Disaggregating Multifaceted Destination Effects on Residential Mobility by Regional and Age Groups in South Korea

Abstract

Residential mobility is not only a demographic process but also a mechanism that reshapes urban form, economic vitality, and spatial inequality. In South Korea, where rapid population decline and stark regional disparities pose urgent planning challenges, analyzing the determinants of residential in-migration provides critical insights into how cities and regions adapt to these demographic shifts. This study addresses the questions of why individuals relocate and how migration drivers vary across regional typologies and age cohorts. Using a Negative Binomial Regression framework applied to spatially disaggregated migration data, this study identifies several key patterns. Housing prices, population density, and network centrality consistently act as strong and positive predictors of in-migration across regions and cohorts. Even shrinking cities retain attractiveness through density, likely reflecting service accessibility and agglomeration benefits. Employment opportunities, school proximity, and road network density play crucial roles in peripheral regions. Katz centrality strongly shapes decisions among younger populations (≤39), while older adults (65+) prefer areas with lower economic intensity and better access to public transportation. These findings advance theoretical understandings of residential mobility and offer policy-relevant insights for age-sensitive and regionally differentiated urban planning.

1. Introduction

In the context of intensifying spatial inequality, population aging, and regional decline, understanding the determinants of residential mobility (e.g., residential in-migration) has emerged as a pressing concern for urban and regional planning [1]. Residential in-migration is not only a demographic process but also a critical driver of spatial restructuring, influencing patterns of urban growth, service demand, and territorial competitiveness [2]. In South Korea, migration trends have amplified socio-spatial polarization, as population and economic activities increasingly concentrate in the Seoul Metropolitan Area while many smaller cities and peripheral localities face sustained demographic shrinkage and economic marginalization [3,4,5,6]. Such uneven flows of people contribute to reinforcing spatial inequality, exacerbating service imbalances, and undermining the long-term viability of lagging regions [5]. Moreover, in-migration reshapes land-use dynamics and neighborhood vitality, thereby influencing broader processes of urban sustainability, resilience, and regional equity [7]. Despite its importance, the mechanisms that drive residential in-migration remain only partially understood; for instance, previous studies have identified that the relocation decisions are inherently complex, shaped by an interplay of urban, economic, infrastructural, and institutional factors [8]. For instance, existing scholarship has examined sociodemographic attributes [9], the impact of life-course events [10], housing market [11], simultaneous home and workplace relocation models [12], and built environment effects of destination [13]. Yet, the studies often overlook how migration determinants operate differently across regional typologies and demographic cohorts, particularly within highly urbanized and spatially stratified contexts such as South Korea.

Therefore, the aim of this study is to disentangle the multifaceted destination effects shaping residential in-migration in South Korea. Specifically, the study pursues three objectives: (1) to assess how diverse factors, such as urban form, vitality, accessibility, and network centrality, influence residential inflows; (2) to examine how these determinants vary across different regional typologies; and (3) to evaluate how migration drivers differ by age cohorts, reflecting life-course heterogeneity. Accordingly, we analyze the determinants of residential in-migration across four analytically defined regional regimes: Seoul Metropolitan Area (SMA), Other Metropolitan Areas (OMA), Urban Areas (URB), and Shrinking Areas (SHR). The study further incorporates a demographically disaggregated lens, recognizing that migration propensities are strongly conditioned by life-course stages [14]: those aged 19 and below, 20–39, 40–64, and 65 and above. To examine the multifactorial influences on residential mobility, the study incorporates variables capturing urbanization intensity (e.g., population density, nighttime light), urban vitality (e.g., floating population), network centrality (e.g., degree and Katz centrality), as well as housing prices, employment accessibility, and urban form attributes such as POI diversity and road network density. A Negative Binomial Regression (NBR) model is employed to estimate the relationship between these variables and in-migration patterns, accounting for the over-dispersed nature of in-migration count data.

2. Literature Review

2.1. Theoretical Background

Urban morphology and residential mobility are deeply intertwined [15]. The historical growth of cities has consistently reflected the cumulative outcome of individual and household relocation choices [16]. Industrialization and improved transportation, for example, enabled outward suburban expansion, while contemporary polycentric patterns emerged as populations sought greater accessibility and diverse amenities beyond traditional cores [17]. Early urban ecological theories illustrate this connection [18]. Burgess’ concentric zone model, Hoyt’s sector model, and Harris and Ullman’s multiple nuclei model each conceptualized how residential mobility produces distinctive morphological patterns of urban growth [19,20,21]. These frameworks underscore that the changing geography of residential settlement is a principal mechanism of spatial reorganization.

Subsequent theoretical perspectives expanded the understanding of why and how people move within and between urban areas. Residential Location Theory highlights rational trade-offs between housing costs, accessibility, and neighborhood quality, explaining the locational decisions that underpin urban expansion or densification [22]. Push–Pull Theory emphasizes the attraction of certain destinations and repulsion of others, linking migration directly to structural inequalities that create uneven patterns of regional development [23]. The Life-Course Perspective further adds that residential preferences vary across age cohorts, thereby introducing temporal differentiation into the spatial evolution of cities [24]. That is, these theories reveal that residential mobility is simultaneously a behavioral process and creates a morphological force.

2.2. Empirical Studies

An extensive body of literature has investigated the determinants of residential mobility [25], with studies ranging from individual decision-making processes [26] to broader structural drivers of the population movement [27]. For instance, Borgers and Timmermans [28] suggested three main categories of factors of residential locations: (1) the residence itself (e.g., dwelling types, costs, and type of neighborhood), (2) transportation infrastructures (e.g., frequency of bus services, availability of railway station, and parking availability), and (3) travel time to workplace. Likewise, Masoumi et al. [29] identified diverse determinants of residential mobility, including socio-economic factors of individuals and the built environment of neighborhoods. These factors can be categorized into micro-level behavioral and macro-level structural perspectives. First, micro-scale studies emphasize household decision-making, highlighting that relocation is often the result of joint deliberation among household members, influenced by subjective factors such as well-being, neighborhood cohesion, and social capital [30,31,32,33]. These perspectives also recognize the interdependence between residential and workplace locations [34,35,36], pointing to employment changes, travel cost, and commuting distance as key factors shaping residential choices [8,37].

