Introduction

Neighbourhood isolation is the precipitation of disparate experiences across diverse life domains. Depending on where individuals live, they experience distinct variations in access to information and proximity to amenities, economic opportunities, and other living conditions, which may result in desirable or undesirable life outcomes (Friedrichs et al., 2003; He et al., 2022; Sampson et al., 2002, 2008; Sharkey and Faber, 2014). Individuals belonging to the highest strata of society tend to reside in areas with access to the best schools, services, and opportunities. In contrast, other people tend to reside in places that offer services and opportunities at a comparatively lower standard. Such spatial disparities have been found to have intergenerational influences, shaping and perpetuating societal inequalities (Chetty and Hendren, 2018; Chetty et al., 2020).

As demonstrated by the aforementioned studies and numerous other research endeavours, the existing discussion of neighbourhood isolation, its consequences, and the subsequent reproduction of interactions between individuals and their environment has largely focused on their residential locations, specifically their home addresses. This approach assumes that the individuals’ lives are spatially formulated around their homes. However, it overlooks the role of non-home locations experienced by individuals and may not provide a complete picture of the intricate mechanisms through which individuals become socially isolated in their lives. As Levy et al. (2020, p. 926) point out, this is because the vitality of a neighbourhood is “a function of the conditions in neighbourhoods its residents visit and are visited by, and not only its residential (or nearby) socioeconomic conditions”. Hence, beyond the differences resulting from the conditions of people’s residences, recent discussions extend the examination of neighbourhood isolation to explore interactions in spaces where individuals engage in various activities.

This study aligned with the expanded approach of Levy et al. (2020). Building on their concept of ‘triple neighbourhood (dis)advantage’ (TND) (Levy et al., 2020), this study examined three key dimensions at the neighbourhood level in Seoul, South Korea (hereafter, Korea): (1) the degree of neighbourhood (dis)advantage in residential neighbourhoods, (2) the degree of (dis)advantage and isolation in neighbourhoods that people visit, and (3) the degree of (dis)advantage and isolation in neighbourhoods from which people visit, based on the socioeconomic characteristics of neighbourhoods.

Measures of (dis)advantage based on patterns of visits and individual mobility were examined in two different life domains (consumption activities and commuting between home and workplace) through the utilisation of ‘big data’, covering up to 99.3% of the entire population in Korea, which enhanced the credibility and generalisability of the inferences. As a byproduct of an individual’s daily activities, big data could increase our understanding of urban problems and various aspects of urban policy issues, providing the opportunity to articulate and engage in a more nuanced exploration of social isolation within an urban context.

Specifically, we examined the interconnections among the three measures of neighbourhood (dis)advantage by exploring the variability of isolation with respect to socioeconomic levels. Neighbourhood isolation is commonly construed as a phenomenon that occurs predominantly in socioeconomically disadvantaged areas. However, considering that socioeconomically affluent locales also spontaneously restrict their networks (Dwyer, 2007; Massey and Fischer, 2003; Reardon and Bischoff, 2011), we examined whether people living in more advantaged neighbourhoods travelled to areas with similarly high advantages in their daily lives. Thus, our concerns were twofold: first, we explored whether residential neighbourhood isolation was exacerbated by the types of socioeconomic environments and people who have contact with non-home activity locations. Second, we examined whether neighbourhood isolation is intensified or (re)produced by both the disadvantaged and the affluent in activity locations, or whether it is solely driven by one group.

The analysis was conducted in Seoul, Korea. Seoul has one of the highest population densities in the world, but little developable land. It is also home to multiple economic centres and stands out as a global city with a transportation system that is not only affordable but also highly efficient, greatly enhancing the mobility of its residents. However, similar to other global cities, Seoul is increasingly grappling with economic and spatial segregation. These distinctive features of Seoul made it an appropriate focus for our examination of both residential and mobility-based disadvantages and the resulting degree of neighbourhood isolation.

To the best of our knowledge, there are only a few studies on activity locations in Korea Hong et al., 2020; Hong and Choi, 2016) and their focus has not extended beyond simply identifying the similarities and differences in the spatial patterns of inter-neighbourhood activities. Furthermore, even less is known about the heterogeneous relationship between residential neighbourhood (dis)advantage (RND) and the other two mobility-based disadvantages, outdegree neighbourhood (dis)advantage (OND) and indegree neighbourhood (dis)advantage (IND) (detailed descriptions of these measures are provided in the ‘Data and variables’ section). Therefore, our study contributes to a better understanding of spatial-societal inequalities in Korea and beyond. It could also help formulate policies for a more balanced provision of diverse urban social services and infrastructure.

