Inequality is rising where social network segregation interacts with urban topology

The study investigates how social networks and geography interact to shape income inequality within cities and towns. While wealth and income disparities are growing and can be reinforced by mechanisms such as homophily and triadic closure in social networks, the spatial embedding of these networks further influences segregation. Individuals tend to form ties with geographically proximate others, and physical or administrative boundaries reduce cross-boundary connections. The research question asks whether and how urban topology—characterized by distances within towns, the spatial concentration of amenities, and physical barriers—relates to social network fragmentation, and how this fragmentation is associated with the level and evolution of income inequality. The authors posit that urban geographic features facilitating sorting and segregation are linked to more fragmented social networks, which in turn are associated with higher and rising income inequality.

Prior work shows social networks provide access to economic opportunities and can maintain or amplify inequalities through homophily and triadic closure. Macro-scale social segregation emerges when socioeconomic status shapes tie formation, affecting access to resources and information. Geography is tightly linked to inequality: residential location predicts economic outcomes, and within-city disparities are pronounced. Social ties are locally biased and less frequent across spatial boundaries, limiting both individual and collective prosperity. Urban science has examined how co-location, urban characteristics, and clustered activities influence tie formation; however, spatial communities usually appear after aggregating individual links. Economic and sociological theories stress that physical space mediates the inequality–network link: central areas can concentrate information and opportunity, while spatial barriers (rivers, railways, major roads) and administrative boundaries (e.g., school districts) facilitate segregation, sorting, and discrimination. Historical and environmental externalities (e.g., pollution) create persistent spatial inequalities. These literatures motivate examining how urban topology dimensions—distance structures, amenity concentration, and physical barriers—relate to network fragmentation and inequality.

Data and setting: The analysis focuses on 474 Hungarian towns (population ≥ 2,500), excluding Budapest due to its unique scale and administrative complexity. Social network data come from the iWiW online social network, capturing 2.8 million users and over 300 million friendship ties by end-2011, with user locations at the town level. Town-level income distributions are obtained from Hungarian Statistical Office personal income tax filings (binned totals and taxpayers). Additional socio-economic and geographic data include population density, unemployment, foreign direct investment (FDI), distance to border, and spatial datasets from CORINE Land Cover and OpenStreetMap (OSM), plus POI/amenity data. Timeline: social network fragmentation F_i is measured from ties up to end-2011; income inequality is measured in 2011 and 2016; geographic features are observed in 2017 (assumed slow-changing). Measures: - Income inequality: Gini index G_it computed from binned income distributions at the town level (2011, 2016). - Social network fragmentation: Using within-town subgraphs (only ties among users residing in the same town), communities are detected via the Louvain algorithm to maximize modularity Q_i. To account for dependence on size/density, Q_i is scaled by theoretical Q_max, defining fragmentation F_i = Q_i / Q_max. - Urban topology indicators (scaled by residential area S_t for comparability across towns): 1) Average Distance from Center (ADC): Randomly sample points across residential polygons (count proportional to polygon area). Compute town center of gravity and average distance of points to center, scaled by town area: ADC = Σ_p D_p / S_t. Higher ADC indicates more spatially dispersed neighborhoods. 2) Segregation by Physical Barriers (SPB): Adapt Railroad Division Index. Cut residential area polygons by rivers, major roads, and railways; compute dispersion of residential area across disconnected components: SPB = 1 - Σ_i (S_i / S_t)^2, where S_i are component areas. Higher SPB indicates more division by barriers. 3) Spatial Concentration of Amenities (SCA): Identify POI clusters via DBSCAN (500 m radius). Compute entropy-based concentration measure: SCA = - Σ_c (p_c log p_c) / S_t, where p_c is the number of POIs in cluster c. Higher SCA reflects amenities concentrated in fewer clusters. Empirical strategy: - Descriptive patterns: Correlate Gini (2011) with F_i and illustrate with sampled town networks. - Dynamics via OLS: Regress G_{i,2016} on baseline inequality, fragmentation, and their interaction: G_{i,2016} = α G_{i,2011} + β F_i + γ (G_{i,2011} × F_i) + Z_{i,2011} + e, where Z includes controls (population density, fraction of iWiW users). The marginal effect of F_i is β + γ G_{i,2011}. - Two-Stage Least Squares (2SLS): First stage predicts fragmentation from urban topology indicators and the iWiW penetration control: F_i = δ + γ IV_i + δ N_i + e_i, where IV_i ∈ {ADC, SCA, SPB}. Second stage regresses G_{i,2016} on predicted fragmentation f_i and controls X_i with county fixed effects: G_{i,2016} = α + β f_i + β_2 X_i + χ + ε_i, where X_i includes FDI (level and change), unemployment, population density, and distance to nearest border. Instrument validity and robustness: Instruments observed after the outcome but assumed slow-changing and exogenous; first-stage F-tests, Wu-Hausman and Sargan tests reported. Additional robustness includes PCA composite of topology indicators, machine learning comparisons with alternative segregation proxies (ethnic, religious, educational, political heterogeneity), and size-restricted samples.

