Station-Level Effects of the COVID-19 Pandemic on Subway Ridership in the Seoul Metropolitan Area

Abstract

This paper investigates the station-level impacts of the coronavirus disease (COVID-19) pandemic on subway ridership in the Seoul Metropolitan Area. Spatial econometric models are constructed to examine the association between ridership reduction caused by the pandemic and station-level characteristics during the pandemic years 2020 and 2021. The results reveal unequal effects on station-level ridership, based on the pandemic waves, the demographics, and the economic features of pedestrian catchment areas. First, the subway system was severely disrupted by the pandemic, with significant decreases in ridership—by about 27% for each of the pandemic years—compared with the pre-pandemic year (2019). Second, the ridership reduction was sensitive to the three waves in 2020 and responded accordingly; however, it became less sensitive to the waves in 2021, indicating that subway usage was less responsive to pandemic waves during the second year of the pandemic. Third, pedestrian catchment areas with higher numbers of younger residents (in their 20s) and older residents (65 years and older), those with more businesses requiring face-to-face interactions with consumers, and stations located in the employment centers were hit the hardest in ridership reduction caused by the pandemic.

Since the first confirmed cases were reported in Wuhan, China, in December 2019, coronavirus disease (COVID-19) had infected 327.7 million people and resulted in 5.54 million deaths globally, as of January 17, 2022, disrupting daily life worldwide ( 1 ). Since the first confirmed case in South Korea was reported on January 20, 2020, there were 696,032 confirmed cases and 6,333 deaths nationwide, as of January 17, 2022 ( 2 ). South Korea has received attention for successfully implementing quarantine measures—there have been relatively few confirmed COVID-19 cases and related deaths—without draconian lockdown measures. The South Korean government tried to control the spread of the disease by taking well-organized epidemic control measures, such as mass testing to confirm positive cases, rigorous contact tracing and quarantine to prevent further spread, limiting the size of social and private gatherings, and treating those who became infected in the early stages. Despite these actions, the number of confirmed cases surged in five pandemic waves during the 2020–2021 period, as shown in Table 1 and Figure 2. After the second wave, which was highly associated with outbreaks in churches in Seoul (the nation’s capital), the Seoul Metropolitan Area (SMA) became the epicenter of the COVID-19 pandemic in South Korea, accounting for 70.8% of the confirmed cases nationwide (492,458) as of January 17, 2022 ( 2 ).

Table 1.

Timeline of Key Events During the COVID-19 Pandemic in South Korea

Date	Event	Social distancing and restrictions
2020/01/20	The first case of COVID-19 was reported in South Korea
2020/03/22–2020/04/19	The first wave	•Mass infections at church services •Recommendation to delay or cancel all meeting, eating out, travel
2020/08/16–2020/10/12	The second wave	•Sporadic cluster infection at publicly used facilities •Ban social gatherings of 50 or more people
2020/11/19–2021/3/30	The third wave	•Nationwide infection •Ban private gatherings of five or more people •Restaurants and cafes must close at 10:00 p.m.
2021/02/26	First vaccination was administered in South Korea
2021/07/01	Distancing in daily life	•Ban private gatherings of six or more people •Restaurants and cafes must close at midnight
2021/07/06–2021/10/30	The fourth wave	•Private gathering: Up to four people allowed until 6:00 p.m., up to two people allowed after 6:00 p.m. •All adult entertainment facilities must be closed all the time •Restaurants, cafes, bathhouses, movie theaters, private academies, hair and beauty salons, PC rooms, theme parks, stores, supermarkets, and department stores must close at 10:00 p.m.
2021/10/24	70% of the population has been vaccinated
2021/11/01	Step-by-step recovery to normal daily life (living with Covid-19)	•No restrictions of operating time in all facilities except entertainment facilities (close at midnight) •Limited up to 10 people for private gatherings in the Seoul Metropolitan Area (SMA) and 12 people outside the SMA
2021/11/15–2021/12/31	The fifth wave	•The first omicron infection was reported (2021/12/01) •Step-by-step recovery stopped (2021/12/18) •Limited up to four people for private gatherings nationwide •Entertainment facilities, restaurants, cafes, gyms, and karaoke bars must close at 9:00 p.m. •Movie theaters, performance halls, and PC rooms must close at 10:00 p.m.

Figure 2.

Weekly subway ridership in the Seoul Metropolitan Area (left axis) and confirmed cases of COVID-19 in South Korea (right axis).

