Metro travel and perceived COVID-19 infection risks: A case study of Hong Kong

Abstract

The COVID-19 pandemic has exerted unprecedented impacts on travel behaviors because of people's increased health precautions and the presence of various COVID-19 containment measures. However, little research has explored whether and how people changed their travel with respect to their perceived local infection risks across space and time. In this article, we relate elasticity and resilience thinking to the changes in metro travel and perceived infection risks at the station or community level over time. Using empirical data from Hong Kong, we measure a metro station's elasticity as the ratio of changes in its average trip length to the COVID-19 cases' footprints around that station. We regard those footprints as a proxy for people's perceived infection risks when making trips to that station. To explore influencing factors on travel in the ups and downs of perceived infection risks, we classify stations based on their elasticity values and examine the association between stations' elasticities and characteristics of stations and their served communities. The findings show that stations varied in elasticity values across space and different surges of the local pandemic. The elasticity of stations can be predicted by socio-demographics and physical attributes of station areas. Stations serving a larger percentage of population with higher education degrees and certain occupations observed more pronounced trip length decrease for the same level of perceived infection risks. The number of parking spaces and retail facilities significantly explained variations in stations' elasticity. The results provide references on crisis management and resilience improvement amid and post COVID-19.

1. Introduction

Tremendous changes in cities have taken place since the COVID-19 pandemic started plaguing the world. As the pandemic persists for years, there has been increased scholarly attention paid to humans' dynamic responses to these changes, the recovery of the economy and social life, and the “new normal” that would dominate the post-pandemic city and long-term urban development. The existing scholarship has considered both the local short-term recovery (Beck & Hensher, 2020) and the path to the future (Forsyth, 2020; Hensher et al., 2021; Pierantoni et al., 2020). In this scholarship, people's travel behavior has been a recurrent topic, which has a lot to do with human activity dynamics and adaption amid COVID-19 (Haraguchi et al., 2022; Schlosser et al., 2020; Wang et al., 2022a). Increased perceived infection risks and non-pharmaceutical measures, such as work-from-home mandates and school closures, have significantly affected people's travel behaviors, which can be measured by indicators such as mode choice and trip frequencies. Notably, a year-on-year reduction rate of over 50 % was once reported in transit ridership in many megacities, e.g., Hong Kong and New York City (Hu & Chen, 2021). To understand the underlying mechanism of travel behaviors amid COVID-19, a body of research on influencing factors of travel during the pandemic has emerged given the presence of perceived infection risks and non-pharmaceutical measures.

At some risk of oversimplifying the overlapping areas across different refereed articles, the existing scholarship on people's travel behaviors and their influencing factors amid COVID-19 can be largely categorized into three streams. Stream 1 focuses on people's travel behaviors, given the presence of increased risks of infection and pandemic containment measures. It is through this kind of research that scholars hope to examine the impacts of different measures on reducing infection risks, which presumably are correlated to how people travel, when, and where and where they meet physically, and for how long (Schlosser et al., 2020; Zhou et al., 2021; Zhou et al., 2022). Models have been developed to predict whether and how the concentration and flow of the population would affect the virus spread (Benzell et al., 2020; Chang et al., 2021b). Stream 2 pays attention to so-called essential services and facilities, and people's trips and accessibility to them before and amid the pandemic. Scholars have advocated changes in transport service operations, accessibility measurement and thresholds, and public policy-making processes to ensure the accessibility to essential services and facilities among people in need (Dong et al., 2021; Kutela et al., 2021; Zhang, 2020). Stream 3 considers the relationship between travel behaviors and communities' or individuals' characteristics. One underlying assumption in this stream is that some communities and individuals can be more vulnerable to the COVID-19 impacts, and inequitable resource allocation among communities can further aggravate the difficult situation and social exclusion of the disadvantaged (Hu & Chen, 2021; Xiao et al., 2022).

