Current Issue

Theoretical discussions on migration decisions provide an academic foundation for understanding interregional migration. Demographic studies traditionally explain the causes of population movement using push and pull factors, based on the push-pull theory proposed by Lee’s (1966) push-pull theory (Park and Kim, 2020; Laajimi and Gallo, 2022). According to this theory, migration occurs through the interplay of individual and regional characteristics, alongside pushing factors that drive people to leave their current residence and pulling factors that attract them to new areas (Lee, 2009). The theory provides a comprehensive framework for analyzing how regional socioeconomic conditions in both origin and destination areas, combined with individual characteristics, influence migration decisions at the individual and regional levels.

The gravity model is a traditional framework used to explain macro-level population movement based on the distance and differences in characteristics between two areas. In Ravenstein’s (1885) first gravity model, the main factors influencing population mobility were population size and distance between two locations. Subsequently, Lowry (1966) emphasized the role of economic factors as major determinants of population migration. He created a modified version of the conventional gravity model that incorporated regional economic disparities. Lowry (1966) specifically examined how the economic circumstances in two areas affect workers’ migration decisions, proposing that people typically migrate from areas with higher unemployment and lower wages to areas with lower unemployment and higher wages.

The push-pull theory and the gravity model are representative frameworks for explaining population migration at macro level. Both models emphasize the significant impact of the socioeconomic characteristics of the origin and destination areas on individuals’ migration decisions. The gravity model which accounts for migration based on differences in characteristics between two areas and can be extended by incorporating the in- and out-migration factors proposed in the push-pull theory (Ramos, 2016). In this extended gravity model, the push and pull factors influencing individual migration decisions are conceptualized as gravitational forces, represented by socioeconomic variables reflecting each area’s relative attractiveness (Karemera et al., 2000; Poot et al., 2016).

A destination’s attractiveness to migrants is influenced by the area’s various socioeconomic factors. Economic opportunities, educational environments, and housing conditions, in particular, are critical determinants when migrants choose a new area. Traditional economic variables, such as regional average wages, income levels, GRDP (gross regional domestic product) per capita, and employment rates, indicate the economic opportunities available in a destination and significantly affect population inflow (Etzo, 2011; Piras, 2017). High-income areas, which offer abundant economic opportunities, act as a pull factor for migrants. Conversely, individuals from areas with lower economic levels may choose to migrate in search of better opportunities (Cavalleri et al., 2021). Based on this theoretical framework, the present study posits that disparities in regional economic levels are a primary driver of migration and examines how spatial disparities in economic opportunities contribute to interregional migration.

According to the push-pull theory, propelling and attraction factors are perceived differently based on migrants’ individual characteristics and circumstances. This suggests that the same regional conditions may lead to different migration outcomes (Lee, 2008). This provides an empirical basis for considering migration characteristics by life cycle in addition to regional-level characteristics, when studying interregional migration (Lee, 2009; Park and Kim, 2020).

From a life-cycle perspective, migration is a cumulative process shaped by an individual’s life-cycle stages and social needs. Common patterns are observed among individuals with similar life-cycle trajectories (Bernard et al., 2014; Cavalleri et al., 2021; Lee and Choi, 2023). In South Korea, for example, migration activity peaks during the 20s, which corresponds with the transition into higher education and employment. This activity then gradually declines as individuals approach retirement age (Lee and Kim, 2020). During the young adult period (ages 20-29), migration is driven primarily by the pursuit of occupational success and personal growth opportunities (Hong and Yoo, 2012; Rindfuss, 1991). During the family formation period (ages 30-39), residential choices are largely influenced by child-rearing and family stability considerations. Migration decisions at this stage are shaped by a combination of factors, including opportunities for career advancement, educational quality, and housing conditions (Mulder and Hooimeijer, 1999; Choi et al., 2010). Similarly, during the child-education period (ages 40-49), migration patterns reflect a focus on improving the educational environment for children. Migration during this stage involves strategic decisions aimed at enhancing the overall quality of life for the family while balancing job stability and educational needs (Hong and Yoo, 2012). During the middle-aged period (ages 50-64), migration tends to align with preparations for life after retirement. Key motivations for migration at this stage include career transitions, homeownership, and retirement planning (Choi et al., 2010). Other important factors influencing migration decisions during this period also include housing quality, lower living costs, and access to community welfare services (de Jong, 2022).