Another significant body of literature has adopted a macro-level perspective to explore economic factors and built environment variables of neighborhoods [16,27]. Residential mobility is frequently framed as a response to interregional disparities in employment opportunities, housing affordability, and neighborhood quality [14]. For instance, much of the literature has framed residential mobility as a rational response to interregional disparities in employment opportunities [38] and housing affordability [11]. Several empirical works have paid attention to the spatial characteristics of destinations themselves [39,40], recognizing that the spatial configuration and functional attributes of built environments significantly shape their attractiveness to potential movers [41]. Scholars have identified the significance of land-use diversity, population density, street network connectivity, and proximity to services in shaping perceived quality of life and locational utility [42]. For example, Zhang and Luo [13] analyzed intra-urban migration in Sydney and found that population inflows were positively associated with destinations offering detached housing, adequate housing supply, and a low concentration of non-private dwellings.

Further empirical studies have emphasized the role of demographic selectivity, particularly the life-course perspective in explaining residential preferences and mobility propensities [43,44,45]. Age, in particular, has been shown to influence spatial behavior, with different cohorts displaying varying locational needs; for example, young adults pursuing employment and lifestyle amenities, and older adults seeking accessibility to healthcare and affordability [10,46,47]. Moreover, the growing body of literature addresses the regional disparities between metropolitan growth hubs and shrinking peripheries, a pattern increasingly evident in South Korea. While core metropolitan areas such as Seoul attract a disproportionate share of population and investment, many regional cities face demographic contraction, economic decline, and infrastructure erosion [39,48]. Despite the recognition of these trends, empirical models of migration often fail to account for spatial heterogeneity, frequently treating urban space as uniform or focusing exclusively on metropolitan areas.

In the South Korean context, two main strands of research can be discerned. The first focuses on the morphological transformation of cities as reflected in patterns of residential mobility. This body of work has consistently shown that SMA attracts a disproportionate share of population growth, while many provincial cities face persistent depopulation, economic stagnation, and service contraction [3,49]. Recent studies further underscore the demographic dimension of in-migration, with younger cohorts gravitating toward Seoul and other large metropolitan centers in pursuit of education, employment, and lifestyle opportunities [50], whereas older populations display stronger attachments to suburban or rural areas [51]. The second strand of literature examines the determinants of residential mobility preferences, identifying key drivers such as socio-demographic characteristics, accessibility to infrastructure, social networks, environmental perceptions, and health conditions [27,52,53,54].

2.3. Research Gaps

Despite the growing body of work on residential mobility, several critical gaps remain. First, much of the literature in the South Korea context has emphasized either individual-level determinants (e.g., life-course factors, household decisions) or macroeconomic conditions such as employment opportunities and housing affordability. While valuable, these perspectives often neglect how critical attributes, such as urban vitality, urbanization rate, and network centrality, shape inter-urban migration flows at an aggregate level. Existing South Korean studies have tended to focus on metropolitan in-migration, particularly into Seoul and its surrounding region, without fully accounting for how diverse regional environments influence population inflows or decline. Furthermore, migration research in South Korea has often overlooked regional and demographic variation, despite evidence that regional typologies and age cohorts respond differently to locational opportunities and constraints.

3. Materials and Methods

3.1. Study Area

This study focuses on South Korea, as shown in Figure 1, for investigating spatial determinants of residential in-migration. South Korea represents a theoretically robust and empirically rich context due to its high urbanization levels, pronounced metropolitan primacy, and increasing regional inequality [55]. The nation has undergone rapid urban expansion over the past several decades, with growth heavily concentrated in SMA, a megaregion encompassing the capital Seoul, Incheon, and Gyeonggi Province, which accounts for nearly half of the national population and a disproportionately large share of GDP [49]. In contrast, many mid-sized and peripheral cities have experienced sustained population decline, largely driven by demographic aging, youth outmigration, and uneven economic development [56].

To capture the heterogeneity of these urbanization trajectories and the differentiated attractiveness of localities, this study classifies the national territory into four analytically distinct regional typologies: SMA, OMA, URB, and SHR (see Figure 1). This categorization is grounded in both administrative definitions and functional spatial differentiation, allowing for comparative analysis that reflects institutional, demographic, and economic realities. The SMA and OMA classifications follow official administrative boundaries: SMA includes Seoul, Incheon, and municipalities within Gyeonggi Province. OMA encompasses six additional metropolitan cities, such as Busan, Daegu, Daejeon, and Gwangju. These two groups are separated analytically due to significant differences in population density, economic capacity, urban scale, and infrastructural systems [57].

Additionally, the fourth category, SHR, is based on official designations by Ministry of the Interior and Safety (MOIS). These municipalities are identified as shrinking cities using a composite index that includes indicators such as annual average population growth rate, population density, net youth migration, daytime population levels, and fiscal self-reliance ratio. Lastly, the third category, URB, comprises all non-metropolitan municipalities classified as cities that are neither part of SMA and OMA, nor designated as shrinking cities by MOIS. By adopting this four-tiered spatial typology, the study ensures a spatially disaggregated and policy-relevant framework for evaluating migration determinants across differential urban systems.

3.2. Variables

3.2.1. Dependent Variables

The dependent variable in this study is the annual number of residential in-migrants, measured at the Eup/Myeon/Dong (EMD) level for the year of 2022 (see

Table 1. Description and Descriptive Statistics of Dependent Variables.