Literature review

A long-running body of research on neighbourhood isolation indicates that neighbourhood socioeconomic conditions are central to residents’ economic, health, and educational outcomes (Chetty et al., 2016; Chyn and Katz, 2021; He et al., 2022; Sampson et al., 2008; Sharkey, 2013; Sharkey and Faber, 2014). When people are exposed to and experience concentrated poverty, unemployment, and significantly higher rates of violent crime in their neighbourhoods, they are likely to have limited access to information, resources, and networks that can provide opportunities and enhance life chances for social advancement. For example, as Chyn and Katz (2021) show, employment rates, adult life expectancy, the frequency of upward social mobility, and children’s academic achievement all decline as the neighbourhood poverty rates increase.

However, elucidating neighbourhood isolation solely through the lens of individuals’ residential locations and their immediate neighbourhoods does not provide a complete picture of how isolation operates and emerges in people’s daily lives because the difference in the experiences and degree of isolation that individuals encounter in their homes and proximal neighbourhoods reflects what occurs in only one of the many spaces in which individuals engage in their activities (Levy et al., 2020). People are also exposed to different class, racial, and ethnic groups in ‘activity spaces’, which encompass non-home locations sought out for work, consumption, leisure, and so on (Candipan et al., 2021).

An activity space is a sphere in which neighbourhood isolation and spatial inequalities can be generated and perpetuated (Tammaru et al., 2021). To illustrate this, we have considered a simplified scenario: a significant proportion of affluent individuals in certain residential areas of a city commute to areas where high-skilled jobs are plentiful. In these locales, these individuals may actively interact and have opportunities to engage in networks that share employment opportunities and information that may be inaccessible to others. Consequently, those who secure employment in such areas have an increased chance of attaining and maintaining high-paying jobs and can establish themselves in distinct residential neighbourhoods with a financial foundation provided by these jobs, creating a cycle that connects advantaged home neighbourhoods and advantaged activity spaces. In contrast, socioeconomically disadvantaged neighbourhoods may experience a recurrence of isolation and segregation within their communities, replicating similar patterns in other disadvantaged neighbourhoods where they frequently visit or spend time.

Alternatively, a completely different scenario may emerge. This is because commuting patterns and interactions in activity spaces where consumption activities take place can mitigate the social and spatial inequalities observed in residential neighbourhoods. The coexistence of certain low-skilled occupations, such as security or cleaning services, in areas characterised by the prevalence of high-income jobs, could facilitate interactions between different socioeconomic strata. Furthermore, during shopping activities, individuals from disadvantaged neighbourhoods may travel to locations that provide better amenities and resources, thereby heightening their exposure to diverse values, norms, and cultural environments and helping to cultivate a network of social ties. As a result, neighbourhood isolation could be diminished by the type of environment and people one comes into contact with within the activity spaces.

To examine whether residential neighbourhood isolation is exacerbated or mitigated by mobility patterns in activity spaces and which scenario is more consistent with it, we constructed three neighbourhood (dis)advantage measures, following Levy et al. (2020): (1) RND, (2) IND, and (3) OND. RND represents the socioeconomic characteristics of residential neighbourhoods; IND and OND show the extent of (dis)advantage in inter-neighbourhood mobility based on the composition of visitors.

The second aspect we focused on, which has been previously overlooked, was that neighbourhood isolation in both residential and activity spaces may be driven not only by disadvantaged groups but also by advantaged groups. Existing research on neighbourhood isolation has overwhelmingly focused on disadvantaged groups, revealing that they are forced into isolation mainly because of segregation and structural constraints (Krivo et al., 2013). However, neighbourhood isolation can be just as powerful for the affluent, and studies have found that the spatial concentration of the affluent is even greater than the poverty concentration (Dwyer, 2007; Krivo et al., 2013; Massey and Fischer, 2003; McAvay and Verdugo, 2021; Reardon and Bischoff, 2011). Unlike disadvantaged groups, the affluent group self-secludes to preserve its economic power and social privilege (Krivo et al., 2013). This is likely to lead to economic segregation and opportunity hoarding, which act as “barriers to the entry of people and the exit of resources” (Dawkins, 2023, p. 793).