A comprehensive and systematic analysis of migration throughout the life cycle is imperative because the primary drivers of migration vary significantly at each stage. Migration is a complex and dynamic phenomenon that cannot be fully explained by economic factors or regional characteristics alone. This study aims to gain a more accurate understanding by comprehensively analyzing the evolving needs and drivers of migration across the different stages of an individual’s life cycle.

The gravity model is a traditional framework for explaining population migration. It focuses on the relationship between the areas of origin and destination. It provides a simplified method for modeling the intricate mechanisms that underlie migration (Poot et al., 2016). Several studies on interregional migration have used the gravity model, considering socioeconomic disparities and physical distances between regions as the main factors influencing migration (Etzo, 2011; Wajdi et al., 2017; Cavalleri et al., 2021; Yoon, 2015; Lee and Won, 2017). However, the gravity model has a significant limitation: it only considers the relative differences between the origin and destination areas, ignoring the potential influence of other areas. In reality, migration decisions are not solely determined by the relative differences between two specific areas. They are also influenced by the relative differences and similarities of other potential destinations. Consequently, failing to account for the spatial relationships among alternative areas can lead to biased estimates of migration determinants (Bertoli and Moraga, 2013).

To overcome the limitations of the gravity model and more clearly illustrate spatial relationships among different regions, some migration studies have adopted spatial econometric models that presume spatial dependence between neighboring regions (LeSage and Pace, 2008; Pu et al., 2019). These studies examine the impact of spatial dependence on interregional migration by constructing spatial weight matrices that capture endogenous interactions with neighboring areas. However, interregional migration is influenced not only by spatial interactions based on physical distance and proximity, but also by the socioeconomic relationships between areas. Indicators of physical proximity do not adequately capture these factors. Several studies have addressed this limitation by using spatial weight matrices that incorporate socioeconomic relationships, such as technological proximity (Parent and LeSage, 2008), economic proximity (Boarnet, 1998), and residential substitutability (Rincke, 2010). Furthermore, it is difficult to assume that spatial relationships between regions are symmetrical (Parent and LeSage, 2008). Instead, asymmetrical socioeconomic relationships observed in real-world contexts should be incorporated into the analysis. Consequently, efforts have been made to incorporate asymmetric interactions into spatial weight matrices. For instance, gravitational influence has been applied (Bu et al., 2022; Kim et al., 2025), and hierarchical relationships between areas have been reflected (Jeong, 2024). These asymmetric interactions more accurately capture the complexity of spatial relationships among regions and enable more reliable estimates for explaining the intricate mechanisms underlying interregional migration (Elhorst and Halleck Vega, 2017).

This study examines the gravitational influence of in- and out-migration intensities (which reflect the asymmetry of economic opportunities) and regional substitutability (which reflects the substitutability of residence) on interregional migration. Specifically, we consider the asymmetric nature of spatial interactions when modeling migration patterns to identify how these factors influence migration dynamics.

This study analyzes the factors that influence migration across the different stages of the life cycle. It focuses particularly on how residential substitutability and economic opportunity affect migration decisions. The analysis covers 226 administrative units (si, gun, and gu) in South Korea, excluding island regions, over the period from 2011 to 2020. Notably, this period captures a range of pivotal policy shifts and demographic inflection points, including the implementation and spatial diffusion of the Innovation Cities policy, deregulation in metropolitan development, and the early intensification of youth concentration in the capital region. These developments reflect the onset of irreversible structural transformations, such as persistent regional depopulation and the reconfiguration of life-stage-specific migration dynamics.

To reflect spatial interactions, we construct spatial weight matrices that exclude island or geographically isolated regions, since commuting via land routes is not feasible. This exclusion is a deliberate design consideration that ensures spatial contiguity–based models do not systematically marginalize certain regions and that the spatial interaction structure reflects actual regional connectivity. Using spatial econometric models and spatial weight matrices, this study empirically analyzes spatial migration patterns while incorporating interregional interactions. <Table 1> provides a summary of the variables used in this study.