Despite the above progress made in the three streams, there remain at least three notable gaps to be filled in the existing scholarship. First, little has been done on whether and how the relationship of travel behaviors and their “legendary” explanatory variables such as built environment has changed during the pandemic. Because of the pandemic, there can be new factors influencing such relationships. Most notably, individuals' risk perceptions on travel and various governmental anti-COVID-19 policies emerged during the pandemic. They could have played a more significant role in travel decisions and can compound the impacts of those legendary explanatory variables on travel behaviors (Ozbilen & Akar, 2023). For instance, people could reduce their travel to where they perceive higher infection risks, even if there are certain facilities or services that used to be so attractive. Second, even though the existing scholarship has disclosed that there is a strong correlation between people's (travel) behavior and (perceived) infection risks (Eisenmann et al., 2021; Rahimi et al., 2021), they rarely considered such a correlation could vary across locales and over time. Third, the existing scholarship has recognized the importance of examining people's behavioral changes across the space longitudinally. However, it rarely considers both people's behavioral changes by subarea within a given city or region and such changes across a long time. The most probable reason underlying this gap is the availability of longitudinal empirical data for scholars to operationalize relevant indicators.

To address these research gaps, we collected longitudinal smartcard data for millions of metro riders in Hong Kong and exploited other publicly available data from sources such as censuses, governments' COVID-19 dashboards, and online map services. We then introduced two concepts: elasticity and resilience, to help us better understand and measure riders' travel behaviors by metro station for different periods amid the pandemic. To quantify perceived infection risks around metro stations, we exploited the COVID-19 dashboards, which recorded venues and trip trajectories, i.e., footprints of locally confirmed COVID-19 cases. By assembling other publicly available data, we constructed a series of probable variables that would, together with people's perceived infection risks, influence riders' behaviors by metro station before and amid the pandemic.

Specifically, we argue that riders to and from one metro station can also be a unit of analysis for elasticity, which is the ratio of changes in two related variables. Riders' travel behaviors can be multidimensional, e.g., trip length, destination choice, and departure and arrival times, and as mentioned, they can be explained by many influencing factors. Elasticity can thus be used to measure how the change in one influencing factor would affect one dimension of riders' behaviors. Prior to the outbreak of COVID-19, perceived infection risks had rarely been considered separately as an influencing factor of riders' behaviors. Smartcard and COVID-19 dashboard data mentioned above has allowed us to better capture heterogeneous changes in both riders' behaviors and perceived infection risks across space (e.g., stations) and time. Elasticity helps us quantify those changes simultaneously using only one variable.

Increasingly, scholars are interested in not only elasticity but also resilience, which concerns how systems can best prepare for, adapt to, and recover from abrupt changes. Metro systems are composed of subsystems, i.e., metro riders, planners, operators, regulators, and various infrastructures and institutions. Their resilience involves a dynamic process that covers different time periods and is embedded in a more complicated urban system (Sajjad et al., 2021; Walker et al., 2004; Zhou et al., 2010). There have been more and more calls for resilience consideration in urban (system) planning and policy (Banai, 2020; Chen et al., 2021; Keenan, 2020). Therefore, it is meaningful for urban scholars to link elasticity and resilience so that elasticity-based studies can be more relevant to urban (especially metro system) resilience planning and policy amid and after COVID-19. Our argument is that, on the one hand, a system must be elastic to be resilient; on the other hand, elasticity and resilience might share a common set of explanatory variables. Therefore, once we know what can help us predict elasticity, we might also know better how to increase resilience, which is often more difficult to be defined and measured than elasticity as it involves so many components of a system.

Bearing the above backdrop in mind, we assembled a series of data for us to empirically examine metro riders' elasticity in Hong Kong, a city where residents highly rely on the metro to travel. Based on this, we then consider resilience of the city. Our data and analysis cover four local COVID-19 case surges during the year 2020. Overall, we attempted to attack the following research questions:

• (1)how can we understand and measure the elasticity of metro riders' behaviors, which have to do with the resilience of the corresponding metro and urban systems?
• (2)whether and how the elasticity of metro riders' behaviors changed across space and time before and amid COVID-19 and across different surges of COVID-19?
• (3)What can predict the changes in the elasticity of metro riders' behaviors across space and time?
• (4)How can answers to Questions (1) to (3) inspire urban/metro system resilience planning and policy?