Variable	Description	Mean	St. Dev	Source
Total	Total Count of in-migrations at EMDs	2314	3580	MOIS
Four Regional Typologies
SMA	Count of in-migrants at EMDs in SMA	4042	4496	MOIS
OMA	Count of in-migrants at EMDs in OMA	1897	1342	MOIS
URB	Count of in-migrants at EMDs in URB	2168	3800	MOIS
SHR	Count of in-migrants at EMDs in SHR	386	829	MOIS
Four Age Cohorts
Aged <19	Count of in-migrants aged 19 years and younger	340	620	MOIS
Aged 20–39	Count of in-migrants aged 20 to 39	1052	1750	MOIS
Aged 40–64	Count of in-migrants aged 40 to 64	606	922	MOIS
Aged >65	Count of in-migrants aged 65 and above	317	413	MOIS

). This measure captures legally documented residential relocations, thereby excluding short-term or informal moves. The use of absolute migration counts follows prior empirical research in migration studies that emphasizes migration flows as indicators of urban attractiveness, regional competitiveness, and demographic restructuring [58,59].

To capture spatial heterogeneity, migration counts are further disaggregated by four regional typologies (Figure 1), which correspond to distinct structural regimes of the South Korean urban system: SMA, OMA, URB, and SHR. This typology builds on comparative urbanization frameworks that emphasize hierarchy, functional embeddedness, and morphological differentiation in shaping mobility incentives. In addition, the dependent variable is stratified into four age cohorts (≤19, 20–39, 40–64, and 65+ years), reflecting the importance of life-course dynamics in migration research. As theorized by the Push–Pull Model [23], residential mobility is jointly shaped by locational attributes and demographic positioning.

3.2.2. Independent Variables

This study incorporates a wide array of 18 independent variables to capture the multifactorial drivers of residential in-migration (see Table 1 and Figure 2). The selection of these variables was guided by both theoretical frameworks of residential mobility and empirical evidence from prior studies [13,28,47]. The variables were chosen to reflect not only the individual determinants of migration but also the structural, spatial, and economic attributes of place, ensuring alignment with the research objective of disentangling the role of diverse urban characteristics in shaping in-migration patterns. The independent variables are organized into six thematic domains: (1) urbanization intensity, (2) urban vitality, (3) accessibility, (4) network centrality, (5) economic opportunity, and (6) urban form (see Table 2). All variables are operationalized at the EMD level and reference conditions in 2022. Data were primarily sourced from official statistics, including the Korean Statistical Information Service (KOSIS), the Ministry of Land, Infrastructure and Transport (MOLIT), the Ministry of Education, and night-time light (NTL) data from satellite imagery.

First, urbanization intensity was measured through population density and NTL intensity. These proxies, derived from census data and satellite-based luminosity, capture spatial concentration, agglomeration economies, and urban scale, which are fundamental indicators in migration theory for assessing development level and functional capacity [60,61]. Second, urban vitality was operationalized using floating population (FP) inflow and outflow densities, derived from mobile phone-based mobility datasets. These measures represent the temporal dynamics of human activity, providing a real-time proxy of neighborhood vibrancy and social interaction opportunities [62]. Third, accessibility indicators included distance to KTX stations, bus terminals, elementary schools, and metropolitan boundaries, calculated using GIS-based network analysis. These metrics reflect locational advantage, commuting costs, and proximity to essential services, aligning with relocation decisions that emphasize residential convenience, work access, and family needs [63].

Fourth, network centrality measures were developed using population mobility networks constructed from inter-EMD movement data. Degree centrality and Katz centrality, derived through graph-theoretical approaches [64], capture the embeddedness of each location within broader migration systems. Areas with higher connectivity are hypothesized to be more visible, accessible, and attractive destinations for settlement [65]. Fifth, economic opportunity variables—including housing prices, business density, employment opportunities, and Gross Regional Domestic Product (GRDP)—were compiled from MOLIT, KOSIS, and regional accounts data. These indicators serve as proxies for market signals, labor accessibility, and overall economic attractiveness [66]. Finally, urban form was measured through Point-of-Interest (POI) diversity and road coverage rate, obtained from national GIS datasets and commercial mapping services. These variables reflect land-use heterogeneity, infrastructure density, and morphological complexity, all of which influence perceptions of livability and locational utility [67].

Table 2. Description and Descriptive Statistics of Independent Variables.

Variable	Description	Mean	St. Dev	Source
Urbanization
Population Density	Log-transformed number of registered residents per square kilometer; reflects residential intensity and built-up saturation in an area	7.1	2.6	NGII
NTL Intensify	Log-transformed nighttime Light Intensity, extracted from VIIRS satellite data; used as a proxy for economic activity, infrastructure concentration, and urban development	3.6	0.7	Li et al. [68]
Urban Vitality
FP Inflow Density	Log-transformed floating Population Inflow Density, measuring the inflow of non-resident mobile individuals (e.g., commuters, visitors) per unit area; serves as a proxy for urban dynamism	6.9	2.5	KT
FP Outflow Density	Log-transformed floating Population Outflow Density, capturing the density of outgoing mobile individuals, often reflecting commuting patterns and urban permeability	6.9	2.4	KT
Transportation Accessibility
Distance to KTX	Log-transformed road network distance (in km) from the administrative unit centroid to the nearest Korea Train Express (KTX) station	9.2	1.0	PDP
Distance to Bus	Log-transformed road network distance to the nearest intercity or regional bus terminal	5.7	1.1	PDP
Distance to Elementary	Log-transformed road network distance to the nearest elementary school used as a proxy for accessibility to basic educational infrastructure	6.6	1.0	PDP
Distance to Metropolitan	Log-transformed road network distance to the nearest metropolitan city boundary, used to assess spatial peripherality and proximity to urban cores	10.1	0.9	PDP
Network Centrality
Degree Centrality	Log-transformed network measure representing the number of direct connections (edges) an administrative unit has to others; reflects local connectivity within the mobility network.	3.6	0.6	KT
Katz Centrality	Log-transformed measure of global centrality in the mobility network, taking into account both direct and indirect connections with decay factors; reflects embeddedness within broader regional flows.	0.6	0.4	KT
Economic Factors
Housing Price	Log-transformed average real estate transaction price or publicly reported housing value in the area; used to capture cost-of-living and market demand	19.0	1.1	NGII
Business Density	Log-transformed number of registered businesses per square kilometer; indicates economic clustering, retail/services availability, and job concentration	4.9	2.4	NGII
Employment Opportunity	Log-transformed number of jobs per working-age population; measures access to labor market and regional economic pull	6.2	2.5	NGII
GRDP	Log-transformed gross Regional Domestic Product, measured per capita or total, indicating economic productivity and resource availability at the municipal level	9.1	3.4	URIS
Urban Form
POI Diversity	Log-transformed index of Point-of-Interest (POI) functional mix, calculated using the number of different types of POIs; reflect land-use diversity	1.6	0.9	NGII
Road Rate	Log-transformed ratio of road area to total land area (%); captures infrastructure density and vehicular accessibility within a locality	1.3	0.9	URIS
Control Factors
Area	Log-transformed total land area (in square kilometers) of the EMD administrative unit; used to normalize spatial measures and control for unit size variability	2.4	1.5	NGII
DEM	Log-transformed digital Elevation Model, representing the mean elevation (in meters); included to account for terrain effects on accessibility and development constraints	4.0	1.2	VW