This study models the impact of interregional interactions on migration by incorporating these interactions into a spatial weight matrix. First, the spatial econometric model considers the spatial spillover effects on neighboring regions by applying a spatial weight matrix based on regional adjacency. Next, the spatial weight matrix is extended to include economic opportunities and residential substitutability. Finally, we integrate predictor variables into the fixed effects of the panel model to account for unobserved regional heterogeneity.

1) Spatial Econometric Model

Most studies attempt to explain the dynamic changes in migration patterns over a specific period. To address the challenges of population stagnation, depopulation, and concentration in metropolitan areas, it is crucial to examine the drivers of migration across various life cycle stages. Thus, this study analyzes a 10-year period (2011-2020) to derive general results.

Spatial econometric models are useful for controlling for spatial interactions in continuous space (Kim et al., 2025) and enable the evaluation of economic opportunities and residential substitutability between areas. This research’s primary objective is to empirically analyze and present the differential effects of economic opportunities and residential substitutability on migration in an intuitive manner, using the SAR panel model (LeSage and Pace, 2008) in an intuitive manner. Fixed effects models are generally considered more appropriate than random effects models when using spatio-temporal data at the regional level (Elhorst, 2014), a finding that is further confirmed by the statistical verification in this study. Consequently, this analysis employs SAR panel model with two-way fixed effects.

(4)

In the above equation (4), the outcome variable vector Y_i,t is N × 1 dimensional and represents the in-migration and out-migration intensity for each region at time t. X_i,t contains the predictor variables and constant term vectors and is N × K dimensional, and β is K × 1 dimensional parameter of the regression. W_ij is the element i, j of the spatial weight matrix that reflects the spatial interaction of the N regions that comprise the regional panel. All spatial weight matrices were standardized to have row sums equal to unity. The parameter ν_i is a fixed effect for the region and ψ_t is a fixed effect for time. ∑j=1NWijYj,t reflects the endogenous interaction between the outcome variable of region i and the outcome variables of all other regions j, and ρ is a parameter associated with it.

2) Model Construction Using Spatial Weight Matrix

The spatial weight matrix captures migration intensity based on the relationships between adjacent regions and the influence of economic opportunities and residential substitutability. <Table 2> outlines the process for constructing the spatial weight matrix. In the context of multilateral migration, the intensity of migration between two areas is influenced not only by the relative attractiveness of the destination area but also by the attractiveness of other potential destinations (Bertoli and Moraga, 2013; LeSage and Pace, 2008). Consequently, the spatial weight matrix b_ij was constructed using an adjacency-based method to account for the impact of neighboring areas. To incorporate the effect of residential substitutability, the adjacency-based weight matrix was multiplied by the Hadamard product(∘) of the residential substitutability between regions. Spatial influence can be uniquely defined by applying the Hadamard product to both spatial relationships and socioeconomic factors (Dubé and Legros, 2013). This approach enables the spatial weight matrix (wijs) to capture both residential substitutability and the spatial relationships between areas simultaneously. Similarly, economic opportunity was incorporated into the contiguity-based weight matrix (wije), with economic gravity weights applied when selecting migration destinations (Bu et al., 2022; Parent and LeSage, 2008). Finally, recognizing that migration decisions are influenced by multiple factors simultaneously (Kim et al., 2025), the two matrices were combined to reflect the simultaneity, resulting in the final spatial weight matrix wijse.

Table 2. 
Development of different models

All models were estimated using the SAR panel model with the Quasi Maximum Likelihood Estimation (QMLE) method (Yu et al., 2008). The analysis was conducted using the ‘SDPDmod’ package in R version 4.2.1. The variance inflation factors (VIFs) for all models were below 5, indicating that multicollinearity was not a concern. This was not examined further in the models. All models also supported the appropriateness of the two-way panel structure (p<0.001). Additionally, the Hausman test confirmed that the fixed-effects model was more suitable than the random-effects model (p<0.01).

