Simulation Framework Reveals Major Spatial Blind Spots in Classical Migration Theories

Standard mathematical theories of human migration can achieve outstanding statistical goodness-of-fit in empirical regression models while simultaneously failing to reproduce the real-world spatial organization of population movements, according to an extensive study published in the European Journal of Population. Demographer Micol Matilde Morellini introduced an innovative four-step simulation framework designed to evaluate theoretical models not merely through conventional explanatory metrics, but by their ability to generate accurate spatial geography. Applying this methodology to international mobility across thirty-one European countries between 2002 and 2021, the investigation tested two foundational theoretical frameworks: the classical gravity model and Migration Systems Theory. Although both frameworks attained high explanatory power in standard econometric estimations, computer-simulated counterfactual migration systems derived from their theoretical assumptions proved unable to replicate essential spatial features of the European migration landscape. In particular, the models failed to capture the high bilateral reciprocity that characterizes contemporary European cross-border mobility and struggled to mirror the observed inflow profiles of Eastern and Northern European nations.

The research addresses a persistent methodological divide that has characterized migration studies for decades, during which scholarship has remained fractured between two parallel intellectual traditions. On one side, researchers in sociology, economics, and social demography have focused primarily on identifying migration drivers, examining the micro-level aspirations, meso-level networks, and macro-level structural factors that motivate or constrain relocations. On the other side, scholars working within human geography, spatial demography, and population ecology have concentrated on migration patterns, analyzing the emergent spatial configurations, corridor densities, and aggregate structures that describe how human flows organize across physical territory. Although these two dimensions are conceptually complementary, empirical methodologies have seldom examined them in tandem. Standard research designs operationalize migration drivers as independent variables in statistical models, using goodness-of-fit measures to determine whether predictors explain flow volumes without evaluating whether resulting spatial arrangements resemble actual geographic systems. Existing spatial econometric methods incorporate spatial lags or autocorrelations primarily as control inputs to correct bias, treating space as a nuisance rather than an outcome of substantive theoretical interest.

To investigate how theory-guided drivers translate into spatial patterns, the study examined two paradigms representing opposite ends of theoretical complexity and spatial articulation. The gravity model, established by Ravenstein in the late nineteenth century and widely applied across demographic research, is an intentionally parsimonious macro-structural formulation. It posits that bilateral migration volume increases with the population sizes of origin and destination locations while decaying with the physical distance between them. Because its explanatory power relies on dyadic and largely symmetric attributes, the gravity model carries implicit expectations of relatively even and balanced exchanges across geographic space, though it rarely makes explicit claims about system-wide spatial organization. In sharp contrast, Migration Systems Theory, extensively synthesized by Massey and colleagues, integrates a comprehensive set of economic differentials, geographic proximity, institutional ties, shared languages, and cumulative causation through prior migrant stocks. Migration Systems Theory conceptualizes migration systems as structured by historically embedded hierarchical relationships, explicitly anticipating core-periphery configurations characterized by concentrated corridors, strong directional asymmetries, and low levels of bilateral reciprocity.

To implement the empirical comparison, Morellini formalized a four-step sequence linking theoretical drivers to simulated spatial configurations. In Step 1, bilateral migration flows were modeled using Poisson Pseudo-Maximum Likelihood (PPML) regressions that incorporated origin, destination, and year fixed effects. PPML estimation was chosen for its mathematical robustness to heteroskedasticity, its capacity to handle zero-valued migration flows, and its consistency under correct conditional mean specifications. To control for unobserved heterogeneity, common macro shocks, and dyadic clustering, standard errors were three-way clustered by origin, destination, and year. A baseline model containing only fixed effects served as an initial benchmark. The gravity specification incorporated geographic distance between national capitals alongside the lagged population ratio between origins and destinations. The Migration Systems Theory model expanded upon gravity by incorporating contiguity, shared official languages, lagged bilateral GDP per capita ratios, bilateral migrant stocks from 2000 to proxy cumulative network effects, joint former Eastern Bloc membership, and lagged joint European Union membership.

The econometric results demonstrated that both theoretical models achieved substantial statistical explanatory power. All models except the baseline passed the Ramsey Regression Equation Specification Error Test (RESET) with p-values greater than 0.10, confirming correct model specification. In the gravity regression, corridors with 10 percent greater geographic distance experienced migration flows that were approximately 8 percent lower, while the lagged population ratio remained statistically insignificant. The gravity model achieved a squared correlation of 0.59 between observed and fitted values and an adjusted pseudo-R-squared of 0.77. Migration Systems Theory demonstrated even higher explanatory fit, achieving a squared correlation of 0.65 and an adjusted pseudo-R-squared of 0.82. In the systems regression, physical distance lost statistical significance, whereas socio-historical linkages became dominant. Bilateral flows between countries sharing a common official language were estimated to be approximately 67 percent larger, and shared geopolitical history in the former Eastern Bloc increased flows by 169 percent. Network effects also proved influential, as a 10 percent increase in bilateral migrant stocks in 2000 was associated with a 3.9 percent increase in subsequent annual flows.