Source: National Geographic Information Institute (NGII), V-World (VW), the Public Data Portal (PDP), and the Urban Regeneration Information System (URIS), KT Mobile Movement Data (KT).

Table 2. Description and Descriptive Statistics of Independent Variables.

Variable	Description	Mean	St. Dev	Source
Urbanization
Population Density	Log-transformed number of registered residents per square kilometer; reflects residential intensity and built-up saturation in an area	7.1	2.6	NGII
NTL Intensify	Log-transformed nighttime Light Intensity, extracted from VIIRS satellite data; used as a proxy for economic activity, infrastructure concentration, and urban development	3.6	0.7	Li et al. [68]
Urban Vitality
FP Inflow Density	Log-transformed floating Population Inflow Density, measuring the inflow of non-resident mobile individuals (e.g., commuters, visitors) per unit area; serves as a proxy for urban dynamism	6.9	2.5	KT
FP Outflow Density	Log-transformed floating Population Outflow Density, capturing the density of outgoing mobile individuals, often reflecting commuting patterns and urban permeability	6.9	2.4	KT
Transportation Accessibility
Distance to KTX	Log-transformed road network distance (in km) from the administrative unit centroid to the nearest Korea Train Express (KTX) station	9.2	1.0	PDP
Distance to Bus	Log-transformed road network distance to the nearest intercity or regional bus terminal	5.7	1.1	PDP
Distance to Elementary	Log-transformed road network distance to the nearest elementary school used as a proxy for accessibility to basic educational infrastructure	6.6	1.0	PDP
Distance to Metropolitan	Log-transformed road network distance to the nearest metropolitan city boundary, used to assess spatial peripherality and proximity to urban cores	10.1	0.9	PDP
Network Centrality
Degree Centrality	Log-transformed network measure representing the number of direct connections (edges) an administrative unit has to others; reflects local connectivity within the mobility network.	3.6	0.6	KT
Katz Centrality	Log-transformed measure of global centrality in the mobility network, taking into account both direct and indirect connections with decay factors; reflects embeddedness within broader regional flows.	0.6	0.4	KT
Economic Factors
Housing Price	Log-transformed average real estate transaction price or publicly reported housing value in the area; used to capture cost-of-living and market demand	19.0	1.1	NGII
Business Density	Log-transformed number of registered businesses per square kilometer; indicates economic clustering, retail/services availability, and job concentration	4.9	2.4	NGII
Employment Opportunity	Log-transformed number of jobs per working-age population; measures access to labor market and regional economic pull	6.2	2.5	NGII
GRDP	Log-transformed gross Regional Domestic Product, measured per capita or total, indicating economic productivity and resource availability at the municipal level	9.1	3.4	URIS
Urban Form
POI Diversity	Log-transformed index of Point-of-Interest (POI) functional mix, calculated using the number of different types of POIs; reflect land-use diversity	1.6	0.9	NGII
Road Rate	Log-transformed ratio of road area to total land area (%); captures infrastructure density and vehicular accessibility within a locality	1.3	0.9	URIS
Control Factors
Area	Log-transformed total land area (in square kilometers) of the EMD administrative unit; used to normalize spatial measures and control for unit size variability	2.4	1.5	NGII
DEM	Log-transformed digital Elevation Model, representing the mean elevation (in meters); included to account for terrain effects on accessibility and development constraints	4.0	1.2	VW

Source: National Geographic Information Institute (NGII), V-World (VW), the Public Data Portal (PDP), and the Urban Regeneration Information System (URIS), KT Mobile Movement Data (KT).

3.3. Data

This study draws upon spatially referenced secondary datasets compiled from authoritative governmental and administrative sources (see Table 1 and Table 2), with all data standardized to the reference year 2022, the most recent year for which comprehensive inter-EMD migration records are available (see Figure 3). The analysis is conducted at the finest available administrative scale, the EMD unit, enabling high-resolution spatial assessment across the national territory. The core dependent variable, annual residential in-migration counts, is derived from the Resident Registration Population Statistics provided by MOIS. This dataset captures official registrations of population inflows at the EMD level. Furthermore, independent variables capturing aspects of the built environment, accessibility, and urban form are sourced from a range of publicly accessible geospatial platforms, including the National Geographic Information Institute (NGII), V-World (VW), the Public Data Portal (PDP), and the Urban Regeneration Information System (URIS). These platforms provide detailed spatial layers such as land-use, road networks, POIs, and public infrastructure locations. To proxy urbanization intensity and capture real-time human activity, this study incorporates NTL imagery, processed and calibrated according to the method developed by Li et al. [68]. Additionally, both floating population and two key measures of network centrality are calculated using KT Mobile Movement Data (KT), which captures the topological embeddedness and relative centrality of each EMD unit within the national mobility network.