• Cross-sectional association: Town-level income inequality (Gini, 2011) is positively correlated with social network fragmentation F_i. For towns >15,000 inhabitants (n = 91), Pearson correlation ≈ 0.44 with linear fit G_{t,2011} = 0.36 + 0.37 F_t (R^2 = 0.198). Across all towns, Pearson’s correlation is 0.29. - Inequality dynamics: Gini increased on average from 0.474 (2011) to 0.484 (2016), significant by Mann–Whitney U-test (p < 0.001). G_{2011} and G_{2016} are highly correlated (≈ 0.9). In OLS with interaction, the effect of fragmentation on future inequality is stronger at higher baseline inequality: the marginal effect β + γ G_{i,2011} becomes significant around the mean G_{i,2011} and grows with G_{i,2011}. A 1 SD increase in F_i associates with a 0.1 SD increase in G_{i,2016} at mean G_{i,2011}, and 0.4 SD at maximum G_{i,2011}. - Urban topology → fragmentation (first stage 2SLS, standardized): Positive and significant relationships between F_i and each indicator: ADC 0.091* (0.045), SCA 0.110** (0.046), SPB 0.168*** (0.044). First-stage F-tests: 47.1, 48.1, 53.3 respectively, indicating strong instruments. - Fragmentation (instrumented) → inequality (second stage 2SLS, standardized): Estimated effect of predicted fragmentation on G_{i,2016}: 0.408* (0.153) using ADC as IV; 0.533** (0.146) using SCA; 0.288* (0.138) using SPB. Controls: Population density tends to be associated with lower inequality (significant in SCA model: -0.118* (0.052)); greater distance to border is associated with lower inequality (-0.231 to -0.254, all p < 0.01), implying towns closer to borders have higher inequality. County fixed effects included. Model fit R^2 ≈ 0.19–0.25. - Validity and robustness: Wu–Hausman and Sargan tests suggest instruments are plausibly exogenous and 2SLS outperforms OLS in most cases (except SCA). Robustness checks (size-restricted samples, comprehensive models including all controls) show SPB is the most robust instrument and fragmentation remains a significant predictor of inequality. - Additional insights: Amenities being spatially concentrated (higher SCA) is associated with higher network fragmentation; physical barriers (SPB) and larger intra-urban distances (ADC) also associate with higher fragmentation. A PCA-based composite topology index and the three individual indicators outperform alternative segregation proxies in predicting fragmentation.

The findings indicate that fragmented town-level social networks are linked to higher income inequality and that fragmentation interacts with existing inequality to predict future increases in inequality. Urban topology—greater average distances within towns, spatial concentration of amenities, and divisions by physical barriers—relates to more fragmented social networks. The geographic division of urban space appears to manifest in social network segregation, which coincides with and may help explain higher inequality. Policy implications follow: while some features like major physical barriers are difficult to change, urban planning decisions concerning the placement of infrastructure and the distribution of amenities can influence social mixing, access to interaction loci, and thus potentially the trajectory of inequality. Ensuring access across neighborhoods and avoiding excessive concentration of amenities may reduce fragmentation. Although causality cannot be definitively established, the instrumental-variable framework and falsification tests reduce concerns about omitted variables and support the interpretation that urban form is associated with social network fragmentation, which in turn is associated with inequality.

This study connects urban topology, social network structure, and income inequality using large-scale OSN data and town-level socio-economic records. It shows that towns with greater network fragmentation have higher and rising inequality, and that fragmentation is strongly related to urban form—larger intra-urban distances, concentrated amenities, and physical barriers. By leveraging 2SLS with geography-based instruments, the analysis provides evidence that urban topology is an important predictor of fragmentation, which is linked to inequality. The work suggests that urban planning that improves connectivity between neighborhoods and distributes amenities more evenly may help mitigate rising inequalities. Future research should extend this framework to other countries and platforms, explore causal mechanisms via natural experiments or longitudinal changes in urban infrastructure, and assess interventions targeting urban connectivity and amenity distribution on network structure and economic outcomes.

Causal claims cannot be proven; results rely on observational data and instrumental variables whose exclusion restrictions, while supported by tests and prior literature, cannot be guaranteed. Some instruments, particularly SCA, may suffer from reverse causality since amenity locations respond to demand and purchasing power. Geographic indicators are measured in 2017, after the inequality outcomes, though assumed to change slowly. Social network data originate from a single OSN (iWiW) with potential selection and representativeness biases and varying penetration across towns; the fraction of users is controlled for but residual bias may remain. The analysis excludes Budapest, limiting generalizability to very large cities. Unobserved confounders and complex long-run neighborhood dynamics may still influence both social networks and inequality. Raw OSN data are not publicly available due to NDA, potentially limiting reproducibility (though town-level aggregates and code for topology indicators are shared).