94; p ≈ 0.091). With destination-only donors, it is also positive (cumulative ATT ≈ 41,156; +102.27%; Post/Pre = 4.71; p ≈ 0.545). Taken together, these checks indicate that the headline effect is not an artefact of donor composition. Table 2. Summary statistics: CL→US Together, these results summarised in Table 2 provide strong and robust evidence that the post-intervention surge in CL→US migration was driven mainly by origin-side dynamics in Chile, rather than by changes in US policy or conditions. The finding is consistent across robustness checks—including leave-one-out analyses, placebo tests, and alternative donor specifications. Additional diagnostics—log-level estimates, donor-specific placebo plots, and sensitivity tests—are also provided in Appendix Table A2 and A3. Interpretation Chile’s post-covid migration landscape helps explain the post-intervention surge. In 2019, Venezuelans were the largest contributor to inflows to Chile, accounting for 44.2% in our dataset. Law 21.325 altered that trend. By shifting most visa applications out of country and tightening in-country renewals/regularisations, the reform raised the risk of status discontinuity for people already in Chile. For Model PreRMSPE PostRMSPE Post/PreMSPE Permutatio n p-value N placebos Cumulative ATT % Change Log Result Combined donors 0.084 1.188 202.25 0.048 20 57450 239.87 Origin-only donors 0.157 1.643 109.94 0.091 10 65918 425.76 Destinationonly donors 0.350 0.759 4.71 0.545 10 41156 102.27 Leave top donor out 0.098 1.117 130.69 0.050 19 55520 214.53 Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 23 Venezuelans navigating expiring permits, backlogs, and stricter renewal rules, the expectation (and experience) of reduced legal continuity plausibly prompted reoptimisation—internal moves, attempts at regional re-entry from abroad, or onward exits from Chile altogether. Given the size and recency of the Venezuelan migrant stock, it is reasonable that a large share of the observed CL→US response reflects Venezuelans moving on, rather than long-settled migrants initiating first-time moves. At the same time, US border enforcement intensified: migrant apprehensions and removals more than doubled between March 2020 and September 2021 (Solano and Massey, 2022). Despite this deterrent environment, outflows from Chile to the United States continued to rise. This pattern suggests Chile’s domestic reform functioned as a strong “push” within a wider regional displacement dynamic, rather than being offset by a US “pull” or by heightened enforcement. Empirically, we observe a tight pre-fit, a sharp and persistent post-break, placebo results consistent with a true effect, and robustness to donor composition. Together, these features demonstrate how a policy shock in a host (Chile) redistributes migration across borders rather than reduces it. The small but detectable rise in Chile’s share of US inflows is consistent with displacement operating through established networks (information, financing, travel companions) that lower the marginal cost of onward movement once regular pathways in Chile narrow (Beine et al., 2011; Munshi, 2003). 4. Case Study: Turkey - Germany 4.1. Policy Context Turkey was host to the world’s largest Syrian refugee population under a national Temporary Protection (TP) regime administered by the Presidency of Migration Management (PMM). By end-2021, about 3.74 million Syrians were registered in Turkey (UNHCR, 2025, 2022), and reports also note additional Syrians present who are not registered (Kachmar, 2023). From December 2021, authorities launched an addressverification campaign to confirm registered addresses (UNHCR, 2022). On 16 May 2022, PMM announced that 781 neighbourhoods in various provinces were closed to new registrations and certain residence/transfer procedures (PMM, 2022b). This was expanded to 1,169 neighbourhoods as of 1 July 2022 (PMM, 2022a), with a tighter 20% foreign-resident threshold from 25%. In parallel, from 6 June 2022 new arrivals seeking TP could be referred to Temporary Accommodation Centres (TACs) for registration under a circular that set limited exceptions (PMM, 2022a). Meanwhile, from late 2021 into 2022, Turkey faced a rapid cost-of-living surge. After the lira’s record lows around 20 Dec 2021 (News Agencies, 2021), the government introduced FX-protected deposits (KKM) to stabilise the currency (20–21 Dec) (Central Bank of the Republic of Turkey (CBRT), 2022, 2021). Price pass-through accelerated and on 1 Jan 2022, administered tariffs rose sharply. Electricity bills increased by about 50% for lowuse households and 100%+ for higher tiers/commercial users, while natural gas rose 25% for households (Reuters, 2022). Official CPI corroborates the squeeze: 36.1% YoY in Dec2021, 48.69% YoY in Jan-2022, peaking at 85.51% YoY in Oct-2022 (Ministry of Treasury and Finance, 2022a, 2022b, 2022c). Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 24 Taken together, the closure of neighbourhoods to new registrations, tighter residence/transfer rules and TAC referrals (which disproportionately targeted Syrians), combined with a rapid inflationary squeeze, constrained regularisation and internal relocation options and compressed real incomes—generating a broad outward push that is not specific to any single nationality. We set December 2021 as our intervention date because it aligns two concurrent triggers of outward pressure: (i) the start of the nationwide address-verification campaign, the first administratively enforceable tightening that could directly affect residency continuity and mobility; and (ii) the onset of the macro cost-of-living shock, crystallised by the lateDecember FX dislocation and immediate price pass-through ahead of the 1 January 2022 administered tariff increases. Anchoring in December captures anticipatory responses to the subsequent neighbourhood-closure steps while treating policy and macro stress as a combined push. The window fits data coverage, and pair and month fixed effects net out EU-side pulls and common shocks. The mobility series do not identify nationality, thus our estimates capture aggregate outward pressure from Turkey. 4.2. Results and Findings Pre-treatment fit The synthetic control closely tracks the pre-intervention evolution of TR→DE migration. As shown in Figure 6c, for the combined donor pool, Pre-RMSPE = 0.098, with a strong prefit (R² = 0.950, corr = 0.975, MAPE ≈ 7.8% in levels). A linear pre-trend test on the pretreatment gap is flat (slope ≈ 0.0017; p = 0.321), supporting design validity. Main effect After the (Dec 2021) intervention, the treated series diverges upward. As shown in Figure 7a and 7b, for the combined donors, Post-RMSPE = 0.775, yielding a Post/Pre MSPE ratio of 62.30. In unit placebos (reassigning treatment across donor units), the permutation p ≈ 0.048; taken with the tight pre-fit, this indicates the post-intervention rise is unlikely to reflect chance or common shocks. Month-shift (timing) placebos were also run (12 pseudointervention dates) and see Figure 8 for the distribution. Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 25 Figure 7. TR→DE Main Treated vs. Synthetic Result (a) Time series with vertical line at intervention date; (b) Gap plot with shaded postperiod; (c) Pre-treatment fit correlation Figure 8. TR→DR Placebo test (a) Unit-placebo MSPE ratios; (b) Timing-placebo distribution of cumulative ATT Magnitude From the log specification, the cumulative post-period ATT ≈ 53,944 migrants (+121.71% relative to the synthetic). A share-of-DE destination test (Figure 9) shows a small, Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 32 7. References Angrist, J.D., Pischke, J.