Following the third wave in November 2020, the government started implementing coercive measures to limit the size of private gatherings and the operating hours of restaurants, cafes, and entertainment facilities. In October 2021, as 70% of the population had received vaccinations, restrictions on business hours—except for some entertainment facilities—were lifted and the limit for private gatherings were eased to 10 people in the SMA and 12 people outside the SMA, in an effort to encourage citizens to return to normal daily life. However, these deregulation measures were suspended on December 18, 2021, because of rapid surges in positive cases as well as severe cases, which led to an overload of the medical system. As cases with the omicron variant infection were first confirmed on December 1, 2021, deregulation measures were stopped and the size of private gatherings was again limited to four people nationwide. In addition, the business hours of restaurants, cafes, and entertainment facilities were limited to nine or ten o’clock at night. At the time of this study, South Korea was in its fifth wave and struggling to contain the spread of COVID-19.

As shown in the SMA, because of their agglomeration and connectivity, urban zones, are becoming the epicenters of the pandemic in numerous countries; as such, urban vulnerability and resilience to the pandemic are actively being discussed. Although the social interactions in metropolitan areas are praised as a source of the urban agglomeration economy and as a driving force for urban growth, such locales contain an infrastructure conducive to the rapid transmission of the virus in a pandemic situation, resulting in metropolitan areas becoming hubs for public health threats and the urban diseconomy ( 3 ). Despite the short period of time since the outbreak of COVID-19, several studies have explored the pandemic’s socio-economic, environmental, health, mobility, transportation, land use, and design impacts on urban regions (e.g., Sharifi and Khavarian-Garmsir [ 4 ]). Among many pandemic-related urban issues, population mobility has drawn scholarly concern, because it is regarded as one of the contributing factors to intra-urban virus transmission and as a good indicator of urban connectivity. Existing studies focus mainly on the associations between mobility (or travel restrictions) and virus transmission ( 5 – 8 ) and on changes in travel behavior in a pandemic situation ( 8 – 11 ). Studies on changes in travel behavior have investigated the pandemic’s effects on the shifts in the mode of transport used, from public transit to private cars or non-motorized modes such as walking and biking ( 12 – 14 ), and on the reduction in transit ridership ( 9 , 14 , 15 ).

Studies on the pandemic’s impacts on transit ridership commonly report a significant decline in transit patronage as the virus spread and restrictions were imposed: −87% in Bogotá, Colombia, in March, 2020 ( 16 ); −88.6% in Madrid, Spain; −80.7% in San Francisco and San Jose; −75% in New York and New Jersey; −77.5% in Boston; and −69.3% in Rio de Janeiro, Brazil in April, 2020 ( 17 ). Fixed-line bus ridership dropped by 66.9% and 65.1% from 2019 baselines in Nashville and Chattanooga, Tennessee ( 15 ). In New York City, the average subway and commuter rail ridership was down by 80%, while bus ridership was down by 50% in the first week of July, 2020, with a peak subway ridership decline of 94% in late March ( 14 , 18 ). Wang and Noland ( 14 ) investigated the impact of the pandemic on the Citi Bike system and the subway system in New York City, after controlling for weather patterns. They found significant decreases in both subway ridership and bikeshare usage initially, but different resilience between the two modes. By the time of that study, bikeshare usage had recovered nearly to pre-pandemic levels; however, subway ridership remained below pre-pandemic level.

Since COVID-19 can have long-lasting and structural effects on travel behavior and mobility, analyzing the pandemic’s effects on transit ridership is of great significance from the perspective of urban resilience, urban and transport planning, public health, and transport infrastructure investment and financial policy. Moreover, as transit ridership is regarded as a derived demand to accomplish urban socio-economic activity, examining the pandemic’s impact on ridership is an effective way to investigate how people’s economic and social activities have changed, and to grasp the relationship between changes in travel behavior and the characteristics of transit stations and their pedestrian catchment areas (PCAs) during the pandemic, especially in cities with high transit dependency, such as Seoul. However, because transit ridership is influenced by diverse factors (e.g., station characteristics, the transit level of services, land use patterns, and the socio-economic and demographic traits of the PCAs), the pandemic’s effects on ridership are likely to vary based on them. Numerous empirical studies have investigated determinants of transit ridership over the last several decades. The determinants include population and employment density, land use features such as density, design, and diversity, the mixture of land uses, intermodal connectivity, and node and place ( 19 – 30 ).