The Seoul context: Neighbourhood isolation in a highly dense, multi-centred, spatially and economically segregated, and well-connected city

Seoul, the capital of Korea, is recognised as an emerging global city in the Asia–Pacific region (Beaverstock et al., 1999). It has a population of approximately 10 million and, partly due to a lack of developable land, has one of the highest population densities in the world (approximately 16,000 people per square kilometre) (H. M. Kim and Han, 2012). Seoul is made up of 25 self-governing districts (‘gu’) and 426 administrative boundaries (‘dong’ or ‘hangjeongdong’). Despite variations in population and geographical size, the administration of a dong is similar to the concept of neighbourhoods, as it is based on the idea of providing citizens with uniform access to public services.

Seoul is the economic epicentre of the capital region of Seoul, Incheon, and Gyeonggi province, which is home to approximately 26 million people, more than half of Korea’s total population. In 2019, approximately 47% of all businesses and 50% of all employees were located in this region (Lee et al., 2021). Traditionally, Seoul has three core central business districts: Jongro, Gangnam, and Yeoido. Through a series of policy efforts to create more sub-centres, Seoul has a growing number of sub-centres (e.g., Yongsan, Yangjae, Magok, and Gasan) and regional centres, each with its own area of expertise.Footnote 1

We have chosen Seoul as a case study for several reasons. First, Seoul is an interesting study area because of its substantial population, high population density, and the presence of multiple central business districts and sub-centres. These conditions allow us to study diverse combinations of places where people reside and visit in their daily lives, which may not be fully captured by examining a relatively small city with a monocentric spatial structure or few central business districts.

Second, Seoul is becoming economically and spatially segregated. Similar to other global metropolitan areas, Seoul has a high degree of housing unaffordability. Recent studies have consistently shown that Seoul suffers from growing inequality and segregation in terms of household income, housing prices, and many other public and private opportunities (S. Hong et al., 2022; Y. S. Lee et al., 2022). As illustrated in Fig. S1 in the Supplementary Materials, high- to very-high-income neighbourhoods are predominantly clustered in the southeastern region of the city, typically referred to as Gangnam areas, whereas low- to very-low-income neighbourhoods are concentrated in the southwestern and northeastern regions. A recent report also compared the spatial distribution of housing prices between 2011–2016 and 2016–2021 in Seoul and revealed that the gap between regional housing price levels has widened and the degree of segregation has increased more rapidly than in other major cities in Korea (Lee et al., 2022).

Third, although Seoul has become increasingly economically and spatially segregated, moving around Seoul is feasible and affordable because of its advanced road and public transportation systems. Mohr et al.’s (2021) analysis of the urban transportation system across 25 global cities shows that Seoul ranks among the top ten for accessibility. Seoul is one of the top three cities in terms of affordability, public transport is relatively inexpensive, and it has the lowest monthly travel card price relative to individual income. Moreover, Seoul’s public transportation is not only affordable but also efficient and safe, making it easy for residents to have a relatively large and diverse range of destinations for their activities.

In summary, Seoul is characterised by high population density, diverse commuting and consumption patterns, increasing economic segregation, and a highly efficient and affordable transportation system. Given these conditions, we expected the IND and OND patterns to differ substantially from those of RND because people can be highly mobile in their everyday routines. However, if this hypothesis is not supported, it could be an indication that neighbourhood isolation is socially constructed in much more nuanced ways in activity spaces and cannot be overcome simply by providing affordable and efficient transportation modes, as some scholars have argued to increase mobility among the poor to overcome spatial inequalities (e.g., Blumenberg and Ong, 2001).

We were also interested in investigating whether neighbourhood isolation in residential and activity locations is reinforced by either disadvantaged or advantaged neighbourhoods. In Seoul, where affordable and efficient public transportation networks cover almost all areas, we expected that residents in disadvantaged neighbourhoods would be less constrained in their mobility, whereas those in advantaged neighbourhoods would show a tendency to separate themselves not only by avoiding visiting disadvantaged neighbourhoods but also by making their neighbourhoods less accessible to those from disadvantaged neighbourhoods (Dwyer, 2007; Krivo et al., 2013). We therefore hypothesised that neighbourhood isolation would be driven by relatively affluent groups.