We present key spatial statistical findings on the intensity and spatial patterns of interregional migration across the life cycle stages in South Korea. Our results reveal how migration motivations and spatial dependencies differ by age group and region, providing valuable insights into population migration dynamics.

To explore the spatial dynamics of interregional migration by age group, we conducted a spatial statistical analysis focusing on the spatial clustering of migration intensity and spatial dependence. <Figure 3> shows the results of spatial statistical analysis. It reveals that Seoul has a high concentration of in-migration among young adults. In contrast, non-metropolitan regions in the southeast and southwest have a high concentration of out-migration among the same age group. However, Seoul shows a relatively high concentration of out-migration intensity, except among the young adults, suggesting that individuals are gradually moving away from central urban areas as they enter the latter stages of life. For the family formation, child-education, and middle-aged groups, the spatial clustering of in-migration intensity is either weak or statistically insignificant, indicating a more dispersed pattern of annual average migration intensity. When assessing overall spatial clustering over time using panel data and yearly global Moran’s I statistics on the residuals of the regression model, however, most years exhibit significant positive spatial dependence (p<0.05). This suggests that migration intensity is generally not randomly distributed but spatially structured, despite the relatively weak clustering observed in specific age groups. This reinforces the importance of incorporating spatial dependence into the modeling framework to more accurately reflect interregional migration dynamics.

Figure 3. 
Spatial statistics of migration intensity of South Korea (average of 2011-2020)

Overall, the findings emphasize that migration intensity is closely tied to spatial clustering and varies by age group and region. These findings suggest the need for regional migration patterns that effectively respond to demographic imbalances and support balanced regional development.

Five models were constructed to investigate the effects of economic opportunity and residential substitutability on interregional migration across different life cycle stages. For each model, the spatial weight matrix (W_ij) was constructed using a contiguity-based weighting matrix that reflects spatial relationships between areas in terms of both residential substitutability and economic opportunity. These effects were applied exclusively to the outcome variables. The main effects were analyzed by life cycle stage, and the results are presented in <Table 3> as Bayesian posterior probabilities. The results of the in-migration and out-migration models are shown in <Tables 4 and 5>, respectively.

Through the comparison of Bayesian posterior probabilities, the results of this study suggest that interregional migration may be influenced by areas adjacent to both the origin and destination areas. The analysis indicates that Model 1, which assumes spatial dependence, is more appropriate than Model 0, which assumes there is no spatial dependence between areas. These findings imply that the migration patterns are better understood when considering the influence of residential substitutability and economic opportunities. Furthermore, the results show that the model’s fit improves when considering multiple factors simultaneously rather than focusing on a single influence. This suggests that migration is not driven by a single factor, but rather by the interaction of multiple complex influences. However, since the analysis is based on panel data from a specific period, its validity is limited to that period.

Considering both the effects of economic opportunity and residential substitutability simultaneously improves the model’s fit. For instance, regarding the in-migration intensity of the child-education group, Model 4 (0.531) was found to be more effective in explaining population inflow between regions than Models 2 (0.363) or 3 (0.069), as it accounts for both effects simultaneously. The final spatial spillover effect is determined by the intensity of spatial dependence (ρ), which is associated with the complex spatial effect (wijse = wijs ∘ wije). Thus, Model 4, which considers economic opportunity and residential substitutability simultaneously, is the most appropriate model. These simultaneous effects also yielded similar results for the population inflow in the family formation group.

However, the results may vary depending on the spatial influence of residential substitutability and economic opportunities as applied to different life cycle stages. Most models demonstrated an improved fit when considering the impact of residential substitutability on migration patterns across different life cycle stages. This suggests that individuals tend to select alternative places of residence based on regional similarities, such as housing costs, commuting patterns, living convenience, and public services. Notably, the effects of residential substitutability on the inflow and outflow of the total population were particularly significant. This indicates that interregional migration generally occurs with consideration of job-housing similarity between regions. Although isolating the effects of population attraction and propelling factors when accounting for residential substitutability is challenging, the analysis indicates that residential substitutability must be adequately considered to more comprehensively explain migration systems. Specifically, residential substitutability, as measured using commuting data, reflects the balance between residential locations and job accessibility. This indicates that migration commonly occurs toward areas with similar job accessibility. While residential substitutability does not directly reveal the direction of migration flows, its influence is incorporated into the spatial weight matrix to better explain spatial migration dynamics. Additionally, this modeling approach enables assessing whether single or simultaneous effects better explain migration patterns (Kim et al., 2025). For example, regarding young adult population inflows, the results suggest that considering a single effect rather than simultaneous effects provides a more accurate model. Findings from Model 3 (0.560), which incorporated economic opportunity, suggest that economic opportunity is the most significant factor in attracting young adults. This result aligns with the tendency of young people to migrate to metropolitan areas or large cities in search of economic opportunities.