-S., 2009. Mostly Harmless Econometrics: An Empiricistís Companion. Princeton university press. Araujo, M., Mejova, Y., I., Weber., 2022. pySocialWatcher: A Social Data Collector from Facebook Marketing API. Araujo, M., Mejova, Y., Weber, I., Benevenuto, F., 2017. Using Facebook Ads Audiences for Global Lifestyle Disease Surveillance, in: Proceedings of the 2017 ACM on Web Science Conference. ACM, New York, NY, USA, pp. 253–257. https://doi.org/10.1145/3091478.3091513 Aslan, D., 2022. Turkey restricts temporary, international protection registry in 16 provinces [WWW Document]. Dly. Sabah. URL https://www.dailysabah.com/politics/turkey-restricts-temporary-internationalprotection-registry-in-16-provinces/news (accessed 11.9.25). Avramescu, A., Wiśniowski, A., 2021. Now-casting Romanian migration into the United Kingdom by using Google Search engine data. Demographic Research 45, 1219–1254. https://doi.org/10.4054/DemRes.2021.45.40 Bakewell, O., 2014. Relaunching migration systems. Migr. Stud. 2, 300–318. https://doi.org/10.1093/migration/mnt023 Beine, M., Docquier, F., Özden, Ç., 2011. Diasporas. J. Dev. Econ., Symposium on Globalization and Brain Drain 95, 30–41. https://doi.org/10.1016/j.jdeveco.2009.11.004 Bertoli, S., Fernández-Huertas Moraga, J., 2013. Multilateral resistance to migration. J. Dev. Econ., Migration and Development 102, 79–100. https://doi.org/10.1016/j.jdeveco.2012.12.001 Beverelli, C., 2022. Pull factors for migration: The impact of migrant integration policies. Econ. Polit. 34, 171–191. https://doi.org/10.1111/ecpo.12191 Brekke, J.-P., Røed, M., Schøne, P., 2017. Reduction or deflection? The effect of asylum policy on interconnected asylum flows. Migr. Stud. 5, 65–96. https://doi.org/10.1093/migration/mnw028 Brzozowski, J., Coniglio, N., 2021. International Migration and the (Un)happiness Push: Evidence from Polish Longitudinal Data. Int. Migr. Rev. 55, 1089–1120. https://doi.org/10.1177/01979183211004835 Callaway, B., Sant’Anna, P.H.C., 2021. Difference-in-Differences with multiple time periods. Journal of Econometrics, Themed Issue: Treatment Effect 1 225, 200– 230. https://doi.org/10.1016/j.jeconom.2020.12.001 Central Bank of the Republic of Turkey (CBRT), 2022. 2022 Annual Report. Merkez Bankasi. Central Bank of the Republic of Turkey (CBRT), 2021. Press Release on Encouraging Conversion of FX Deposits to TL Time Deposits [WWW Document]. Merk. Bankasi. URL https://www.tcmb.gov.tr/wps/wcm/connect/EN/TCMB%2BEN/Main%2BMenu/ Announcements/Press%2BReleases/2021/ANO202162?utm_source=chatgpt.com (accessed 11.11.25). Charles-Edwards, E., Bernard, A., Rowe, F., Abel, G., 2025. International Migration and Development: The Changing Impact of Migration on Redistributing Global Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 33 Population. Int. Migr. Rev. 59, 850–866. https://doi.org/10.1177/01979183231175114 Chi, G., Abel, G.J., Johnston, D., Giraudy, E., Bailey, M., 2025. Measuring global migration flows using online data. Proc. Natl. Acad. Sci. 122, e2409418122. https://doi.org/10.1073/pnas.2409418122 Chiaramello, D., Cortinovis, R., Kajander, N., Loeschner, J., Majorano Sarapo, F., Rosinska, A., Scapolo, F., Sermi, F., Talò, T., Tarchi, D., Bongiardo, D., Crespi, I., Dara, A., Sofos, I. (Eds.), 2024. Atlas of migration 2024. Publications Office of the European Union, Luxembourg. https://doi.org/10.2760/5025638 Czaika, M., de Haas, H., 2017. The Effect of Visas on Migration Processes. Int. Migr. Rev. 51, 893–926. https://doi.org/10.1111/imre.12261 Czaika, M., De Haas, H., 2013. The Effectiveness of Immigration Policies. Popul. Dev. Rev. 39, 487–508. https://doi.org/10.1111/j.1728-4457.2013.00613.x Czaika, M., Hobolth, M., 2016. Do restrictive asylum and visa policies increase irregular migration into Europe? Eur. Union Polit. 