Although numerous studies report significant declines in subway ridership at the urban level during the pandemic, few studies could be found that investigated the outcomes of the pandemic on subway ridership at the station level, despite spatiotemporal variations in impacts by station. Another characteristic of the current study is that while most existing pandemic studies analyzed the impacts of the early period of the pandemic, this study analyzed the dynamic subway usage patterns during the two years of the pandemic. It is also notable that the pandemic’s impacts on ridership with regard to station-level characteristics can be quantified through statistical modeling.

For this study, the effects of the COVID-19 pandemic on subway ridership in SMA were examined, focusing on spatial variations in the pandemic’s impacts at the station level. In doing so, logistic regression models were developed to examine the association between ridership reduction caused by the pandemic and station-level characteristics during the 2020 to 2021 pandemic years. Zones within 600 m of each subway station were delineated as PCAs. For the methodology, a spatial econometric approach was employed to statistically test the pandemic’s impacts given the existence of spatial autocorrelation in the level of ridership reduction.

Data and Research Method

Data

SMA is one of the largest, densest metropolitan areas in the world and had more than 25 million inhabitants as of 2019. SMA has an extensive subway system, which includes 22 lines covering approximately 1,200 km, with more than 500 stations facilitating a daily average of about 14.9 million passenger trips in 2019 ( 31 ). Daily ridership data were obtained from smart card data from January 1, 2019, to December 31, 2021 ( 32 ). The ridership data contain details on daily boarding and alighting passengers by station. Data on land use and the socio-economic aspects of the PCAs were derived from the South Korea Statistical Office’s jipgyegu dataset of 2020, which provides highly and spatially disaggregated data on the population, housing, households, and industry in 50,464 spatial units of the SMA. Information was obtained on location and connectivity-related variables, including stations located in the central business district (CBD), the Gangnam and Yeongdeungpo subcenters, the numbers of bus stops and nodes, and the link length in a PCA using geographic information system spatial analysis tools. Details of subway service-related variables, such as the number of transfers, headway, and express line stops, were obtained from subway operation organizations, as these factors are known to affect subway ridership. Figure 1 shows the locations of the subway stations in the metropolitan region.

Figure 1.

The subway system and employment centers in the Seoul Metropolitan Area.

Research Methods

To investigate the pandemic’s effects on subway ridership, both exploratory and statistical analyses were conducted. First, spatiotemporal changes in ridership before (2019) and after the start of the pandemic (2020 and 2021) were explored. Specifically, changes in ridership were analyzed after the first confirmed case was reported and during the five waves of the pandemic, as well as changes in ridership based on the day of the week and the time of day, before and after the pandemic. For the spatial exploratory analysis, stations that were hit the hardest in ridership reduction were identified by comparing the ridership of 2019 and 2020, and that of 2019 and 2021.

Second, logistic regression models were conducted to examine the association between ridership reduction caused by the pandemic and station-level characteristics. Two models were constructed to analyze how the pandemic reduced the ridership in 2020 and 2021, compared with the ridership in 2019. For the dependent variable, the log-odds of the shares of 2020 and 2021 to 2019 ridership were used for the logistic model. As shown in Table 2, as the percentage change in ridership from 2019 to 2020 is between −2.4% and −75.1%, and the change from 2019 to 2021 is between −0.1% to −79.7%, the proportions of 2020 and 2021 ridership of that of 2019 are bounded by 0 and 1. For the dependent variable of the logistic model, first the shares were converted into odds that removed the upper bound and took the natural logarithm of the odds (log-odds) to remove the lower bound ( 33 ). This conversion is beneficial in that the coefficients of the logistic model can be interpreted as odds ratios after exponentiation, meaning the probability of the pandemic having an impact over the probability of it having no impact. Further, pandemic wave dummy variables were included in the logistic model to account for the effects of each wave by grouping daily ridership data into four groups (waves 1–3 and a non-wave period) in 2020 and three groups (waves 4–5 and a non-wave period) in 2021. The ridership figures during each wave in 2020 and 2021 were then compared with the corresponding days in 2019. This approach allowed different impacts of the pandemic on ridership by year and by wave to be measured.

Table 2.