By addressing the above questions, we hope to make three research contributions. First, we introduce elasticity to examine how metro travel is sensitive to locally perceived infection risks, which can be spatiotemporally heterogeneous. Second, we identify a set of explanatory variables that can help predict spatiotemporal variations of the elasticity of metro travel with respect to locally perceived infection risks. Third, we relate elasticity and its variations to resilience.

In the remainder of this article, Section 2 conducts a literature review on travel behaviors, perceived infection risks, their relationship, and influencing factors. The elasticity is introduced and existing studies on resilience in travel behaviors during the pandemic are also revisited. 3, 4 are our empirical studies of Hong Kong. We measure the elasticity of metro travel by station and examine the resilience of the metro system. Section 3 describes the data collected and used. Section 4 presents the data analysis results. Section 5 discusses and Section 6 concludes. The two sections synthesize our answers to the above questions, discribe policy implications, and outline possible future research directions.

2. Literature review

2.1. Travel behaviors and their influencing factors

Prior to the outbreak of COVID-19, there was a substantial body of literature on what would influence travel behaviors, which can be measured by indicators such as trip frequency, trip length, mode choice, departure and arrival times, and origins-destinations. The influencing factors could help us understand the dynamics of travel behavior across people, places, and times, as well as explore probable social disparity and exclusion in cities (Lucas, 2012). In a nutshell, five types of “legendary” influencing factors of travel behaviors could be found:

• (1)the built environment and land use pattern (Ewing & Cervero, 2010; Hong et al., 2014),
• (2)the transportation network and infrastructures (Meyer, 1999; Tiwari et al., 2016),
• (3)direct or indirect policy strategies (De Vos, 2015; Sharaby & Shiftan, 2012),
• (4)socio-demographic characteristics of communities (Hanson & Hanson, 1981; Tal & Handy, 2010), and
• (5)individual attributes and preferences (Lin et al., 2017; Mokhtarian & Cao, 2008).

Specifically, the built environment and land use, including measures like designs of the physical environment, the number and types of facilities and land use, and functional mix, as well as the transport networks, can shape peoples' mode choice and accessibility to resources, services, and opportunities (Cao et al., 2009; Hadas, 2013; Levinson & Kumar, 1994; Næss, 2015). The relationship between these physical determinants and travel behaviors would be moderated by subjective factors such as individuals' self-selection and travel preferences (Lin et al., 2017). Socioeconomic factors like income, age, gender, race, and education level of individuals or communities have to do with people's economic resources and job-housing choices, which often well predict travel behaviors (Handy, 1996; Stead, 2001; Wang, 2003).

Amid the pandemic, the above factors can still predict (spatiotemporal) variations of travel behaviors. For instance, physical attributes such as diverse, dense, and accessible land use, street designs, and transport facilities of subareas or communities in a city increased people's walking and cycling behaviors during the pandemic (Shaer et al., 2021). Essentially, physical attributes determine the distribution of facilities and services across communities, which impacts whether and to what degree people can get access to what kind of services, in what quantity, and when (Mouratidis, 2021; Mouratidis & Peters, 2022). Not surprisingly, socioeconomic characteristics also still predict travel behaviors across different units of analysis: individual, station, and community (Chang et al., 2021b; Hu & Chen, 2021). Notably, it was more likely for the high-income and high-educated to decrease travel, work remotely, and change travel mode amid the pandemic (Do Lee et al., 2021; Kar et al., 2022; Wang et al., 2022b).

However, as many studies have mentioned or implied, the relationship between travel behavior and its predictors could have changed after the COVID-19's outbreak. Besides, there could be an emerging set of predictors, such as people's perceived infection risks and governments' pandemic containment measures came into play (Beck & Hensher, 2020; Ma & Cao, 2019; Park et al., 2022; Shamshiripour et al., 2020). For instance, white-collar workers would decrease their travel to downtowns where they perceived higher infection risks. Therefore, the attractiveness of downtowns at least temporally changed, and so did its impacts on people's travel to and from there. To better manage cities and sustain people's gains from cities, policymakers must account for the above new situations and adapt if necessary.

