Segregated by whom? Examining triple neighbourhood disadvantages and activity spaces in Seoul, South Korea

The study investigates how neighbourhood isolation extends beyond residential contexts into non-home activity spaces, asking: (1) whether residential neighbourhood isolation (RND) is reinforced or mitigated by the socioeconomic conditions and social interactions encountered in consumption and commuting activity spaces, and (2) which socioeconomic groups drive the association between isolation in residential and activity spaces—disadvantaged, advantaged, or both. Set in Seoul—a dense, multi-centred city with efficient and affordable public transport yet rising socioeconomic segregation—the research is motivated by the idea that neighbourhood vitality reflects not only residential conditions but also the conditions of places residents visit and places from which visitors originate. The study hypothesises that, given Seoul’s transit accessibility, disadvantaged neighbourhoods face fewer mobility constraints, while advantaged neighbourhoods are more likely to self-seclude, thereby driving neighbourhood isolation through homophilous use of urban spaces.

Extensive literature links neighbourhood socioeconomic conditions to outcomes in employment, health, education, and intergenerational mobility. Traditional research emphasises residential exposure, but recent work argues that everyday activity spaces (work, consumption, leisure) are crucial arenas where segregation and isolation are reproduced. Two opposing scenarios are proposed: activity-space interactions can reinforce residential isolation (e.g., affluent residents frequent advantaged areas, concentrating opportunities), or they can mitigate it (e.g., cross-class interactions in mixed employment or shopping areas). The paper adopts Levy et al.’s (2020) triple neighbourhood (dis)advantage (TND) framework—residential (RND), outdegree (OND: places residents visit), and indegree (IND: places visitors come from)—and highlights that prior research has focused largely on disadvantaged populations, whereas affluent self-seclusion and opportunity hoarding can also strongly drive isolation. In Seoul, given broad transit access, disadvantaged residents may be less mobility-constrained, while affluent groups may actively avoid disadvantaged spaces and reduce access to their own neighbourhoods, potentially intensifying isolation from the top of the socioeconomic hierarchy.

Study area: Seoul, South Korea, a dense, multi-centred global city with efficient and affordable public transit but increasing spatial and economic segregation.

Data sources:

• Korea Credit Bureau (KCB) aggregated neighbourhood (dong) data covering approximately 44.5 million individuals (≈99.3% of those aged ≥19): median household income; household size composition; distribution across nine income brackets; number/share of households with overdue payments >90 days; number/share of earners in large corporations.
• KCB origin–destination (OD) commuting data (home–workplace), ≈3 million commuting records per month, Dec 2019–Dec 2022; covers commuting only.
• BC Card weekly transaction data for consumption activities, 236 spending categories, Oct 2018–Dec 2021; covers a large share of the population (≈36 million individuals).
• Demographics and transit accessibility controls: number of residents, share of men, median age (KOSIS, 2018–2022); subway accessibility (area within 500 m of stations / neighbourhood area) and bus accessibility (number of stops / area), from Seoul Open Data and Rail Portal.

TND measures:

• Residential neighbourhood (dis)advantage (RND): constructed via principal factor analysis (single factor) of four variables—median household income, poverty rate (≤60% of median income), share of earners at large corporations, share of individuals with overdue payments >90 days. Factor scores used as RND (Table S1 shows robust loadings).
• Outdegree (OND) and indegree (IND): measure socioeconomic (dis)advantage in activity-space connections based on mobility ties weighted by visit volumes and the RND of connected neighbourhoods. • For consumption (Card-OND, Card-IND): link BC Card transaction location to cardholder home neighbourhood; compute inter-neighbourhood ties by counts of customer transactions (Oct 2018–Dec 2021). • For commuting (OD-OND, OD-IND): link home to workplace from KCB OD (Dec 2019–Dec 2022); ≈3M records/month.

Formulas:

• OND for neighbourhood i: sum over j≠i of [V_ij(ni) / (V(ni) − V_ii(ni))] × RND_j, where V_ij(ni) is visits from i to j, V(ni) total visits from i, V_ii within-i visits (excluded), and RND_j the destination’s RND.
• IND for neighbourhood i: [Σ_{j≠i} RND_j × V_ij(ni) × P_j] / [Σ_{j≠i} V_ij(ni) × P_j], weighting visiting neighbourhoods’ RND by tie strength and population P_j.

Scope and exclusions:

• Exclude within-neighbourhood visits (focus on extra-local structural connections).
• Limit movement analysis to within Seoul.

Analytic strategy:

• Spatial clustering using local Moran’s I to map RND, Card-OND/IND (2018–2021), OD-OND/IND (2019–2022).
• Pearson correlations among TND measures across years.
• Multivariate regressions of Card-OND, Card-IND, OD-OND, OD-IND on RND with controls (median age, share of men, transit accessibility), estimated for all neighbourhoods and separately by RND quintiles (Q1–Q5) to assess heterogeneity. COVID-19 period effects considered when interpreting 2020–2022 results.

