Exploring urban housing disadvantages and economic struggles in Seoul, South Korea

The study addresses how housing and living poverty manifest spatially across Seoul, with the goal of identifying the most socio-economically vulnerable urban areas during and after the COVID-19 pandemic. Motivated by global and local shocks—including the pandemic, climate events, and rising inequality in Korea—the paper highlights the need for targeted, data-informed policies for urban poor populations. Prior work on Korea has often been historical or descriptive and lacked fine-grained identification of vulnerable neighborhoods (dongs). This study proposes a data-driven approach to characterize housing poverty (e.g., Gosiwon and Jiha/Banjiha) and living poverty (e.g., basic livelihood security recipients and older adults living alone with low income) at the dong level (n=426) in Seoul. By extracting latent factors and clustering neighborhoods, the research aims to reveal spatial disparities and inform inclusive urban policy.

The paper synthesizes research that frames poverty as a complex social issue rooted in structural inequalities, with within-country inequality rising despite declines in absolute poverty. COVID-19 intensified vulnerabilities, disproportionately affecting low-income groups through reduced access to health care, education, and secure employment. In Korea, scholarship identifies drivers of poverty such as unequal opportunities, educational disparities, labor market characteristics, weak safety nets, and especially housing inequality that perpetuates wealth gaps intergenerationally. Urban studies emphasize multidimensional and spatial understandings of poverty, including neighborhood (area) effects on health, education, employment, safety, and social capital. Evidence shows urban poor face limited access to medical services, inadequate sanitation, higher exposure to violence, and increased vulnerability to natural hazards. In Seoul specifically, historical development under an Asian developmental state model, policy focus on economic growth (e.g., Gangnam-centric development), and shocks like the 1997 Asian financial crisis intensified socio-spatial stratification. Despite Seoul’s strong global competitiveness, its livability lags, reflecting stark intra-urban disparities. The literature distinguishes housing poverty from living poverty, both central to understanding disadvantage in Seoul.

Design: A two-part approach is used. First, a preliminary survey-based analysis assesses perceived COVID-19 impacts on housing conditions among Seoul residents using the 2022 Koreans’ Happiness Survey and an Ordered Probit model. Second, the core spatial analysis applies Principal Component Analysis (PCA) and clustering to quantify and classify housing and living poverty at the dong level in Seoul.

Study area and units: Seoul’s 25 gus (boroughs) and 426 dongs (smallest administrative units) are analyzed as spatial units.

Variables and data sources (Table 1):

• Housing poverty: Gosiwon (low-cost, micro-rooms; n≈5,582 units) from commercial facility registry via government OpenAPI (data.go.kr); Jiha/Banjiha (habitable basements/semi-basements; n≈202,520) estimated from registered building structure data (open.eais.go.kr).
• Living poverty: Basic living (households receiving basic livelihood security; n=289,518) and Old in poverty (older adults living alone with basic livelihood security or low income; n=124,654) from Seoul government statistics (stat.eseoul.go.kr). Data are aggregated by dong. Log transformation is applied to socio-economic variables prior to clustering so group differences reflect ratios rather than absolute levels.

Preliminary survey analysis: The 2022 Koreans’ Happiness Survey (N=2000 for Seoul in Fig. 1) is examined. An Ordered Probit model evaluates the ordered response: “COVID-19 has negatively affected my housing environment” (1=Not at all to 5=Very much). Models consider age, income, education, sex, marital status, and housing type. Odds ratios are reported for key predictors.

Dimensionality reduction: PCA is run separately on the housing poverty indicators (Gosiwon, Jiha/Banjiha) and the living poverty indicators (Basic living, Old in poverty). For housing poverty, PC1 explains 70.3% and PC2 29.7% of variance. For living poverty, PC1 explains 99.1% and PC2 0.9% of variance. The first principal component from each domain is retained as a summary feature for clustering.

Clustering: K-means++ is applied using the two retained PCs (housing PC1 and living PC1). The optimal number of clusters K is selected via silhouette analysis (scores in [-1,1]). Hierarchical Clustering with Ward linkage is also performed as a robustness check; the dendrogram and results corroborate the optimal K.

Model selection and validation: Silhouette scores are computed for K=2 to K=6. The highest silhouette score is at K=2 (0.426), with lower scores at K=3 (0.334), K=4 (0.356), K=5 (0.345), and K=6 (0.338). Hierarchical clustering also supports K=2 and yields the same cluster membership by dong. Maps visualize spatial distributions of indicators and cluster assignments.

Software and reproducibility: Data extraction and analysis are implemented in Python. Code covering data collection, preprocessing, PCA, clustering, and visualization is provided in Supplementary Tables 6 and 7 with detailed comments. Data supporting findings are available in the article and Supplementary Material.