17, 345–265. https://doi.org/10.1177/1465116516633299 Data for Good at Meta, 2025. International Migration Flows | Humanitarian Dataset. de Haas, H., 2025. Towards a global migration paradigm. Comp. Migr. Stud. 13, 76. https://doi.org/10.1186/s40878-025-00481-9 Doña-Reveco, C., 2022. Chile’s Welcoming Approach to Immigrants Cools as Numbers Rise [WWW Document]. migrationpolicy.org. URL https://www.migrationpolicy.org/article/chile-immigrants-rising-numbers (accessed 8.31.25). Donato, K., Jacobs, E., Singh, L., Arab, A., Wycoff, N., 2025. Using Organic Data in Migration Research. Int. Migr. Rev. 59, 5–36. https://doi.org/10.1177/01979183231193627 Filauri, F., Pazzona, M., Rosina, M., 2024. New Evidence on Post-Brexit UK Migration Trends (Technical Report). Brunel University London. Grow, A., Perrotta, D., Del Fava, E., Cimentada, J., Rampazzo, F., Gil-Clavel, S., Zagheni, E., Flores, R.D., Ventura, I., Weber, I., 2022. Is Facebook’s Advertising Data Accurate Enough for Use in Social Science Research? Insights from a CrossNational Online Survey. J R Stat Soc Ser A Stat Soc 185, S343–S363. https://doi.org/10.1111/rssa.12948 Guichard, L., Machado, J., 2025. The externalities of immigration policies on migration flows: the case of an asylum policy. J. Econ. Geogr. 25, 59–74. https://doi.org/10.1093/jeg/lbae016 Human Rights Watch, 2022. Turkey: Hundreds of Refugees Deported to Syria. URL https://www.hrw.org/news/2022/10/24/turkey-hundreds-refugees-deportedsyria (accessed 11.12.25). Iacus, S.M., Porro, G., Salini, S., Siletti, E., 2020. Controlling for Selection Bias in Social Media Indicators through Official Statistics: a Proposal. Journal of Official Statistics 36, 315–338. https://doi.org/10.2478/jos-2020-0017 ImmiChile, 2022. Entra en vigencia la nueva Ley de Migraciones. ImmiChile. URL https://immichile.cl/entra-en-vigencia-la-nueva-ley-de-migraciones/ (accessed 11.9.25). Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 34 Instituto Nacional de Estadísticas, 2025. Demografía [WWW Document]. Demografía. URL http://www.ine.gob.cl/estadisticas/sociales/demografia-yvitales/demografia-y-migracion/demografía (accessed 10.23.25). International Monetary Fund, 2025. Chapter 3 Journeys and Junctions: Spillovers from Migration and Refugee Flows. International Monetary Fund, Washington, D.C. https://doi.org/10.5089/9798400289583.081 Kachmar O., 2023. Protecting Syrian Refugees in Turkey from Forced Repatriation. New Lines Inst. URL https://newlinesinstitute.org/displacement-andmigration/protecting-syrian-refugees-in-turkey-from-forced-repatriation/ (accessed 11.7.25). Kveladze, I., Jokinen, J.C., Tapia, C., Hansen, H.S., 2024. What can location-based social media reveal on human migration patterns in Europe? Journal of Maps 20. https://doi.org/10.1080/17445647.2024.2329168 Leasure, D.R., Kashyap, R., Rampazzo, F., Dooley, C.A., Elbers, B., Bondarenko, M., Verhagen, M., Frey, A., Yan, J., Akimova, E.T., Fatehkia, M., Trigwell, R., Tatem, A.J., Weber, I., Mills, M.C., 2023. Nowcasting Daily Population Displacement in Ukraine through Social Media Advertising Data. Popul. Dev. Rev. 49, 231–254. https://doi.org/10.1111/padr.12558 Lechner, M., 2011. The Estimation of Causal Effects by Difference-in-Difference Methods. ECO 4, 165–224. https://doi.org/10.1561/0800000014 Lyu, L., Cheng, M., Chen, Y., Min, X., 2025. Spatial Mobility Under Crisis: Policy Responses and Migration Dynamics in Five Countries. Appl. Spat. Anal. Policy 18, 121. https://doi.org/10.1007/s12061-025-09734-7 Massey, D.S., 1988. Economic Development and International Migration in Comparative Perspective. Popul. Dev. Rev. 14, 383–413. https://doi.org/10.2307/1972195 Mesnard, A., Savatic, F., Senne, J.