3.4. Method

To empirically examine the relationship between spatial attributes of urban environments and residential in-migration, this study employs a NBR framework in R Studio 2025.05.1+513. The use of NBR is motivated by both the statistical properties of migration data and the specific objectives of this study. Formally, the probability mass function of the NBR model can be expressed as [69]:

$$\mathrm{Pr}\left(Y_i= y_i|u_i,a\right)= {\displaystyle \frac{r\left(y_i+a^{-1}\right)}{r\left(a^{-1}\right)r\left(y_i+1\right)}}{\left({\displaystyle \frac{1}{1+a{u}_i}}\right)}^{a^{-1}}{\left({\displaystyle \frac{a{u}_i}{1+a{u}_i}}\right)}^{y_i}$$

where $Y_i$ is the count of in-migrants for unit $i$, $u_i$ is the expected value of $Y_i$, and $a$ is the dispersion parameter.

The NBR models are stratified to explore the heterogeneity of in-migration determinants, along two key dimensions: (1) regional typologies: SMA, OMA, URB, and SHR, and (2) age cohorts: ≤19 years, 20–39 years, 40–64 years, and 65+ years. The disaggregated models serve three primary analytical objectives: (1) quantifying the influence of built environment and economic factors on in-migration, (2) comparing how these influences differ across regional typologies, and (3) identifying age-specific migration logics in response to place characteristics.

In the migration literature, NBR is particularly useful because in-migration counts are typically skewed, with some areas experiencing disproportionately high flows while others receive very few migrants [70]. The effectiveness of NBR in migration research has been demonstrated in multiple studies [71]. For instance, Vakulenko and Mktchyan [72] explored the drivers of inter-regional migration in Russia and found significant factors, such as socio-economic and geographical factors.

NBR is well-suited for this study. Unlike linear regression, which assumes a continuous outcome with constant variance, count data models account for the integer and non-negative nature of the dependent variable [73]. The Poisson regression model is often the default method for such data; however, Poisson assumes equidispersion (the equality of mean and variance), an assumption that rarely holds in real-world migration contexts where unobserved heterogeneity across places generates overdispersion (variance exceeding the mean). Furthermore, the NBR model addresses this limitation by introducing an additional dispersion parameter, which relaxes the strict equidispersion assumption and allows the variance to exceed the mean [69]. This feature makes NBR particularly effective for capturing heterogeneity across spatial units, where migration inflows may vary considerably depending on urban form, economic conditions, or accessibility. NBR thus prevents biased parameter estimates and underestimated standard errors that would result from using Poisson regression in the presence of overdispersion [74].

4. Results

4.1. Preliminary Analysis Results

Figure 4 presents the preliminary analysis results using hot spot analysis, community detection, and spatial distribution mapping. These serve as an essential foundation for operationalizing the core analytic strategy of this study to model residential in-migration through a spatially disaggregated and demographically sensitive lens. Specifically, Figure 4a illustrates the results of a Getis-Ord Gi* hot spot analysis, applied to total residential in-migration counts at the EMD level. The results reveal statistically significant hot spots (at 90%, 95%, and 99% confidence levels) clustered primarily in SMA and selected high-growth corridors such as Daejeon, Pohang, and Changwon. Figure 4b displays the results of a community detection analysis based on a modularity optimization algorithm applied to the inter-EMD migration flow network. This method identifies 68 empirically derived spatial communities, each representing a functionally cohesive migration sub-region.

Additionally, Figure 4c depicts the spatial distribution of in-migrants aged 19–39, a cohort typically aligned with key life-course transitions such as labor market entry, household formation, and higher residential mobility. The highest concentrations of this group are located in core metropolitan centers, including Seoul, Incheon, Suwon, Daejeon, and Busan. In contrast, Figure 4d maps the spatial distribution of in-migrants aged 65 and above, a cohort associated with post-retirement mobility, aging-in-place preferences, and non-economic migration logics. The spatial pattern for this group is more diffuse, with significant inflows observed in both urban peripheries and rural or semi-rural areas, especially in southern Jeolla, northern Gyeongsang, and coastal Gangwon. Collectively, the results in Figure 4 underscore the necessity of treating residential in-migration as a heterogeneous spatial process, one shaped by both regional development typologies and life-cycle contingent decision-making.

4.2. Determinants of Residential In-Migration by Regional Typologies

Table 3. Negative Binomial Regression Results for the Four Regional Typologies.