Data and variables

The primary data for constructing the variables were obtained from the Korea Credit Bureau (KCB), the largest credit bureau in Korea, and BC Card, which has the largest market share (approximately 24%). The majority of KCB’s shares are held by 16 large domestic financial institutions, ranging from banks to insurance companies. The company collects financial information from approximately 150 firms that are members of the KCB covering approximately 44.5 million individuals (approximately 99.3% of individuals aged ≥19 years).

For our analysis, we used two types of data from the KCB. First, we employed a series of variables that contained aggregated household information at the neighbourhood level (i.e., at the administrative dong level). Specifically, we used the median household income, composition of household size and distribution of households across nine household income brackets,Footnote 2 number (and share) of households with overdue payments exceeding 90 days, and number (and share) of earners in large corporations.Footnote 3

Second, to measure activity space, we utilised the KCB’s origin and destination (OD) trip data for commuting based on home and workplace addresses. These data have several advantages over government administrative data, especially since household income data are not available through the census or other administrative sources at the neighbourhood level in Korea, so the KCB data bear merit. One limitation of KCB OD data is that it only covers commuting activities.

To supplement non-commuting activities, and more specifically consumption activities, we used information on consumption activities from weekly credit card transaction data from BC Card, which serves a significant majority of the country’s customers, encompassing 36 million individuals out of a total population of 51 million (Lee et al., 2023). BC Card provides real-time credit card transaction data and the data consists of 236 expenditure categories (e.g., restaurant or clothing).

Construction of three neighbourhood disadvantage measures

To construct the RND, we consulted Levy et al. (2020) and incorporated the following four neighbourhood-level variables: median household income, poverty rate, share of earners working for large corporations (i.e., companies listed on the stock market), and share of individuals with overdue payments exceeding 90 days. The poverty rate was calculated using 60% of the median household income, which is a benchmark used to establish the minimum cost of living for an individual rehabilitation system in Korea.Footnote 4 Next, we transformed these variables into a single-factor score based on the results of the principal factor analysis. All four variables showed robust loading on a single factor with high reliability, and the resulting factor scores were used as the RND (detailed results of the principal factor analysis are included in Table S1 in the Supplementary Materials).

To build OND and IND for work and consumption activities, it was first necessary to measure how neighbourhoods were interconnected. For consumption activities, we linked the neighbourhood where a certain customer’s transaction occurred to the neighbourhood of the customer’s home address. We then gauged the inter-neighbourhood connection by analysing the matched pairs and computing the number of customers who initiated card transactions from October 2018 to December 2021.

The OND and IND for commuting-related activities were calculated by connecting the neighbourhood of the worker’s home address to the workplace address and summing the number of people sharing the OD trip. The OD data covered December 2019–December 2022, aggregating approximately three million commuting records per month. OND and IND were calculated using the following formula:

$${\rm{OND}}=\mathop{\sum }\limits_{j=1}^{N}\left(\frac{V\left({n}_{{ij}}\right)}{V\left({n}_{i}\right)-\,V\left({n}_{{ii}}\right)\,}* {{\rm{RND}}}_{j}\right),i\,\ne\, j$$

OND is the weighted disadvantage level of extra-local neighbourhoods to which any neighbourhood \({n}_{i}\) is structurally connected. \(V\left({n}_{i}\right)\) denotes the total number of visits from neighbourhood \(i\), \(V\left({n}_{{ii}}\right)\) represents visits within neighbourhood \(i\), \(V({n}_{{ij}})\) represents visits from neighbourhoods \(i\) to \(j\), and \({{\rm{RND}}}_{j}\) denotes the RND score of the visited neighbourhood.

Visits within neighbourhoods were excluded because we focused on how neighbourhoods were structurally connected through various activities. In addition, the analysis was limited to movements within Seoul to investigate divergent mobility patterns within the area, which is characterised by a similar socio-spatial structure.

The IND was calculated using the following formula:

$${\rm{IND}}=\frac{{\sum }_{j=1}^{N}{{\rm{RND}}}_{j}* V({n}_{{ji}})* {P}_{j}}{{\sum }_{j=1}^{N}V({n}_{{ji}})* {P}_{j}},\,i\ne j$$

IND shows the weighted average disadvantage level of extra-local neighbourhoods to which each neighbourhood \({n}_{i}\) is connected through visits received from other neighbourhoods. Here, population \({P}_{j}\) was adjusted to consider differences in population size. Combined, OND and IND present the RND scores of visited or visiting neighbourhoods, as well as data on the strength of a mobility tie between neighbourhoods. Hereafter, OND and IND based on consumption activity are referred to as Card-IND and Card-OND, respectively, whereas OND and IND based on workplace OD data are referred to as OD-IND and OD-OND.