Descriptive Statistics of Variables Used in the Statistical Models

Variable	Mean	Std dev	Minimum	Maximum
Total ridership in 2019 (million)	11.2	11.5	4.E-02	74.7
Total ridership in 2020 (million)	8.1	8.0	3.E-02	51.9
Total ridership in 2021 (million)	8.2	7.8	5.E-02	49.7
Change in 2019 and 2020 ridership (%)	−26.1%	7.6%	−75.1%	−2.4%
Change in 2019 and 2021 ridership a (%)	−24.2%	9.4%	−79.7%	−0.1%
Number of buses stopped per day in a PCA	130.4	77.8	2.0	617.0
Length of links in a PCA (km)	13.4	4.7	0.6	27.2
Number of nodes in a PCA	19.6	11.6	0.0	67.0
Number of transfers	1.2	0.5	1.0	4.0
Express line stops (dummy)	0.3	0.5	0.0	1.0
Headway (minutes)	5.5	4.8	2.0	23.6
Number of train vehicles	411.5	212.3	36.0	814.0
Proportion of the population aged between 20 and 29	0.15	0.05	0.05	0.57
Proportion of the population aged 65 or above	0.14	0.05	0.01	0.40
Proportion of manufacturing firms	0.06	0.09	0.00	0.64
Proportion of wholesale and retail businesses	0.31	0.11	0.00	0.93
Proportion of transportation businesses	0.01	0.02	0.00	0.25
Proportion of accommodation and restaurant businesses	0.24	0.10	0.00	0.88
Proportion of publishing, broadcasting, and information businesses	0.01	0.03	0.00	0.19
Proportion of finance, insurance, and real estate businesses	0.06	0.05	0.00	0.43
Proportion of scientific and technical services, and business support businesses	0.05	0.06	0.00	0.44
Proportion of education and public administration services	0.05	0.06	0.00	0.48
Proportion of health entertainment, culture, and sports-related services	0.06	0.04	0.00	0.26
N	476

Note: PCA = pedestrian catchment area.

^aThe number of stations is 467.

As to the explanatory variables for both models, three variable types for the influencing factors on subway ridership were selected, based on the existing literature: station characteristics (six variables), the transit level of services (four variables), and a PCA’s socio-economic and demographic characteristics (11 variables). Demographic and economic variables include proportions of residents from different age cohorts and proportions of businesses, according to industry. Subway connectivity variables include the number of buses that stopped per day, as well as the node number and link length within a PCA. Subway line or station characteristic variables include the number of transfers, express line stops, train vehicles, and headway. Three dummy variables were included to indicate whether subway stations are located in the employment centers, including the CBD, Gangnam, and Yeongdeungpo.

Table 2 presents the summary statistics for the variables used in the statistical models. Of the 508 stations for which ridership data were available, the analysis included data on 480 stations located within SMA. It was found that, out of the sample, 476 stations had a decrease in ridership in 2020 and 467 stations in 2021, compared with ridership in 2019. The annual mean values of ridership in 2019, 2020, and 2021 were 11.2 million, 8.1 million, and 8.2 million, respectively. The ratios of total ridership in 2020 and 2021 to 2019 ridership were 72.4% and 73%, respectively; this indicates a reduction of approximately 27% during the pandemic. Moreover, ridership did not recover from the 2020 reduction in 2021. The average percentages of ridership reduction per station before and after the pandemic were 26.1% in 2020 and 24.2% in 2021, indicating a minor recovery from 2020 in 2021. The proportions of firms by sector ranged from 1% to 31%, while shares of the population in their 20s and 65 years old or above were about 15% in each cohort. The mean subway headway was 5.5 min, while the mean number of train vehicles per line was 412.

For the statistical models, spatial econometric models were built to determine the number of ridership reduction clusters, based on each station and its PCA characteristics. The Moran’s I Index values of log-odds (the dependent variable) for 2020 and 2021 were then calculated, as well as their p-values, to evaluate the significance of that index. The Moran’s I indices for 2020 and 2021 were 0.216 and 0.432, respectively, and were significant at the 1% significance level. This indicates the presence of positive spatial autocorrelation; high values are close to other high values. To address the spatial dependence of the dependent variable, a spatial autoregressive model (SAR) was applied to address the endogenous interaction effects of a spatially lagged dependent variable. The SAR model takes the following mathematical forms ( 34 ):

$$\mathrm{SAR}:Y=\rho \mathrm{WY}+\alpha l_N+X\beta +\epsilon$$ (1)

where Y is an N×1 vector of the dependent variable, $l_N$ is an N×1 vector, X denotes an N×K matrix of explanatory variables, $\mathrm{\alpha },\beta$ are the parameters to be estimated, $\epsilon$ is the vector of error terms, W is an N×N spatial weight matrix, and $\rho$ denotes the spatial dependence coefficient.