2.2. Travel behaviors, perceived infection risks, and elasticity

Like it or not, COVID-19 is not the only health crisis we have ever faced and will face. Prior to COVID-19, authors had found that people's perceived infection risks of infectious diseases like SARS (Brug et al., 2004) and H1N1 flu (Lee et al., 2012) impacted their travel behaviors. During the COVID-19 pandemic, perceptions of virus spread and infection again decreased individuals' willingness to travel (Bohman et al., 2021; Lee et al., 2021) and actual outgoings, e.g., physical meetings and shopping somewhere (Amram et al., 2022; Chou et al., 2020; Zhang et al., 2022a). However, the poor would like to but still had to travel more than the rich amid the pandemic, which resulted in probably more divided patterns of mobility and virus spread across social groups (Bargain & Aminjonov, 2021). Even for the same group, perceived infection risk elasticity of travel demand varied across different trip purposes during the pandemic (Parady et al., 2020). In addition, a large body of research found that changes in travel mode occurred because of the increased fear of infection posed by the pandemic. People tended to reduce collective modes of travel, like public transit and ridesharing services, and prefer private modes of travel, such as driving, biking, and walking amid the pandemic (Bagdatli & Ipek, 2022; Eisenmann et al., 2021; Kim et al., 2021; Rahimi et al., 2021). These can have not only social equity implications but also environmental ones. On the one hand, not everyone can afford to drive and those cannot could have limited choices, which can be smaller in quantity amid COVID-19; on the other hand, those start driving more can use publicly financed roads to produce more pollutions that nobody can avoid.

The relationship between risk perceptions and travel behaviors varied across time and space. Several studies measured such a relationship and produced more empirical evidence on its existence (Qin et al., 2021; Schneider et al., 2021). The locally confirmed cases in a neighborhood and its adjacent areas could also impact people's perception and their travel to and from those areas. The findings have been generated in cities such as Seoul (Lee & Lee, 2022) and Singapore (Janssen & Shapiro, 2021), where there exist COVID-19 dashboards that publicize detailed information about confirmed cases at different geographical levels over time.

There were not many but a growing number of studies on why the impacts of perceived infection risks on travel behaviors varied across time and space. Prior to COVID-19, there had already been some studies on the possible association between community- or individual-level characteristics such as income and availability, diversity, and quality of facilities and services and people's level of health concerns and trip-making decisions (Mehdizadeh et al., 2018; Simpson & Siguaw, 2008). After COVID-19 hit the world, the association tended to become more salient. Notably, people perceived fewer risks when living in more compact, accessible, and walkable neighborhoods (Ozbilen & Akar, 2023). Besides, income and educational levels can well predict the degree of risk perception. Together, they significantly impacted travel (Lee & Lee, 2022; Ozbilen et al., 2021).

However, to our best knowledge, the existing studies rarely considered spatiotemporal heterogeneity of the relationship between travel behaviors and perceived infection risks. This is a research gap to be filled. To address this gap, we can focus on one dimension of travel behaviors and one impact factor for one fixed unit of analysis at one time. This results in the re-introduction of elasticity into the emerging scholarship on the association. Prior to COVID-19, elasticity was once an important variable that scholars used to look at desirable changes and their probable explanatory variables. Maria Kockelman (1997), for instance, examined the elasticities of various travel modes choices with respect to accessibility, land use patterns, and socio-demographic characteristics in San Francisco Bay Area. Guerra et al. (2018) elaborated on how the probability of commuting to work by car changed relative to average changes in urban forms, transit supply, and household income across the 100 largest urban areas in Mexico. Crôtte et al. (2009) investigated the income elasticity of metro travel demand based on empirical data from the Mexico City metro. In this article, we aim to focus on metro travel elasticity with respect to the perceived infection risk of COVID-19. Such elasticity has rarely been considered in the existing scholarship. Plus, metro stations and station areas are commonly seen as units of analysis for urban and transport planning and policy, and metro travel data is readily available by metro station because of the adoption of smartcard technologies. We, therefore, use such units of analysis when operationalizing variables for our ensuing empirical study.