Control variables

To examine the associations among RND, OND, and IND in a multivariate context, we included a set of controls for neighbourhood demographic characteristics and access to public transportation. Mobility patterns can vary according to the population size and composition. Therefore, the number of residents, the share of men, and the median age of each neighbourhood were included from 2018 to 2022. These demographic data are available from the Korean Statistical Information Service.

The accessibility of public transport can also significantly impact the frequency and convenience with which individuals visit various neighbourhoods. Therefore, we measured the accessibility of subway stations by the area covered within a 500 m radius of the station relative to the total administrative area of the neighbourhood, following the method proposed by Lee et al. (2012). For buses, accessibility was measured by the number of stops relative to the administrative area of the neighbourhood. Subway data were collected from the Rail Portal, while bus data were sourced from the Open Data Portal of Seoul City. Table S2 in the Supplementary Material provides summary statistics of the key variables.

Using the constructed TND measures and control variables, we first analysed the similarities and differences in the spatial patterns of RND, OND, and IND. Then, we investigated the associations among the TND measures. Finally, we further examined whether the associations between each mobility-based neighbourhood disadvantage measure (i.e., Card-OND, Card-IND, OD-OND, and OD-IND) and RND were consistent across neighbourhoods, and whether mobility patterns differed between affluent and economically lagging neighbourhoods, even after controlling for demographic characteristics and access to public transport.

Results

Spatial patterns of TND measures

The spatial patterns of TND measures are presented in Figs. 1–3. As expected, spatial clusters with high RND scores, highlighted in red, were abundant in the southeastern areas, along with a few neighbourhoods in the western part of the city. In contrast, parts of the southwestern, northern, and central areas exhibited a clustering of low RND scores (highlighted in blue).

Fig. 1
figure 1

Cluster analysis of RND using local Moran’s I from 2018–2022.

Full size image
Fig. 2: Cluster analysis of Card-IND and Card-OND from 2018–2021.
figure 2

A Card-IND. B Card-OND.

Full size image
Fig. 3: Cluster analysis of OD-IND and OD-OND from 2019–2022.
figure 3

A OD-IND. B OD-OND.

Full size image

The spatial patterns observed for Card-OND, Card-IND, OD-OND, and OD-IND (see Figs. 2–3) closely resembled those of the RND but, these mobility-based measures clustered over a broader area. First, the overlapping clustering of Card-OND and Card-IND appeared to be linked to the pattern of visits for consumption purposes, which often occurred in locations close to people’s homes and in socioeconomically similar areas. The results are consistent with the findings of Hong et al. (2020) in their analysis based on survey data, which is one of the few studies examining isolation patterns in Korea’s activity space. On the other hand, the similar yet broader concentration in the spatial distribution of commuting trips (Fig. 3) seems to correlate with labour market segmentation and its geographical representation, which will be discussed in the sections ‘Correlation analysis’ and ‘Discussion and conclusion’.

Correlation analysis

The results of the correlation analysis between the neighbourhood disadvantage measures are presented in Table 1. All correlation coefficients were positive, and the magnitudes of the relationships were moderately strong, with a minimum coefficient of 0.535 and a maximum coefficient of 0.720, although the coefficients for the RND OD-OND relationship showed a decline as the COVID-19 pandemic prolonged and severely impacted many industries and jobs in 2021 and 2022.

Table 1 Pearson correlation coefficients between RND and other neighbourhood disadvantage measures.
Full size table

The results suggest that neighbourhoods with residential disadvantages are likely to experience concentrated mobility-based disadvantages, whether for consumption activities or commuting, while neighbourhoods with residential advantages are likely to experience concentrated advantages. Despite the relatively limited occurrence of mobility barriers, mobility patterns in Seoul appeared to be socially patterned, with evidence of strong social and neighbourhood homophily.

Heterogeneity in the relationship among TND measures

Are the relationships among measures of neighbourhood (dis)advantage primarily influenced by the socio-spatial isolation of disadvantaged groups, as emphasised in previous studies? Alternatively, as some recent studies have suggested (Dwyer, 2007; Krivo et al., 2013), might the concentration of advantaged groups be a more significant source of increasing levels of neighbourhood isolation? In this section, we explored possible heterogeneous associations by dividing neighbourhoods into five quintiles based on the level of RND and conducting regression analyses for each quintile explaining Card-OND, Card-IND, OD-OND, and OD-IND.