-N., Thiollet, H., 2024. Revolving Doors: How Externalization Policies Block Refugees and Deflect Other Migrants across Migration Routes. Popul. Dev. Rev. 50, 607–642. https://doi.org/10.1111/padr.12650 Mills, M., 2024. Mapineq Link: Leveraging real-world data for research. Ministry of Treasury and Finance, 2022a. Update on Turkish Economy. Ministry of Treasury and Finance, 2022b. Update on Turkish Economy. Ministry of Treasury and Finance, 2022c. Update on Turkish Economy. Minora, U., Bosco, C., Iacus, S.M., Grubanov-Boskovic, S., Sermi, F., Spyratos, S., 2022. The potential of Facebook advertising data for understanding flows of people from Ukraine to the European Union. EPJ Data Sci. 11, 59. https://doi.org/10.1140/epjds/s13688-022-00370-6 Munshi, K., 2003. Networks in the Modern Economy: Mexican Migrants in the U. S. Labor Market*. Q. J. Econ. 118, 549–599. https://doi.org/10.1162/003355303321675455 News Agencies, 2021. Turkish lira rockets after Erdogan’s promise to protect deposits [WWW Document]. Al Jazeera. URL https://www.aljazeera.com/economy/2021/12/21/turkish-lira-rebounds-aftererdogans-anti-dollarization-plan (accessed 11.11.25). Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 35 Paglino, E., 2024. Estimating excess migration associated with tropical storms in the USA 1990–2010. Popul Environ 46, 11. https://doi.org/10.1007/s11111-02400452-9 Palotti, J., Adler, N., Morales-Guzman, A., Villaveces, J., Sekara, V., Herranz, M.G., AlAsad, M., Weber, I., 2020. Monitoring of the Venezuelan exodus through Facebook’s advertising platform. PLOS ONE 15, e0229175. https://doi.org/10.1371/journal.pone.0229175 Paul, A.M., 2011. Stepwise International Migration: A Multistage Migration Pattern for the Aspiring Migrant. Am. J. Sociol. 116, 1842–1886. PMM, 2022a. Mahalle Kapatma Duyurusu Hk. [WWW Document]. TC İçişleri Bakanl. Göç İdaresi Başkanl. URL https://www.goc.gov.tr/mahalle-kapatma-duyurusuhk?utm_source=chatgpt.com (accessed 11.9.25). PMM, 2022b. Mahalle Kapatma Duyurusu hk. [WWW Document]. TC İçişleri Bakanl. Göç İdaresi Başkanl. URL https://www.goc.gov.tr/mahalle-kapatma-duyurusuhk2?utm_source=chatgpt.com (accessed 11.9.25). Rampazzo, F., Bijak, J., Vitali, A., Weber, I., Zagheni, E., 2021. A Framework for Estimating Migrant Stocks Using Digital Traces and Survey Data: An Application in the United Kingdom. Demography 58, 2193–2218. https://doi.org/10.1215/00703370-9578562 Rampazzo, F., Rango, M., Weber, I., 2023. New Migration Data: Challenges and Opportunities, in: Bertoni, E., Fontana, M., Gabrielli, L., Signorelli, S., Vespe, M. (Eds.), Handbook of Computational Social Science for Policy. Springer International Publishing, Cham, pp. 345–359. https://doi.org/10.1007/978-3031-16624-2_18 Reuters, 2022. Turkey jacks up electricity, gas prices for New Year [WWW Document]. MarketScreener. URL https://www.marketscreener.com/news/latest/Turkeyjacks-up-electricity-gas-prices-for-New-Year-37448707/ (accessed 11.11.25). Ribeiro, F.N., Benevenuto, F., Zagheni, E., 2020. How Biased is the Population of Facebook Users? Comparing the Demographics of Facebook Users with Census Data to Generate Correction Factors, in: Proceedings of the 12th ACM Conference on Web Science, WebSci ’20. Association for Computing Machinery, New York, NY, USA, pp. 325–334. https://doi.org/10.1145/3394231.3397923 Schapendonk, J., 2012. Turbulent Trajectories: African Migrants on Their Way to the European Union. Societies 2, 27–41. https://doi.org/10.3390/soc2020027 Solano, P., Massey, D.S., 2022. Migrating through the Corridor of Death: The Making of a Complex Humanitarian Crisis. J. Migr. Hum. Secur. 