• COVID-19 perceived impacts (Ordered Probit): Age is positively associated with perceiving more negative housing impacts (odds ratio=1.072), while income is negatively associated (odds ratio=0.893). Older and lower-income respondents more often report adverse effects of COVID-19 on their housing environment. Education and marital status are significant in intermediate models but attenuate in the full model; age and income remain robust across specifications.
• Economic shock and elderly vulnerability: Among 3,239 Seoul respondents in the 2021 survey, 597 reported income decreases due to COVID-19; 38.69% of these were aged 60+. Suicide rates in 2022 rise sharply from age 70 onward (e.g., 70–74: 26.6; 75–79: 37.8; 80–84: 53.5 per 100,000), indicating heightened psychological and social vulnerability among older adults.
• Spatial distributions: Gosiwon are most concentrated in southeastern and eastern Seoul; Jiha/Banjiha are concentrated in southeastern, northern, and western dongs. Living poverty indicators (Basic living, Old in poverty) cluster notably in southwest, north, and northeast Seoul.
• PCA: High correlations within each domain allow strong dimensionality reduction—housing PC1 explains 70.3% of variance; living PC1 explains 99.1%—indicating coherent latent dimensions of housing and living poverty.
• Clustering (K=2 optimal; silhouette=0.426): Seoul’s dongs partition into two clusters that differ markedly in vulnerability. Over 75% of dongs fall into one large, higher-vulnerability cluster. The smaller, lower-vulnerability cluster is concentrated around the Han River and in the southeast. Affluent Gangnam-3 gu dongs are largely absent from the high-vulnerability lists.
• Poverty gap by cluster (means, Cluster 0 vs Cluster 1): Cluster 0 exhibits higher averages—Gosiwon 1.43×, Jiha/Banjiha 1.74×, Basic living 5.37×, Old in poverty 5.15× those of Cluster 1—signaling substantially greater housing and living poverty burdens.
• Overall: More than three-quarters of Seoul’s urban areas are exposed to vulnerabilities associated with poverty, implying that many residents live in or adjacent to vulnerable neighborhoods, even if not poor themselves.

Findings reveal a pronounced socio-spatial bifurcation in Seoul, tracing back to developmental state policies prioritizing economic growth and competitiveness. The large share of dongs in the high-vulnerability cluster demonstrates that poverty and deprivation are pervasive rather than confined to isolated pockets. The spatial pattern—central and southeastern areas around the Han River showing comparatively lower vulnerability—aligns with well-known socio-economic geographies (e.g., Gangnam vs. Gangbuk). COVID-19 disproportionately affected older and lower-income residents, compounding existing disparities; concurrently, suicide rates escalate at older ages, underscoring intertwined economic, social, and psychological risks. These insights support a policy shift from growth-centric strategies toward inclusive urban governance. Seoul’s recent initiatives (e.g., inclusive city index, public rental housing expansions, targeted assistance for disadvantaged groups, and SSIP) are steps in this direction. However, the scale and distribution of vulnerability identified by the clustering suggest broader, integrated interventions are necessary—especially those addressing housing precarity (Gosiwon, Jiha/Banjiha), living poverty, and the specific needs of elderly residents. The data-driven framework—combining PCA, clustering, and spatial visualization—offers actionable intelligence to prioritize neighborhoods for intervention and to monitor progress toward inclusivity.

This study advances the literature by applying a transparent, data-driven methodology to map and classify housing and living poverty in Seoul at high spatial resolution. By reducing multidimensional indicators to latent components and clustering dongs, the analysis uncovers a stark two-cluster structure with over 75% of neighborhoods registering elevated vulnerability. These results underscore the widespread nature of urban poverty and the risks faced by marginalized populations, especially older adults. The approach provides a replicable template for evidence-based urban policy, enabling targeted allocation of resources and ongoing disparity monitoring. Future work should expand data coverage (including unregistered and informal housing and homeless populations), integrate ethnographic and administrative sources, refine estimates for Jiha/Banjiha, and conduct longitudinal analyses to assess policy impacts. Extending the framework to other cities—particularly in Asian developmental contexts—can inform broader discussions on inclusive and sustainable urban development (e.g., SDG1).

• Data coverage: Housing poverty data rely on registered structures and commercial facilities; unregistered or informal housing may be omitted. The spatial distribution of homeless individuals is not included.
• Estimation: Jiha/Banjiha counts are estimated from building registry data, introducing potential error margins.
• Scope: Results are specific to Seoul; direct generalization to other cities should be cautious. Nonetheless, lessons may extend to similar developmental contexts.
• Methodological: PCA/clustering results depend on chosen indicators and transformations; alternative specifications may yield different boundaries. Findings are provisional and benefit from validation with additional datasets and qualitative/ethnographic inquiry.
• Survey analysis: Perceptions of COVID-19 impacts are self-reported and subject to response biases; model results reflect associations, not causal effects.