Models	NBR Model 1	NBR Model 2	NBR Model 3	NBR Model 4
Dependent Variable	SMA	OMA	URB	SHR
Variables	Coef. (St. Err)	Coef. (St. Err)	Coef. (St. Err)	Coef. (St. Err)
Urbanization
Population Density	0.778 *** (0.050)	0.407 *** (0.069)	1.124 *** (0.061)	1.157 *** (0.058)
NTL Intensify	−0.050 (0.057)	0.032 (0.095)	0.069 (0.051)	−0.051 * (0.030)
Urban Vitality
FP Inflow Density	−1.043 *** (0.127)	0.511 *** (0.161)	0.109 (0.117)	−0.087 (0.081)
FP Outflow Density	0.940 *** (0.135)	−1.014 *** (0.199)	−0.267 * (0.150)	−0.317 *** (0.094)
Transportation Accessibility
Dist to KTX	−0.171 *** (0.030)	0.038 (0.030)	0.003 (0.020)	−0.014 (0.015)
Dist to Bus	−0.076 *** (0.025)	−0.048 * (0.028)	−0.050 ** (0.021)	−0.024 ** (0.011)
Dist to Elementary	−0.062 ** (0.031)	−0.031 (0.034)	0.041 (0.030)	0.033 * (0.017)
Dist to Metropolitan	0.397 *** (0.047)	−0.012 (0.036)	−0.371 *** (0.035)	−0.218 *** (0.025)
Network Centrality
Degree Centrality	0.589 *** (0.128)	0.567 *** (0.120)	0.236 *** (0.070)	0.140 *** (0.037)
Katz Centrality	0.464 *** (0.078)	0.771 *** (0.154)	0.064 (0.098)	0.478 *** (0.096)
Economic Factors
Housing Price	0.304 *** (0.042)	0.225 *** (0.058)	0.076 * (0.043)	0.062 * (0.036)
Business Density	−0.258 *** (0.069)	−0.114 (0.081)	−0.252 *** (0.073)	−0.152 *** (0.046)
Emp Opportunity	−0.098 * (0.058)	0.239 *** (0.068)	0.186 *** (0.057)	0.164 *** (0.036)
GRDP	−0.086 *** (0.010)	0.061 (0.043)	−0.075 *** (0.008)	−0.033 (0.044)
Urban Form
POI Diversity	−0.074 ** (0.035)	−0.106 *** (0.038)	−0.036 (0.030)	−0.024 (0.017)
Road Rate	0.094 ** (0.041)	0.207 *** (0.056)	−0.006 (0.050)	0.194 *** (0.049)
Control Factors
Area	0.304 *** (0.074)	−0.133 (0.140)	0.888 *** (0.089)	0.727 *** (0.053)
DEM	0.084 *** (0.026)	0.007 (0.024)	−0.002 (0.018)	−0.001 (0.011)
Constant	−5.740 *** (1.270)	−0.393 (1.434)	0.488 (0.964)	0.476 (0.694)
Model Performance
Observations	1087	542	871	826
Log Likelihood	−9597.13	−4368.49	−6537.58	−4646.13
theta	2.789 *** (0.114)	4.294 *** (0.252)	3.632 *** (0.168)	12.215 *** (0.634)
AIC	19,232.27	8774.97	13,113.16	9330.25

Significance Level: *** p < 0.001; ** p < 0.01; * p < 0.05.

presents the estimated coefficients from four stratified NBR models, each corresponding to a distinct regional typology. The log-transformation of all independent variables permits elasticity-based interpretation of the results, facilitating direct comparison across spatial contexts. The findings highlight both shared and divergent drivers of residential in-migration, shaped by the spatial logic and development trajectory of each typology.

Urbanization variables reveal consistent and robust effects across all regions. Population density is a statistically significant and positive determinant of in-migration in every regional model, with the largest effect sizes observed in URB (β = 1.124, p < 0.001) and SHR (β = 1.157, p < 0.001). This suggests that even in regions experiencing demographic decline, higher-density settlements retain a comparative advantage due to superior service provision, land-use intensity, or agglomeration economies. In contrast, NTL intensity, a proxy for urban activity and spatial development, emerges as significant only in SMA (β = 0.407, p < 0.01), underscoring the relevance of urban vibrancy primarily in dense metropolitan cores.

With regard to urban vitality, FP inflow density exhibits a strong negative effect in both SMA (β = −1.043, p < 0.001) and URB (β = −1.009, p < 0.01), suggesting that inflow-heavy, high-mobility areas may deter permanent residential settlement due to congestion, instability, or perceived transience. This variable is not statistically significant in SHR, indicating its limited relevance in regions with minimal mobility churn. Conversely, FP outflow density is positively associated with in-migration in SMA (β = 0.940, p < 0.01), potentially reflecting the draw of dynamic districts with high turnover, such as university towns or innovation hubs. In contrast, it is negatively associated with in-migration in URB and SHR, suggesting that outward mobility in these contexts may reflect neighborhood instability or declining local attractiveness.

Transportation accessibility reveals spatially differentiated effects. In SMA, proximity to key transport infrastructure, including KTX stations (β = −0.171, p < 0.001), bus terminals (β = −0.076, p < 0.001), and elementary schools (β = −0.062, p < 0.01), significantly enhances residential in-migration, confirming the importance of multimodal connectivity and service access in dense urban cores. These effects are partially echoed in OMA and URB, albeit with smaller coefficients and lower levels of significance. In contrast, accessibility indicators are largely insignificant in SHR, highlighting the diminished role of transport infrastructure in attracting migrants in structurally declining regions, where limited spatial sorting or destination choice may prevail.

Network centrality metrics, derived from mobile-based inter-EMD movement patterns, are among the most powerful predictors in metropolitan contexts. Both degree centrality and Katz centrality exert strong and statistically significant effects in SMA (β = 0.599 and 0.464, respectively, both p < 0.001) and OMA (β = 0.567 and 0.772, respectively, both p < 0.001), reinforcing the notion that spatial embeddedness within broader mobility systems is a core dimension of urban attractiveness. While these effects are attenuated in URB, they remain positive and significant. Notably, degree centrality retains a positive effect in SHR (β = 0.140, p < 0.05), suggesting that functional connectivity, even in low-growth regions, can partially offset locational disadvantages and bolster migration inflows.

Economic variables yield differentiated effects by region. Housing prices are positively associated with in-migration in SMA and OMA, reflecting the role of real estate values as a proxy for amenity-rich, economically vibrant neighborhoods. The absence of significance in URB and SHR may indicate greater sensitivity to affordability constraints in these regions. Business density is negatively associated with in-migration in SMA (β = −0.253, p < 0.01), possibly reflecting saturation, congestion, or land-use competition in hyper-dense commercial districts. Meanwhile, employment opportunities, capturing job accessibility within commuting thresholds, positively affect in-migration across SMA, OMA, and URB, but are non-significant in SHR, reaffirming the diminished labor market relevance in areas with systemic economic decline. GRDP shows a small but significant effect in SMA and OMA, but not elsewhere.

Finally, urban form variables underscore the varying impact of morphological characteristics. POI diversity, representing functional land-use heterogeneity, has a negative effect in SMA and OMA, suggesting that greater land-use mixing may contribute to congestion or reduce perceived residential stability in high-density settings. Road rate, representing street network density, is positively associated with in-migration only in SHR (β = 0.176, p < 0.05), indicating that physical connectivity may serve as a compensatory mechanism in peripheral or under-resourced locales.

4.3. Determinants of Residential In-Migration by Age Cohort

Table 4. Negative Binomial Regression Results for the Four Age Cohorts.