Figure S2 in the Supplementary Materials presents the neighbourhoods divided into quintiles. First, areas in the Q4 and Q5 quintiles were clustered along the east-west axis of Seoul, characterised by better socioeconomic conditions. These Q4 and Q5 areas included major employment centres for relatively high-wage jobs and industries and were well served by transport and amenities.Footnote 5 In contrast, Q1 and Q2 areas located in the southwestern and northern parts of Seoul overlapped with regions dominated by lower-wage jobs and industries and had the lowest levels of public transportation accessibility in the city (Yoo et al., 2020). This geographic segmentation may influence neighbourhood isolation in mobility patterns, which will be discussed later.

Heterogenous RND, Card-OND, and Card-IND relationships

The results of the analysis of the relationships among RND, Card-OND, and Card-IND are presented in Tables 2 and 3. For comparison, the estimated associations among RND, Card-OND, and Card-IND for all neighbourhoods are shown in the first column of each table. The associations remained positive and statistically significant, even after controlling for demographic characteristics and access to public transportation, as shown in Table 1.

Table 2 The results of regression models that explain Card-IND from 2018 to 2021.
Full size table
Table 3 The results of regression models that explain Card-OND from 2018 to 2021.
Full size table

However, in the regression analysis by quintile (column headings Q1–Q5 in Tables 2 and 3), the relationships among the three disadvantage measures did not consistently achieve significance across all quintiles, except for the most advantaged neighbourhoods (Q5). In these neighbourhoods, we found a close and statistically significant relationship between residential neighbourhood advantage and advantage in consumption activity locations over time, suggesting that individuals residing in Q5 areas tend to visit similarly advantaged areas for consumption activities and restrict their networks to members of similar groups, as shown in Krivo et al. (2013). That is, neighbourhood homophily is not uniform in Seoul; however, it is notably stronger in the most advantaged neighbourhoods. This trend remained unchanged even after the COVID-19 outbreak in 2020. The results of the regression analysis based on the OD data, which will be discussed later, suggest a slightly different pattern, likely due to differences in the nature of visits.

The tendency towards neighbourhood homophily became even more apparent when examining the quintile neighbourhoods belonging to Q5 that fell into Card-OND and Card-IND (see Table S3 in the Supplementary Materials). Specifically, of the neighbourhoods classified in Q5 by RND, approximately 85% belonged to Q5 or Q4 by Card-OND, suggesting that the majority of visits originating from Q5 of RND were directed to similarly advantaged neighbourhoods. Similarly, for Card-IND, more than 82% of neighbourhoods in Q5 of RND were categorised under Q5 or Q4 in 2019. In contrast, within the Q5 of RND, where Card-OND fell within Q1, it comprised only 1.2%. For Card-IND, there was not a single neighbourhood belonging to Q5 of RND that fell under Q1.

For the lower quintiles from Q1 to Q3 of RND, no consistently significant relationships were found between RND and Card-IND (Table 2). This finding suggests that disadvantaged areas did not necessarily experience a higher rate of visits from similarly disadvantaged areas; rather, visits occurred significantly from more advantaged areas, indicating a more mixed social environment in the Q1 to Q3 neighbourhoods. This visiting pattern contrasts with the self-secluding pattern identified above in RND Q5 areas. The relationship between RND and Card-OND in Table 3 follows a similar pattern, with the exception of some positive coefficients observed in the lower quintiles (i.e., Q1 in 2019 and 2021 and Q2 in 2020), although these coefficients lack consistency.

Again, for Card-IND and Card-OND, we examined the quintile into which the areas classified as Q1 to Q3 of the RND were placed (Table S3 in the Supplementary Materials). For instance, in 2019, in areas belonging to Q3 of the RND, the proportion of Card-IND in Q3 was 22%, the proportion in Q2 of Card-IND was even higher at 33%, and the proportions in Q1 and Q4 were 16% and 20%, respectively. This again indicates a considerable mix and a substantial number of visits from both relatively advantaged and disadvantaged areas.