For this study, the spatial autocorrelation was modeled using six different spatial weight matrices for the SAR model: rook contiguity, two distance-based (4 and 8 k-nearest-neighbors) and three kernel functions (uniform, triangular, and gaussian). Though there was no significant difference in the model results according to the spatial weighting matrix, the SAR models for both logistic models with the 4 k-nearest-neighbors spatial weight matrix were finally selected as the best models, as they have the lowest Akaike Information Criteria.

Results

Exploratory Analysis Results

Figure 2 illustrates the changes in ridership before and after the first confirmed case was reported (January 20, 2020). Every year, sharp declines can be seen in February and October, regardless of the pandemic; these are because of the major national holidays of Chinese New Year and Chuseok (South Korean thanksgiving). There were also substantial ridership declines in March, August, and November, 2020, during the three pandemic wave periods. However, although the overall ridership in 2021 was substantially lower than in 2019, ridership seemed to be less sensitive to the pandemic waves in 2021 than in the previous year.

Figure 3 presents the ridership based on the day of the week and the time of day, before (2019) and after the start of the pandemic (2020–2021). It shows significant ridership declines on all days of the week and at all times of the day after the start of the pandemic. Ridership decreased by about 22% to 27% on weekdays and by about 37% and 40% during the weekend in both pandemic years. Ridership decreased by 18% to 20% during morning peak hours (7:00–9:00 a.m.) and by 21% to 26% during evening peak hours (5:00–7:00 p.m.) in both pandemic years.

Figure 3.

Ridership by (a) day of the week and (b) time of day between 2019 and 2021.

Figure 4 presents the spatial distribution of reduction rates between 2019 and 2020 (2020 ridership/2019 ridership) and between 2019 and 2021 (2021 ridership/2019 ridership). Both maps show that the stations with the largest ridership reduction are found in the CBD, Gangnam, Dongdaemoon, and Jamsil, where shops, restaurants, and cultural and entertainment facilities (such as theme parks) are concentrated. The maps show that reductions in ridership were relatively larger at stations located in the employment centers, in both 2020 and 2021.

Figure 4.

Spatial distribution of reduction rates in ridership (a) between 2019 and 2020, and (b) between 2019 and 2021.

Effects of the Pandemic on Subway Ridership

Table 3 shows the logistic regression results of the analysis of the pooled data set by wave, to measure the effects of explanatory variables on ridership reduction that was caused by the pandemic. Overall, the models performed well, with an R² of 0.77 for the 2020 model and 0.43 for the 2021 model. The significant positive values of ρ in both models reflect the existence of spatial dependence inherent in the data. Several findings can be highlighted. First, the coefficients of the three wave dummy variables in the 2020 model had statistically significant negative values; this indicates a more pronounced reduction in subway ridership during the wave periods than during the non-wave period in 2020. The odds ratios of the three wave dummy variables indicate that the first wave period had the largest reduction, at 66.4%, while the second and third waves were likely to see similar reductions to those in the non-wave period, at 57.6% and 60.0%, respectively. However, ridership reductions during the waves of 2021 show a different pattern from that of 2020. The fourth wave period showed a decrease of only by about 10%, compared with the non-wave period, while ridership in the fifth wave increased by 6.6%, indicating that, in 2021, the wave effect was insignificant. This seems to be related to subway use being less responsive to the waves during the second year of the pandemic, as the number of confirmed cases continues to increase in 2021.

Table 3.

Outcomes of Spatial Autoregressive Model for Pandemic Effects on Subway Ridership