Models	NBR Model 5	NBR Model 6	NBR Model 7	NBR Model 8
Dependent Variable	Aged < 19	Aged 20–39	Aged 40–64	Aged > 65
Variables	Coef. (St. Err)	Coef. (St. Err)	Coef. (St. Err)	Coef. (St. Err)
Urbanization
Population Density	0.802 *** (0.046)	0.754 *** (0.036)	0.840 *** (0.032)	0.659 *** (0.029)
NTL Intensify	−0.517 *** (0.041)	−0.236 *** (0.032)	0.069 **(0.028)	0.094 *** (0.026)
Urban Vitality
FP Inflow Density	0.217 ** (0.097)	0.247 *** (0.076)	−0.204 *** (0.067)	0.001 (0.062)
FP Outflow Density	−0.538 *** (0.122)	−0.585 *** (0.096)	−0.199 ** (0.085)	−0.282 *** (0.078)
Transportation Accessibility
Dist to KTX	−0.031 * (0.018)	−0.031 ** (0.014)	−0.019 (0.013)	−0.024 ** (0.012)
Dist to Bus	−0.115 *** (0.016)	−0.084 *** (0.013)	−0.104 *** (0.011)	−0.101 *** (0.010)
Dist to Elementary	−0.077 *** (0.023)	−0.019 (0.019)	−0.012 (0.016)	−0.016 (0.015)
Dist to Metropolitan	0.073 *** (0.021)	0.008 (0.016)	0.041 *** (0.015)	0.009 (0.013)
Network Centrality
Degree Centrality	0.693 *** (0.060)	0.566 *** (0.048)	0.399 *** (0.042)	0.311 *** (0.038)
Katz Centrality	0.864 *** (0.071)	0.871 *** (0.057)	0.684 *** (0.050)	0.518 *** (0.045)
Economic Factors
Housing Price	0.282 *** (0.029)	0.202 *** (0.023)	0.168 *** (0.020)	0.091 *** (0.018)
Business Density	−0.565 *** (0.055)	−0.347 *** (0.043)	−0.254 *** (0.038)	−0.056 (0.035)
Emp Opportunity	0.287 *** (0.044)	0.340 *** (0.035)	0.112 *** (0.030)	−0.047 * (0.028)
GRDP	−0.059 *** (0.007)	−0.071 *** (0.005)	−0.079 *** (0.005)	−0.089 *** (0.004)
Urban Form
POI Diversity	−0.055 ** (0.025)	−0.047 ** (0.019)	−0.054 *** (0.017)	−0.024 (0.016)
Road Rate	−0.087 *** (0.030)	−0.048 ** (0.024)	−0.003 (0.021)	0.026 (0.019)
Control Factors
Area	−0.018 (0.070)	0.212 *** (0.055)	0.281 *** (0.049)	0.326 *** (0.045)
DEM	−0.031 ** (0.015)	−0.025 ** (0.012)	−0.026 ** (0.010)	0.001 (0.010)
Constant	−2.173 *** (0.715)	−1.617 *** (0.566)	−1.391 *** (0.498)	0.530(0.456)
Model Performance
Observations	3316	3316	3316	3316
Log Likelihood	−20,094.23	−22,739.69	−21,545.60	−19,770.20
theta	1.727 *** (0.040)	2.556 *** (0.060)	3.196 *** (0.076)	3.857 *** (0.093)
AIC	40,226.46	45,517.39	43,129.20	39,578.41

Significance Level: *** p < 0.001; ** p < 0.01; * p < 0.05.

presents the estimation results from four age-specific NBR models, revealing how built environment, accessibility, economic, and urban form variables differentially shape residential in-migration across the life course. These findings empirically substantiate the life-cycle perspective in migration theory, which posits that locational preferences are mediated by age-contingent needs, constraints, and lifestyle orientations.

Across all age cohorts, population density consistently exerts a statistically significant and positive influence on in-migration. The effect is most pronounced among the 0–18 age group (β = 0.802, p < 0.001) and the 40–64 cohort (β = 0.840, p < 0.001), suggesting a general preference for denser urban environments where educational, employment, and service opportunities are more accessible. Notably, while still significant, the 65+ cohort exhibits the lowest elasticity (β = 0.659, p < 0.001), indicating a more moderate affinity for dense settings among older adults. Conversely, NTL intensity, a commonly used proxy for urban activity and spatial vibrancy, demonstrates a negative and significant effect across all cohorts. This is especially evident among children (β = −0.517, p < 0.001) and young adults (β = −0.236, p < 0.01), suggesting that excessive urban intensity may deter families and individuals seeking affordability, tranquility, or residential stability.

The role of FP inflow density is more nuanced. It is positively associated with in-migration only for the 20–39 cohort (β = 0.247, p < 0.05), indicating a preference among younger adults for dynamic, high-turnover neighborhoods—possibly due to proximity to employment centers, nightlife, or peer networks. In contrast, FP outflow density exhibits strong and consistent negative effects across all cohorts, with the most pronounced disamenity perceived by youth (β = −0.538, p < 0.001) and elderly residents (β = −0.323, p < 0.001). This suggests that population volatility may undermine perceived neighborhood stability and social cohesion, particularly for more risk-averse groups.

Transportation accessibility variables demonstrate distinct life-stage patterns. Proximity to bus stops significantly predicts in-migration across all cohorts, with especially large effects observed among children (β = −0.115, p < 0.001) and working-age adults (40–64: β = −0.104, p < 0.001). Likewise, distance to elementary schools negatively affects migration among the 0–18 cohort (β = −0.077, p < 0.001), reflecting family preferences for educational infrastructure. While distance to metropolitan centers is positively associated with in-migration across all age groups, the effect is most significant for older adults (β = 0.051, p < 0.01), possibly reflecting preferences for peripheral locations that balance accessibility with residential quietude.