In some cases, the fit of the model may not improve compared to Model 1, which accounts for spatial dependence, even when spatial effects related to residential substitutability and economic opportunity are included. According to the results, the model incorporating spatial effects of residential substitutability and economic opportunity demonstrates a better fit than the model considering only spatial dependence in the case of population inflow. Among the population outflow models, however, the Bayesian posterior probabilities for family formation and middle-aged population outflow were 0.825 and 0.719, respectively. This indicates that the model accounting solely for spatial dependence yielded a better fit. These results suggest that households tend to migrate shorter distances rather than longer ones and highlight the need for a deeper understanding of policy and structural changes that cannot be fully explained by socioeconomic factors alone.

Most of the predictors used in this study were found to have a statistically significant direct effect on migration intensity. Specifically, regional population density, land prices, and local development budgets were found to positively affect in-migration intensity and negatively affect out-migration intensity. Regions with higher population density and land prices tend to have better-developed infrastructure. Areas with a larger share of development-related budgets, such as those allocated for water resource development, regional and urban development, and industrial complex development, typically prioritize infrastructure improvement. These characteristics suggest that such regions are more likely to attract and retain a population.

Additionally, educational attainment was found to significantly impact in- and out-migration intensities. This implies that regions with higher levels of education are more likely to experience increased mobility because highly educated individuals tend to seek better opportunities elsewhere. This finding aligns with the arguments of Faggian et al. (2017) and Winters (2011), who emphasize the mobility of highly educated populations. Among young adults, in particular, the proportion of the creative class (a proxy for human capital) significantly affected in-migration. This corresponds with Higa et al. (2019), who found that regions with a high concentration of highly skilled workers tend to attract more diverse industries and populations. These findings suggest that young people are drawn not only by job opportunities or cost-of-living advantages, but also by connections to knowledge-based industries and innovation infrastructure.

From an industrial structural perspective, a higher tertiary sector share was significantly associated with in-migration, while a higher secondary sector share was associated with out-migration. This pattern reflects the accelerating deindustrialization of urban manufacturing and the broader transition toward a service-oriented economy. This transformation may lead to increased rural-to-urban migration, which could exacerbate population decline in some non-urban areas.

Most importantly, this study finds that migration decisions are driven not only by socioeconomic disparities between regions but also by spatial relationship structures that cannot be captured by absolute differences alone. The analysis incorporates a spatial weight matrix reflecting non-physical linkages, such as residential substitutability and disparities in economic opportunity, to capture the spatial interdependencies among regions. Even within the relational framework previously established, significant spatial effects were identified. These effects indicate that migration is influenced by the intricate interplay of life cycle stages and interregional connectivity. This influence is more complex than a simple linear movement toward “better” areas.

This study provides empirical evidence that interregional migration is shaped by multi-layered processes involving spatial interdependencies and life-cycle-specific dynamics, rather than being driven solely by regional socioeconomic disparities. Integrating migration theory, spatial effects, and a life-cycle perspective, the study offers a comprehensive understanding of interregional migration mechanisms. Specifically, the study examines the spatial effects of residential substitutability and economic opportunity, emphasizing their simultaneous influence across life cycle stages. Through comparative modeling with alternative spatial weight matrices, the study reveals how spatial interactions impact migration patterns. Due to the consistent statistical relationships between the outcome and predictor variables across the models, the focus of the interpretation shifts from individual predictors to broader structural dynamics. The following discussion derives policy implications based on the results of the analysis.