10, 147–172. https://doi.org/10.1177/23315024221119784 Tagliacozzo, S., Pisacane, L., Kilkey, M., 2024. A system-thinking approach for migration studies: an introduction. J. Ethn. Migr. Stud. 50, 1099–1117. https://doi.org/10.1080/1369183X.2023.2279708 UNHCR, 2025. Republic of Turkey [WWW Document]. UNHCR US. URL https://www.unhcr.org/us/where-we-work/countries/republic-tuerkiye (accessed 11.12.25). UNHCR, 2022. Address registry with PDMMs. Help Turkey. URL https://help.unhcr.org/turkiye/2022/04/01/address-registry-with-pdmms/ (accessed 11.9.25). UNHCR, 2019. Guidance on Responding to Irregular Onward Movement of Refugees and Asylum-Seekers. Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 36 Urbański, M., 2022. Comparing Push and Pull Factors Affecting Migration. Economies 10, 21. https://doi.org/10.3390/economies10010021 Xiang, B., Lindquist, J., 2014. Migration Infrastructure. Int. Migr. Rev. 48, 122–148. https://doi.org/10.1111/imre.12141 Zagheni, E., Weber, I., Gummadi, K., 2017. Leveraging Facebook’s Advertising Platform to Monitor Stocks of Migrants. Population and Development Review 43, 721– 734. https://doi.org/10.1111/padr.12102 Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 37 8. Appendix A: Supplement Figure A1. Intra-regional Migration Networks by Country (a) Africa; (b) Americas; (c) Asia; (d) Europe Note: Only migration links exceeding 20,000 migrants are shown, with ribbon thickness reflecting the non-directional strength of the migration connection (sum of flows in both directions over the Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 38 four-year period). Oceania is excluded from the figure because only one country pair (Australia– New Zealand) exceeds this threshold. Table A1. Policy event coding and timeline Date Policy/Event Situation Chile Jan 2021 New Migration and Foreigners Law (Law 21,325) Anticipation anchor Apr 2021 (Doña-Reveco, 2022) New Migration and Foreigners Law (Law 21,325) — promulgation & publication Promulgated Apr 11; published Apr 20 Feb 2022 (ImmiChile, 2022) Law 21,325 — entry into force (regulation published) Law becomes operative upon publication of the regulation Turkey Dec 2021 (UNHCR, 2022) Address verification & TP ID deactivations for non-verified addresses Address checks launched Feb 24, 2022 (Aslan, 2022; Kachmar, 2023) Registration closed/ restricted in 16 provinces (TP/IP); 25% neighbourhood cap signalled Public announcement by Dep. Presidency of Migration Management (PMM) May 16, 2022 (PMM, 2022a) Neighbourhood closures (781) First PMM notice operationalizing closures Jun 30, 2022 (PMM, 2022b) Closures expanded to 1,169; foreignershare cap lowered to 20% PMM update confirming effective date 1 July 2022 Table A2. CL (log) Top donor weights and group shares. Destination Control Origin Control Donor Weight Donor Weight TW→US 0.2293 CL→CO 0.2328 BD→US 0.1235 CL→CA 0.1947 JM→US 0.0471 CL→PE 0.0972 TH→US 0.0000 CL→VE 0.0623 Subtotal (dest.) 0.3999 Subtotal (orig.) 0.5870 Other donors 0.0131 Group shares ≈ 0.40 Group shares ≈ 0.60 Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 39 Table A3. CL→US Full Result Model PreRMSPE PostRMSPE Post/PreMSPE Permutation p-value N placebos Cumulative ATT % Change Log Result Combined donors 0.084 1.188 202.25 0.048 20 57450 239.87 Origin-only donors 0.157 1.643 109.94 0.091 10 65918 425.76 Destinationonly donors 0.350 0.759 4.71 0.545 10 41156 102.27 Leave CL- >CO out 0.098 1.117 130.69 0.050 19 55520 214.53 Count Result Combined donors 50.475 2810.618 3100.68 0.048 20 57975 247.5 Origin-only donors 127.536 3231.367 641.96 0.091 10 66920 462.14 Destinationonly donors 292.942 2140.413 53.39 0.182 10 41184 102.4 Leave CL- >CA out 51.411 2856.652 3087.44 0.05 19 59119 265.34 Table A4. TR (log) Top donor weights and group shares. Destination Control Origin Control Donor Weight Donor Weight RO→DE 0.5393 TR→FR 0.1717 IN→DE 0.0607 TR→IQ 0.1046 GR→DE 0.0000 TR→SY 0.0669 BA→DE 0.0000 TR→NL 0.0568 Subtotal (dest.) 0.6000 Subtotal (orig.) 0.4000 Other donors 0 Group shares ≈ 0.60 Group shares ≈ 0.40 Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 