	2020 model (2019 versus 2020)			2021 model (2019 versus 2021)
	β	z-Value	OR	β	z-Value	OR
Number of buses stopped per day in a PCA	−2.E-04	−2.4**	1.000	−0.001	−2.8***	0.999
Length of links in a PCA (km)	0.009	3.6***	1.009	0.016	2.9***	1.016
Number of nodes in a PCA	−0.003	−2.7***	0.997	−0.007	−3.4***	0.993
Number of transfers	0.011	0.8	1.011	0.027	0.9	1.027
Express line stops (dummy)	−0.056	−3.9***	0.946	−0.126	−4.1***	0.882
Headway (minutes)	0.001	0.8	1.001	0.012	3.9***	1.012
Number of train vehicles	−4.E-05	−1.3	1.000	−2.E-04	−2.7***	1.000
Proportion of the population aged between 20 and 29	−0.557	−4.1***	0.573	−1.003	−3.4***	0.367
Proportion of the population aged 65 or above	−0.567	−3.5***	0.567	−1.384	−4.0***	0.250
Proportion of manufacturing firms	0.113	1.2	1.120	0.506	2.5**	1.658
Proportion of wholesale and retail businesses	−0.179	−2.0**	0.836	−0.009	0.0	0.991
Proportion of transportation businesses	−0.961	−2.8***	0.383	−0.242	−0.3	0.785
Proportion of accommodation and restaurant businesses	−0.012	−0.1	0.988	0.270	1.3	1.309
Proportion of publishing, broadcasting, and information businesses	0.712	2.1**	2.038	2.428	3.4***	11.338
Proportion of finance, insurance, and real estate businesses	0.343	2.3**	1.410	1.398	4.4***	4.046
Proportion of scientific and technical services, and business support businesses	−0.309	−1.8*	0.734	−0.596	−1.7*	0.551
Proportion of education and public administration services	−0.419	−2.8***	0.658	−0.100	−0.3	0.905
Proportion of health entertainment, culture, and sports-related services	−0.209	−1.0	0.812	−1.008	−2.3**	0.365
CBD dummy (1 if station locates in CBD, 0 else)	−0.246	−3.9***	0.782	−0.295	−2.2**	0.744
Gangnam dummy (1 if station locates in Gangnam, 0 else)	−0.176	−2.8***	0.839	−0.379	−2.9***	0.685
Yeongdeungpo dummy (1 if station locates in Yeongdeungpo, 0 else)	−0.159	−2.6***	0.853	−0.152	−1.2	0.859
Wave 1	−1.091	−63.8***	0.336	na	na	na
Wave 2	−0.858	−50.4***	0.424	na	na	na
Wave 3	−0.916	−53.7***	0.400	na	na	na
Wave 4	na	na	na	−0.101	−3.3***	0.904
Wave 5	na	na	na	0.064	2.1**	1.066
Constant	1.192	14.3***	na	0.881	5.0***	na
ρ	0.668	49.0***	na	0.523	22.3***	na
R 2	0.77			0.43
N	1906			1403

Note: PCA = pedestrian catchment area; na = not applicable.

^***p < 0.01; **p < 0.05; *p < 0.1.

Second, among connectivity variables, the more buses that stopped in a PCA, the greater the number of nodes, and the more the express line stops, the larger the effect of ridership being reduced by the pandemic in both years. Third, among demographic variables, the higher the proportion of the younger population (in their 20s) or older population (65 years and older) within a PCA, the greater the effect on the ridership being reduced. A one-unit increase in the proportions of those in their 20s and 65 years and older contributed to decreasing ridership by 42.7% and 43.3%, respectively, in 2020 and by 63.3% and 75%, in 2021, compared with other age groups. This finding implies that the pandemic affected young and older ridership disproportionately; this may be explained by a reduction in the use of subways by students, as universities were not allowed to open in-person classes from March 2020, as well as a reduction in the outdoor activities of older adults because COVID-19 poses a greater threat to older people. Fourth, four industry-related variables were associated with higher ridership reduction in 2020: (i) wholesale and retail businesses, (ii) transportation-related businesses, (iii) scientific and technical services and business support businesses, and (iv) education and public administration services. This shows that PCAs with a higher proportion of these industries were hit the hardest by the pandemic in 2020. The odds ratios of these variables clarify that a one-unit increase in the proportions of each industry contributed to ridership decreasing by 16.3%, 61.7%, 26.6%, and 34.2%, respectively, compared with other industry groups. However, three of these four industry-related variables are not statistically significant in 2021: wholesale and retail businesses, transportation-related businesses, and education and public administration services. This finding may be related to the recovery of activities for these three industries within the PCA as the pandemic period lengthened, reducing gaps among industry-related activities. Another notable finding on the industry-related variables is that health, entertainment, culture, and sports-related services had negative but insignificant coefficient value in 2020. However, it became statistically significant in 2021, indicating that people were less likely to participate in these activities during the second year of the pandemic. Finally, for stations that are located in the three employment centers, there is likely to be a greater reduction in ridership. Three employment center dummy variables contributed to ridership decreasing by 21.8%, 16.1%, and 14.7% in 2020, respectively, compared with stations in other areas.