Regarding the effects of the control variables, the results were in line with expectations. In general, younger people are more open to diverse cultural environments (Bump, 2019). In addition, better public transportation is expected to lower the physical and mental barriers to interacting with other neighbourhoods. Our findings support existing ideas: as the median age decreased and public transport accessibility improved, individuals tended to engage more with neighbourhoods of diverse socioeconomic backgrounds.

Heterogenous RND, OD-OND, and OD-IND relationships

Tables 4 and 5 show the relationships among RND, OD-IND, and OD-OND. The association between RND and OD-IND for all neighbourhoods in the first column of Table 5 is significant and positive, similar to the relationship observed for RND and Card-IND. Again, this positive relationship revealed heterogeneity depending on the RND level. We found a significant and positive relationship in the most advantaged neighbourhood group, whereas the relationship between RND and OD-IND was not significant in the Q1 to Q3 areas of the RND, except for Q2 in 2022.

Table 4 The results of regression models that explain OD-IND from 2019 to 2022.
Full size table
Table 5 The results of regression models that explain OD-OND from 2019 to 2022.
Full size table

The findings from the relationship between RND and OD-OND aligned consistently with the analysis results for the association between RND and Card-OND, in that they showed significant and positive coefficients when measured for the entire neighbourhood and revealed heterogeneity across quintiles. However, the relationship between RND and OD-OND differed from that between RND and Card-OND in two ways: first, the relationship between the most disadvantaged area, Q1 of RND, and OD-OND showed consistent significance and a strong association, even when compared with the results for Q5 of RND. The significant correlation in the Q1 of the RND area was primarily due to the spatial landscape of jobs and work. As discussed earlier, relatively low-wage industries, such as manufacturing, are primarily distributed within socioeconomically disadvantaged areas of Seoul. A high proportion of the individuals employed in these low-wage industries reside in areas of lower socioeconomic status. It appears that a substantial proportion of residents in disadvantaged areas commute to similarly disadvantaged areas, demonstrating an isolated mobility pattern.Footnote 6

Second, in the analysis focusing on Q5 areas, the RND variable lost its significance in 2021 and 2022. This is likely indicative of the impact of the COVID-19 pandemic on the practices and locations of working professionals. Professionals engaged in professional and management services, as well as in the finance industry which offers high-wage jobs, were concentrated in advantaged areas in Seoul, as in other global cities. After the COVID-19 outbreak, many of these office jobs transitioned to remote work, resulting in a reduction in the frequency and share of visits to advantaged neighbourhoods where jobs in these sectors were concentrated. Therefore, the weakened relationship between RND and OD-OND in 2021 and 2022 needs to be interpreted as a phenomenon arising from changes in office workers’ working methods rather than as an indication of a diminishing trend in the seclusion of the affluent.

Discussion and conclusion

This study explored the potential mechanisms by which neighbourhood isolation is created and reinforced in non-home activity locations. Specifically, we examined: (1) whether residential neighbourhood isolation is reinforced by social interactions and conditions that individuals encounter in consumption and commuting activity spaces, and (2) which socioeconomic group drives the association between isolation in residential neighbourhoods and activity spaces—whether both disadvantaged and privileged groups contribute to isolation through differential use of urban space, or whether one group predominantly drives this association.

Research on neighbourhood isolation is often criticised for focusing exclusively on residential neighbourhoods and ignoring the contact and interactions that individuals experience while performing routine activities such as shopping, working, and other activities outside their homes. Neighbourhood isolation based on RND can either be alleviated by such interactions in activity spaces or exacerbated by neighbourhood segregation in activity spaces. People may travel to areas similar to their home neighbourhoods, thereby reinforcing neighbourhood isolation. Alternatively, people may be exposed to different values, norms, cultural environments, and networks of social ties in activity spaces.

Finding 1: The findings for Seoul were consistent with those of the former scenario. Despite the unique characteristics of Seoul, including its high population density, the presence of multiple business centres, and highly efficient and affordable transportation systems, the patterns of IND and OND were not substantially different from those of RND. This appears to stem from the relationship between the socioeconomic characteristics of the neighbourhood and the characteristics and locations of the labour markets to which people commute. According to labour market segmentation theory, the labour market is divided and segmented into different sectors by occupation, industry, and other factors (Bauder, 2001; Doeringer and Piore, 1971). From a spatial perspective, a simplified description of this segmentation would be that the primary segment, consisting of high-wage, stable jobs, and the secondary segment, characterised by low-wage, less stable jobs, are located in distinct areas and correspond to separate labour markets.