2.3. Changes and resilience of travel behaviors during a crisis

Changes in travel behaviors, as a form of human responses to the pandemic, are indispensable elements of transportation systems. Amid a crisis, greater variations in people's travel behaviors were commonplace in cities (Benzell et al., 2020; Hunter-Jones et al., 2008; Loo & Leung, 2017). During the COVID-19 pandemic, a series of studies have found significant changes in travel behaviors (Schlosser et al., 2020; Zhang et al., 2022b) and travel-related activities, e.g., commuting for working, eating out, and traveling for leisure and social activities (Balbontin et al., 2021; Hamidi et al., 2020; Zhou et al., 2022). To what degree travel changed also varied in space (Chen & Steiner, 2022; Kar et al., 2022; Zhang et al., 2021) and time (Li et al., 2021; Sullivan et al., 2021; Zhang et al., 2021). As a type of public space where people frequently meet and virus transmission occurs, public transport use particularly suffered from the pandemic and decreased significantly (Eisenmann et al., 2021; Vickerman, 2021). A group of studies examined changes in travel behaviors in the metro system to understand who maintained their metro travel, and whether variations in travel change among riders or stations can be related to characteristics of individuals, metro stations, or communities around those stations (Chang et al., 2021a; Jiang & Cai, 2022; Zhang et al., 2021; Zhou et al., 2021).

Beyond changes, there were a limited number of studies on resilience thinking when discussing how the pandemic impacted travel behaviors or, more broadly, human responses in cities (Chen et al., 2021; Keenan, 2020; Xiao et al., 2022). Rather than simply examining to what degree travel behaviors change, the resilience of travel behaviors further emphasizes two aspects. First, resilience describes a dynamic process of human responses in time (Folke, 2006; Walker et al., 2004; Wang & Taylor, 2016; Zhou et al., 2010). It can be short-term, where a transport system could be more resilient if changes in travel behaviors quickly disappear and rebound to the previous status. It can also be permanent adaptation and transformation, where a new equilibrium of different components of the system emerges after the crisis. For pandemic studies and learning, some studies have explored variations in the resilience of travel behaviors of dynamic urban (sub)systems across different periods of the crisis. The changes in travel behaviors reflect instability in human responses to the changing pandemic situation and government orders, e.g., work-from-home and bans on social gatherings (Haraguchi et al., 2022; Thombre & Agarwal, 2021).

Second, urban or transportation system resilience is complex and involves dimensions of tolerable changes and multiple stakeholders (Walker et al., 2004). The resilience of travel behaviors could describe changes in the urban/metro systems, transit stations, communities, and individuals and how they influence the desirable performance or goals of the latter. So far, the citywide resilience of travel behaviors has been conceptualized and measured in the existing studies (Liu et al., 2023; Lu et al., 2022). The resilience of travel behaviors has also been used to examine responses from certain communities and individuals where people travel to and from and their impacts on the performance and goals mentioned above (Hong et al., 2021; Podesta et al., 2021). In particular, the scholarship along this line focused on how public transportation use varied during a crisis. An implicit assumption in the scholarship is that public transportation use would impact the attainment of certain performance and goals, for instance, the accessibility to workplace, amenities, and healthcare (Azolin et al., 2020; Wang et al., 2022a; Xin et al., 2021). So far, however, Xiao et al. (2022) conducted one of the few cases that examined the resilience of travel behaviors by metro stations and its influencing factors over time amid the pandemic. The study measured the impact of COVID-19 on metro ridership by station as the difference between the station's observed ridership under COVID-19 and its predicted ridership if no such crisis occurs. Then, comparing the ridership in a short-term wave of the local pandemic with a long-term wave, a station's resilience is measured as the difference between the COVID-19 impacts on this station in two waves. In their study, however; little has been done on the travel resilience at the metro station level.

Overall, the existing scholarship on the resilience of travel behaviors has shed light on at least two issues. First, travel behaviors varied more significantly over time during a crisis, and such variations are related to (a) emerging factors such as locally confirmed cases and perceived infection risks and (b) “legendary” factors such as built environment and socioeconomic characteristics by different units of analysis. Such situations force us to consider more about the dimensions and extents of tolerable changes in (sub)systems of interest and their achievable performance levels or goals over a longer period of time than ever (Gkiotsalitis & Cats, 2021; Vickerman, 2021). In other words, to ensure the attainment of certain performance levels and goals of these (sub)systems, we must think more about resilience (Kim & Kwan, 2021; Leach et al., 2021).