Conclusion and Implications

A station-level analysis of the impact of the COVID-19 pandemic on subway ridership in SMA during the two years of the pandemic was conducted. Unequal pandemic effects on station-level ridership were found according to each station, its PCA characteristics, and the pandemic wave and year. First, the subway system was severely disrupted by the pandemic, with significant decreases in ridership—by about 27% during both pandemic years—compared with the last pre-pandemic year. As the number of confirmed cases increased significantly in 2021, strict restrictions on the size of private gatherings and on business operating hours are believed to have contributed to hindering the recovery of subway usage in the second year of the pandemic. Second, in 2020, the ridership reduction was sensitive to the three waves, but in the second year, it became less sensitive to the waves, relative to the non-wave periods. This seems to be related to a change in travel behavior that became less responsive to the pandemic waves, as the high number of confirmed cases persisted during the second year of the pandemic.

Third, although the total reduction in ridership remained at a similar level in 2020 and 2021, the spatial distribution of ridership reduction showed a different pattern when considered by pandemic year and activity type. For example, PCAs with more businesses requiring face-to-face interactions with consumers saw the greatest ridership declines, reflecting a tendency to avoid face-to-face contact during the first year of the pandemic; these businesses include wholesale and retail businesses, transportation-related businesses, and education and public administration services. However, these impacts virtually disappeared in the second year of the pandemic, possibly with the recovery of activities for these industries within the PCA. Furthermore, PCAs with higher numbers of young residents in their 20s and older residents aged 65 years and older experienced greater ridership reductions than those serving other age groups in both pandemic years. In addition, stations located in the employment centers, such as the CBD and Gangnam, had higher ridership reductions in both pandemic years.

These results illuminate the following implications for urban spatial structure, land use, and transportation policies that could help to mitigate pandemic impact. First, the result showing that stations located in employment centers experienced a significant ridership reduction suggests that job decentralization and suburbanization could mitigate the impact of the pandemic on ridership. Jun ( 35 ) argued that SMA has significantly fewer employment centers than those reported for Western cities. Having first identified seven employment centers in SMA (three in the central city and four in the suburbs), these were compared with 32 centers in Los Angeles ( 36 ), 15 centers in Chicago ( 37 ), and 22 centers in the San Francisco Bay Area ( 38 ). Before the pandemic, high job concentration and compact development in employment centers tended to attract urban activities and generate agglomeration economies; however, this location advantage turned out to be detrimental to public health during the pandemic. The government should induce job decentralization through various policies, such as suburban R&D centers and development of industrial parks, as well as tax incentives for companies to relocate from the central city to the suburbs, among other initiatives. These measures could contribute to mitigating Seoul’s infamous subway overcrowding and traffic congestion.

Second, it is necessary to reconsider Seoul’s recent attempts to increase the floor area ratio of residential and commercial facilities by up to 700% near subway stations, to thereby supply more residential and commercial space ( 39 ). As SMA is already one of the densest cities in the world, policy directions should rather aim to decrease the floor area ratio to alleviate overcrowding and provide more open spaces around stations to improve social distancing and inhibit the spread of the virus. It is also important to improve ventilation, air filtering, and sanitation within subway stations and nearby buildings. The findings of this study warn of the health threat of high-density land use development, which is an important principle of station area development worldwide. The lesson from the pandemic is that public health considerations must be added to the existing station area development strategies which tend to maximize the economic efficiency of land use and accessibility to public transit.

Third, providing various safe modal choices is crucial to facilitate citizens’ mobility in the era of the pandemic. It is necessary to disperse the demand for public transportation by providing personal means of transportation such as bicycles, scooters, motorcycles, or electric boards, and by creating street layouts conducive to walking while maintaining social distance.

Finally, the findings that the pandemic dealt a major blow to PCAs with higher proportions of young and older adults, and small businesses requiring in-person contact services, imply that the pandemic exacerbated existing socio-economic inequalities in urban areas, causing socio-economically vulnerable groups (such as the young and older adults, small business owners, the self-employed, those who receive lower pay, as well as temporary and female workers) to focus on basic survival. The central and local governments need to re-evaluate existing welfare measures and create new policies to support these groups during the pandemic.