When considering the segmented labour market alongside the varying socioeconomic statuses of neighbourhoods, it can be inferred that there will be stronger interactions between advantaged (disadvantaged) neighbourhoods and neighbourhoods where the primary (secondary) segment of the labour market is located. And since the primary (secondary) segment of the labour market is predominantly located in advantaged (disadvantaged) neighbourhoods with excellent (poorer) amenities, Fig. 1 depicting the distribution of RND, IND, and OND, exhibited similar geographic patterns.

Finding 2: Mobility patterns in activity spaces tend to reinforce neighbourhood isolation, mainly through the self-seclusion of the most advantaged neighbourhood group. This reinforcement of neighbourhood isolation and the reduction of shared routines in activity spaces are understood to be influenced by material constraints and social distance (Browning et al., 2017). Here, material constraints refer to differences in affordability and accessibility, which often restrict the places that lower-income individuals can visit, for example, due to limited transportation options. However, Seoul, with its relatively affordable and well-connected transportation system, shows that transport options contribute, at least in part, to reducing isolation. Nevertheless, this influence is not significant enough to counteract the overall pattern of neighbourhood homophily. Specifically, there was a strong tendency toward self-seclusion in the highest socioeconomic status group (Q5), likely due to the significant role of social distance.

Social distance refers to the differences in socioeconomic status that can result in diverse lifestyles, preferences, and attitudes, potentially diminishing the willingness to share spaces for activities and routines (Browning et al., 2017). To the extent that people foster a sense of group identity as the socioeconomic class becomes more similar, those of similar socioeconomic class may prefer to share spaces with others of similar class, while discouraging the sharing of spaces with different classes (Hipp and Perrin, 2009). Krivo et al. (2013) articulated this phenomenon as the self-seclusion of affluent groups, positing that it may stem from their propensity to limit their activities and visits to areas of similar advantage, driven by economic power and social privilege. Our study also finds evidence to support this tendency.

Finding 3: In contrast, in the lower socioeconomic status quintiles, Q1 to Q3, neighbourhoods showed broader cross-interaction across status groups, probably because of Seoul’s transit-friendly environment. Only when measuring the places visited for commuting purposes in mobility patterns did most of the lower socioeconomic neighbourhoods show a tendency to become more isolated as their status declined. Exposure to diverse and mixed social environments does not guarantee better life outcomes. However, if enhanced social interactions between low- and middle-income groups evolve organically through voluntary and gradual processes, as opposed to being developed through policy interventions or strategies, it is more likely to lead to increased access to economic opportunities (better educational attainment and employment prospects) for low-income groups, as evidenced by Galster et al. (2015). Considering this, the observed interactions within neighbourhoods Q1–Q3 may indicate a positive phenomenon in Seoul, where spatial and economic segregation is increasing. Furthermore, interactions within activity spaces provide tools and ideas to mitigate segregation and its negative influences. Identifying detailed tools to enhance exposure to other environments and socioeconomic groups through a voluntary process becomes a task for policymakers and practitioners or further research. However, it would be safe to argue that promoting a greater mix of diverse groups in activity spaces, for example by providing leisure activities, is likely to be a faster and more expedient way to reduce segregation than altering the configuration of residential environments that have long been constructed by economic, cultural, ethnic and other factors (Hedman et al., 2021). Our findings endorse recent perspectives and highlight the significance of broadening the analysis of neighbourhood isolation. We extend previous research by statistically elucidating the relationship between TND and taking advantage of more extensive ‘big data’ to capture intricate details of socioeconomic status and mobility patterns of neighbourhoods.

There are several avenues to explore. First, future research should examine how isolation in residential and activity spaces has different consequences. Neighbourhood isolation implies that people spend time in different parts of a city, creating variations in access to resources, networks, and information. These differences could influence access to employment opportunities, and public and social services, and these could even affect their values and norms. Second, although information on the amount of time people spend in various neighbourhoods was unavailable in the datasets for this study, having such data would be paramount to fully understanding the differences in consequences and the relationship among TND. Lastly, it would be necessary to conduct an analysis for other cities with different characteristics, considering that they vary in population composition, distribution of employment centres, degree of pre-existing residential segregation, and public transportation systems, among others. Correlational or comparative studies employing other universal and widely used indicators of neighbourhood (dis)advantages may also reveal either generalisable or distinctive patterns of neighbourhood isolation across different urban contexts.