• Spatial clustering: High RND (advantaged) clusters concentrate in southeastern Seoul (e.g., Gangnam/Yeouido areas), while low RND clusters are in southwestern, northern, and some central areas. Card-OND/IND and OD-OND/IND display patterns similar to RND but over broader areas, reflecting homophilous consumption and segmented labour markets.
• Correlations: All pairwise correlations between RND and mobility-based disadvantage measures are positive and moderately strong (min ≈0.535; max ≈0.720). Example values (Table 1): RND–Card-OND up to 0.720 (2020); RND–Card-IND ≈0.613–0.625; RND–OD-IND ≈0.625–0.644 (2019–2021; 0.535 in 2022); RND–OD-OND declines in 2021–2022 (≈0.538 in 2021), likely due to pandemic-induced remote work.
• Heterogeneity by RND quintile (consumption): In the most advantaged neighbourhoods (Q5), RND strongly and consistently predicts higher Card-OND and Card-IND, indicating self-seclusion and homophily among the affluent; e.g., in 2019, >82% of RND Q5 areas fall in Q4–Q5 Card-IND; only 1.2% of RND Q5 fall in Q1 Card-OND and none in Q1 Card-IND. In lower quintiles (Q1–Q3), relationships are weaker/inconsistent, indicating more mixed cross-status interactions—disadvantaged areas are not exclusively connected to disadvantaged activity spaces.
• Heterogeneity by RND quintile (commuting): OD-IND patterns resemble consumption: strong positive and significant association for Q5; mostly insignificant for Q1–Q3 (except Q2 in 2022). For OD-OND, Q1 shows consistently strong positive associations, indicating residents in disadvantaged areas commute to similarly disadvantaged job locations (manufacturing/low-wage clusters), reflecting spatial labour market segmentation. In Q5, the RND–OD-OND association weakens or becomes insignificant in 2021–2022, likely due to remote work among high-wage sectors.
• Controls: Younger median age and better public transport accessibility are associated with more diverse cross-neighbourhood interactions (e.g., negative coefficients on age; transit accessibility often associated with reduced homophily), supporting the idea that mobility infrastructure can reduce some barriers though not eliminating overall homophily.

The findings indicate that activity-space mobility largely reinforces residential neighbourhood isolation in Seoul, aligning with a homophily mechanism. Despite Seoul’s extensive, affordable, and efficient public transit, socioeconomic sorting persists in both consumption and commuting domains, producing TND patterns similar to RND. This likely reflects spatial labour market segmentation: high-wage, stable jobs cluster in advantaged areas, while low-wage, less stable jobs cluster in disadvantaged areas, linking residential status to the socioeconomic nature and location of daily destinations. Importantly, isolation is driven disproportionately by affluent self-seclusion in consumption and inbound/outbound activity networks—residents of advantaged neighbourhoods largely visit and are visited by similarly advantaged areas—while disadvantaged neighbourhoods display broader cross-status interactions, particularly outside of commuting patterns. Pandemic-era reductions in commuting to advantaged office districts among high-wage workers likely weakened some OD-OND associations in 2021–2022 without implying a structural decline in affluent self-seclusion. Overall, the results answer the research questions by showing that: (1) residential isolation is reinforced by conditions and interactions in activity spaces; and (2) the affluent play a leading role in producing and maintaining this isolation through homophilous use of urban space.

This study extends neighbourhood isolation research by integrating residential conditions with everyday activity-space connections using a triple neighbourhood (dis)advantage framework (RND, IND, OND) and large-scale administrative/transaction datasets in Seoul. We show that mobility patterns in consumption and commuting generally mirror residential disparities, reinforcing isolation through neighbourhood homophily. The self-seclusion of advantaged neighbourhoods is a primary driver of isolation, while disadvantaged neighbourhoods often experience more mixed cross-status interactions—though commuting links for the most disadvantaged remain to similarly disadvantaged job areas due to labour market segmentation. Policy implications include promoting voluntary cross-status mixing in activity spaces (e.g., inclusive leisure and service offerings, equitable access to amenities and employment centres) as potentially more expedient than altering entrenched residential patterns. Future research should: (1) differentiate consequences of isolation in residential versus activity spaces; (2) incorporate time-use and duration in locations to refine exposure measures; and (3) compare across cities with varying demographics, employment geographies, segregation levels, and transport systems, potentially leveraging universal indicators to assess generalisability.

• Data coverage and scope: KCB OD data capture commuting only; BC Card transactions reflect consumption activities made via card among cardholders and may miss cash or other payment methods.
• Temporal/spatial scope: Consumption data span Oct 2018–Dec 2021; commuting data Dec 2019–Dec 2022; analysis restricted to movements within Seoul; within-neighbourhood visits excluded.
• Measurement constraints: No data on time spent or duration of exposure in visited neighbourhoods; poverty rate approximated via 60% of median income using available KCB categories; population-adjusted weights used for IND but alternative specifications could yield different sensitivities.
• External shocks: COVID-19 altered commuting patterns (e.g., remote work), affecting OD-based associations in 2021–2022.
• Design: Observational and correlational analyses (spatial clustering, correlations, regressions) limit causal inference; potential unobserved confounding (e.g., preferences, firm sorting) remains.