Both degree centrality and Katz centrality, derived from mobile movement data, are strong and statistically significant predictors of in-migration across all age cohorts. These findings underscore the critical role of network embeddedness in shaping migration decisions. Notably, youth (β = 0.693, p < 0.001) and young adults (β = 0.566, p < 0.001) exhibit heightened sensitivity to Katz centrality (β = 0.864 and 0.878, respectively), which captures both direct and indirect connectivity within the human mobility network, highlighting the importance of integrated, accessible neighborhoods for younger, more mobile populations.

Among economic variables, housing prices exhibit a consistent and positive association with in-migration across all cohorts, particularly among the ≤19 (β = 0.282, p < 0.001) and 20–39 (β = 0.223, p < 0.01) groups. This aligns with the notion that higher-cost areas often signal amenity-rich environments, even when affordability may be a concern. In contrast, business density is negatively associated with in-migration across all age groups, especially for youth (β = −0.565, p < 0.001) and young adults (β = −0.523, p < 0.001). This may reflect aversion to commercial congestion, land-use conflicts, or noise pollution.

Employment opportunity, as measured by job accessibility, is a strong positive predictor of in-migration for working-age adults (20–64), but statistically insignificant for the elderly cohort, consistent with expectations from post-retirement migration logic. Although GRDP is statistically significant across all groups, the 20–39 cohort (β = −0.059, p < 0.001) displays the most pronounced sensitivity. This inverse relationship may indicate a tradeoff where higher productivity regions are less appealing to younger migrants due to elevated living costs or competitiveness.

Lastly, urban form indicators reveal divergent effects. POI diversity, typically used to measure land-use heterogeneity and functional mix, shows a negative association across all groups, particularly among children (β = −0.055, p < 0.01) and young adults (β = −0.108, p < 0.001). This suggests that excessive land-use mixing may be associated with perceived dis-amenities such as noise, traffic, or safety concerns. In contrast, road network rate is positively significant for older age groups (40–64 and 65+), indicating the importance of physical mobility infrastructure for populations with greater travel autonomy or service reliance.

5. Discussion

5.1. Major Findings

This study examined the determinants of residential in-migration in South Korea through a stratified framework that incorporated both regional typologies and age cohorts. There are several major findings. First, population density emerges as a consistently positive determinant of residential in-migration across all regional typologies and age cohorts. Even in SHR, where demographic decline is prevalent, higher-density locales maintain relative attractiveness, suggesting that basic service availability, agglomeration effects, and urban scale continue to shape mobility decisions. Second, urban vitality variables exhibit complex and sometimes counterintuitive associations. While younger adults are drawn to dynamic, high-turnover districts, families and older residents perceive volatility in floating populations as a disamenity.

Third, accessibility exerts powerful but regionally and demographically specific effects. In SMA, proximity to transport nodes and educational facilities strongly predicts in-migration, while in peripheral regions accessibility is less decisive. Fourth, economic variables such as housing prices and employment accessibility display asymmetric impacts: high housing costs signal amenity-rich urban cores attractive to youth, while employment opportunities are particularly salient for working-age cohorts. Finally, urban form indicators reveal that road density enhances migration inflows in shrinking areas and among older cohorts, while excessive land-use diversity is perceived as destabilizing in metropolitan cores.

These findings corroborate and extend prior research. The centrality of urban density aligns with studies in European and U.S. contexts that highlight the enduring pull of service-rich urban cores despite high living costs [75,76]. The differentiated perception of urban vitality mirrors qualitative studies showing that youth value cultural vibrancy while older adults prioritize neighborhood stability [77]. The significance of transport accessibility in SMA resonates with Saghapour and Moridpour [41], who found that multimodal connectivity is a decisive factor in metropolitan migration decisions.

5.2. Policy Implications

These results offer several policy implications. First, urban densification strategies should be pursued even in shrinking regions, as density provides a buffer against population loss by sustaining service provision and local vitality. Second, policies must recognize the ambivalent role of urban vitality: while cultural vibrancy attracts youth, excessive turnover may deter families and elderly populations. Place-making efforts should therefore balance dynamism with stability. Third, investment in multimodal transport infrastructure remains crucial for metropolitan regions, while in peripheral areas, policies should focus on enhancing functional connectivity rather than simply expanding physical infrastructure. Fourth, housing policies should account for the dual role of high housing costs: they signal urban attractiveness but also risk excluding vulnerable populations, calling for affordability interventions in high-demand centers. Finally, the finding that older adults value road density highlights the need to adapt urban form to aging societies, ensuring accessibility not only through transit but also through pedestrian-friendly and service-rich local environments.

6. Conclusions

This study examined the multifaceted determinants of residential in-migration in South Korea using a spatially disaggregated and demographically stratified modeling. Using detailed administrative migration data at the EMD level and integrating a suite of urban form, accessibility, network centrality, and socio-economic indicators, the study employed NBR models to reveal the spatial heterogeneity of migration drivers across both regional typologies and age cohorts. This study makes several theoretical and empirical contributions. Methodologically, it integrates multi-scalar spatial data, network-based indicators, and demographically stratified regression models to offer a robust framework for modeling intra-national migration. Substantively, it advances the understanding of how urban structure and life-course dynamics interact to shape migration outcomes in an East Asian context, particularly within a spatial system marked by extreme metropolitan primacy and regional decline.

Nonetheless, several limitations should be acknowledged. First, this study relies on 2022 data, limiting the capacity to analyze migration trends over time or assess causality. Second, using registered in-migration counts may exclude unregistered or temporary movers, such as university students, informal migrants, or undocumented relocations. Third, while care was taken to use structural and exogenous variables, potential endogeneity between migration and some spatial features (e.g., infrastructure investment or economic growth) cannot be fully ruled out. Fourth, this study does not include attitudinal or perceptual factors (e.g., neighborhood satisfaction, safety, social capital), which are often significant but harder to quantify in secondary data-driven studies. Finally, regional policies and local planning incentives, such as relocation subsidies or urban regeneration programs, were not incorporated, although they may materially influence migration decisions.