40 Table A5. TR→DE Full Result Model PreRMSPE PostRMSPE Post/PreMSPE Permutation p-value N placebos Cumulative ATT % Change Log Result Combined donors 0.098 0.775 62.3 0.048 20 53944 121.71 Origin-only donors 0.207 0.976 22.28 0.091 10 63652 183.9 Destinationonly donors 0.125 0.933 55.71 0.091 10 60081 157.35 Leave RO- >DE out 0.138 0.928 45.32 0.05 19 61281 165.7 Count Result Combined donors 289.057 5855.762 410.39 0.048 20 54193 122.97 Origin-only donors 652.33 6279.402 92.66 0.091 10 63092 179.38 Destinationonly donors 305.997 6151.791 404.17 0.091 10 58065 144.45 Leave RO- >DE out 390.956 6025.675 237.55 0.05 19 57988 143.98 Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 41 9. Appendix B: Basic BACI Method 9.1. Analytical Framework and Methodology 9.1.1. Analytical Approach The analysis adopts a Before–After–Control–Impact (BACI) design to evaluate whether discrete policy events in specific origin or transit countries led to measurable changes in outward migration. The BACI framework isolates intervention effects by jointly exploiting temporal variation (before versus after the policy) and spatial variation (treated versus control migration corridors). It therefore distinguishes policy-induced changes from background fluctuations in mobility that occur over time or across routes. 9.1.2. Model Specification Let 𝐹𝑖𝑗𝑡 denote the monthly number of migrants from origin country 𝑖 to destination country 𝑗 at time 𝑡. We model the logarithm of migrant counts as: 𝑙𝑜𝑔(𝐹𝑖𝑗𝑡) =𝛼𝑖+𝛾𝑡+𝛽𝐵𝐴𝐶𝐼𝑖𝑗𝑡 +𝜀𝑖𝑗𝑡 Where: − 𝛼𝑖are pair fixed effects that absorb all time-invariant characteristics of the migration corridor (e.g., distance, language, historical ties); − 𝛾𝑡are month fixed effects capturing global shocks common to all routes (e.g., pandemic restrictions, seasonal patterns); − 𝛽𝐵𝐴𝐶𝐼𝑖𝑗𝑡 =𝑇𝑟𝑒𝑎𝑡𝑒𝑑𝑖× 𝑃𝑜𝑠𝑡𝑡 is an interaction equal to one for treated pairs after the event; − 𝜀𝑖𝑗𝑡 is an idiosyncratic error term. The parameter of interest, 𝛽, measures the average treatment effect of the policy on the treated route. Because the dependent variable is in logs, 𝑒𝑥𝑝(𝛽)−1 can be interpreted as the approximate percentage change in monthly migrant counts attributable to the event. Standard errors are clustered by migration pair to allow for arbitrary serial correlation within routes. To accommodate potential differences in underlying trends, we also estimate an extended specification with pair-specific linear trends: 𝑙𝑜𝑔(𝐹𝑖𝑗𝑡) =𝛼𝑖+𝛾𝑡+𝛿𝑖𝑡+𝛽𝐵𝐴𝐶𝐼𝑖𝑗𝑡 +𝜀𝑖𝑗𝑡 Where 𝛿𝑖𝑡 allows each route to follow its own gradual evolution over time, improving robustness to non-parallel pre-treatment trajectories. 9.1.3. Dynamic Effects: Event-Study Extension To examine how effects evolve before and after the policy, we implement an event-study (or dynamic BACI) specification. This decomposes the single treatment coefficient into a series of leads and lags relative to the intervention month: Report: Policy-Induced Push Migration: Measuring Cross-Border Mobility with Meta Data 48 Interpretation: The Chilean policy is associated with a sharp rise in recorded outmigration to the United States. Across models, the estimated treatment effect ranges from 0.96 to 1.30 log points (≈ +160 – +265 %), statistically significant under clustered inference. Event-study estimates confirm strong positive post-policy deviations but also reveal statistically significant pre-treatment dynamics, implying that parallel trends are not fully met. Placebo tests reinforce that CL→US is an outlier relative to other corridors, yet specific months in 2019 and 2020–2021 produce similar spikes when treated as policy dates. Consequently, the results should be interpreted as evidence of a large temporal association, consistent with a strong “push” effect from Chile’s domestic policy, rather than definitive causal proof.