Under conventional research paradigms, these regression metrics would suggest that Migration Systems Theory provides a superior explanation of European mobility. However, Steps 2 through 4 examined whether these models could actually reproduce observed spatial geography. In Step 2, fitted flows and their sampling distributions were utilized to simulate 10,000 synthetic migration systems for each theoretical specification and each year in the panel. Draws were taken independently for each origin-destination dyad from truncated normal distributions bounded by regression confidence intervals, ensuring non-negative values and propagating 95 percent estimation uncertainty into the spatial domain. The simulated flows were then calibrated row-wise so that total outflows from each origin country exactly matched observed national outflows in the corresponding year. This calibration preserved the empirical volume of national departures while allowing theoretical predictors to dictate how migrants dispersed among destinations, generating an ensemble of counterfactual migration matrices.

In Step 3, four descriptive spatial indices were computed for each simulated matrix and compared directly against observed data. Spatial concentration was evaluated using three complementary metrics: the Aggregate Coefficient of Variation (ACV), which measures relative dispersion and exhibits sensitivity to dominant corridors; the Gini Index, which measures overall corridor inequality across the complete system; and the Migration Inequality Index, which captures deviations from a hypothetical uniform distribution where all corridors receive equal flows. System reciprocity was measured using a minimum-based index that calculates the share of migrants participating in bidirectional exchanges by summing the smaller of the two directional flows in each bilateral corridor and dividing by total system volume. In Step 4, model performance was evaluated across spatial scales, utilizing visual diagnostic distributions and formal multivariate Mahalanobis distances that compare observed flow vectors against the covariance structure of simulated distributions.

The system-level results revealed striking discrepancies between explanatory power and spatial accuracy. For spatial concentration, both models performed reasonably well, with Migration Systems Theory slightly outperforming gravity across all three concentration indices. The gravity model aligned closely with the observed Gini and Migration Inequality metrics, confirming that physical distance and relative population size suffice to explain general flow concentration, although it underpredicted the Aggregate Coefficient of Variation in certain years due to difficulty capturing extreme corridor dominance. However, both theories failed catastrophically when evaluated on migration reciprocity. In the observed European system, reciprocity frequently surpassed 0.40—meaning that more than 40 percent of all cross-border migrants moved within corridors characterized by two-way flows. In contrast, both theoretical models predicted substantially more asymmetric structures, failing to reproduce observed reciprocity in eighteen out of twenty panel years. While gravity simulated slightly higher reciprocity than Migration Systems Theory, both frameworks systematically underestimated the bidirectional nature of intra-European movement.

The author explained that this widespread failure to replicate reciprocity reflects institutional dynamics unique to Europe that traditional theories fail to incorporate. The European Union’s freedom of movement regime, established under the Maastricht Treaty and Free Movement Directive alongside bilateral pacts with non-member states such as Switzerland, grants citizens reciprocal rights to live, work, and study across borders. This institutional environment encourages temporary labor migration, student exchanges, and circular mobility where individuals return home or move repeatedly between countries. These fluid movements generate balanced, bidirectional corridors even alongside substantial economic disparities. Because both gravity and Migration Systems Theory view migration as predominantly unidirectional flows driven by asymmetric economic incentives or core-periphery structures, their simulated counterfactuals underestimated actual reciprocity. Country-level evaluations using Mahalanobis distances further illustrated this spatial divergence: while Migration Systems Theory achieved lower distances across Central European states like Germany, Austria, and Poland, gravity better captured Nordic inflows, and both models exhibited severe misfit (Mahalanobis distances exceeding 4.00) across peripheral states including Cyprus, Estonia, Lithuania, Slovakia, and Slovenia.

Granular corridor diagnostics and regional analyses highlighted specific blind spots where both frameworks broke down. While contiguous corridors like Austria to Germany were reproduced accurately by both models, and historically linked pairs like Cyprus to Greece were captured exclusively by Migration Systems Theory, neither theory replicated reciprocal corridors like Estonia and Finland, which saw massive northward labor movements from 2010 to 2015 followed by southward return flows from 2015 to 2020. Regional scatterplots confirmed that migration within Eastern Europe, within Northern Europe, and from Northern to Eastern Europe was systematically mispredicted. Although the study relied on aggregate bilateral data from the Human Migration Database without distinguishing individual life courses, the simulation architecture demonstrated remarkable computational scalability, modeling millions of moves in seconds using standard software. Ultimately, Morellini’s findings demonstrate that high statistical goodness-of-fit does not guarantee geographical validity, establishing simulation-based spatial testing as an essential tool to ensure migration theories accurately reflect the real-world architecture of human mobility.

Subject of Research: Spatial accuracy testing of international migration theories using simulation methods

Article Title: A Simulation Approach for the Spatial Testing of Migration Theories

Article References: Morellini, M. M. (2026). A Simulation Approach for the Spatial Testing of Migration Theories. European Journal of Population, 42(1), Article 12. https://doi.org/10.1007/s10680-026-09770-0

Image Credits: AI Generated

DOI: 10.1007/s10680-026-09770-0

Keywords: migration theories, spatial demography, simulation modeling, gravity model, migration systems theory, European migration, Poisson pseudo-maximum likelihood, population mobility, Simulation, Approach, Spatial, Testing