Second, travel behavior resilience involves spatiotemporally heterogeneous responses in various subsystems. In the metro system, for instance, changes in travel behaviors and corresponding resilience varied across metro stations over time. Understanding those variations and their explanatory variables, such as metro riders' willingness to change and capacities to adapt, can help us better address the well-being of metro riders and related planning and policy implications (Iio et al., 2021; Rojas-Rueda & Morales-Zamora, 2021; Valenzuela-Levi et al., 2021). So far, little has been done at the metro station level over time. This is what we aim to do in the following empirical study.

3. Data and methods

3.1. The study site and the research framework

Hong Kong, one of China's two special administrative regions (SAR), is chosen as the study site. We consider that the changes in metro travel in Hong Kong are cases in point for us to understand the city's responses to the crisis for the following reasons. First, as Hong Kong is a transit-reliant city, the Hong Kong Mass Transit Railway (MTR), the local company responsible for railway operations, operates 96 stations as of 2020. On average, MTR carries 41 % of all the local public transport trips each day. It is thus not surprising that which MTR riders well represent a large proportion of the local population. Second, Hong Kong MTR kept operating during the local pandemic and saw a significant fluctuation in metro ridership served during the four local case surges in 2020 since the first local confirmed case occurred on Jan 23, 2020. The metro travel varied spanning the year of 2020, which reflect many locals' responses to the crisis. Third, the Hong Kong SAR government recorded the detailed locations of confirmed cases' footprints and made the information open to the public on local news and official websites (HKGov, 2021). Hence, the infection risks brought about by the case footprints could be, in principle, well known and perceived by the local citizens. The methods and findings in this article thus might also be transferable to other regions with similar information disclosure, like mainland China.

Our research framework is presented in Fig. 1 . Specifically, by collecting various datasets as input, we defined and calculated the elasticity of metro travel relative to the perceived infection risk at the metro station level. Stations were classified based on their elasticity values. Then, to investigate the variations in elasticity across space and time, we geovisualized the spatiotemporal distribution of classified stations during all four local case surges. Statistical models were fitted to examine the association between different elasticities and their probable influencing factors—perceived infection risks, a “new” factor plus some legendary factors such as characteristics of stations and their served communities. As mentioned above, such association has rarely been considered in the existing scholarship on COVID-19 and travel behaviors.

Fig. 1.

The research framework.

3.2. Data for analysis

Three sets of data were used for analysis: (1) metro smartcard data that records riders' trips between stations. Since we did not have data for Airport Express Line, and two stations, Lo Wu and Lok Mau Chau Stations, had been closed since Feb 4, 2020, 88 metro stations would be used for analysis; (2) the local COVID-19 information, including the daily confirmed cases and locational data related to footprints of confirmed cases, acquired from the Hong Kong public coronavirus geodatabase (HKGov, 2021). (3) Data that describes certain characteristics of the station's served communities, including socio-demographic data acquired from the 2011 Hong Kong Census, and built environment data acquired from the Google Map API in 2017/2018 and the OpenStreetMap (OSM) data in 2020.

In this article, we measured the elasticity of metro travel in four case surges spanning the year 2020. To fathom changes caused by COVID-19, we selected one baseline week and one comparative week for each surge (Fig. 2 ). On the one hand, controlling confounding effects of the holiday for Spring Festival, the baseline weeks reflected normal travel patterns of local metro riders pre-pandemic or when the case increase remained low before a surge occurred. On the other hand, the comparable weeks recorded metro riders' travel when the case increased and approached the zenith of a certain surge.

Fig. 2.

Four local case surges of year 2020 and the weeks selected for elasticity measurement.

To measure the perceived infection risks in cities, we acquired the footprints of confirmed cases emerging between the baseline week and the comparative week of each surge. For instance, we collected the confirmed cases and venues visited by these cases between Jan 6 and Feb 9 for the first surge. The same venues visited by n cases would be counted n times, with a multiplier effect on the infection risk considered (samples as shown in Table 1 ).

Table 1.

Sample of footprints of confirmed cases.

Case ID	Date	Venues visited	Within 800 m of a metro station
1	01/23/2020	Empire Hotel Kowloon-Tsim Sha Tsui	Jordan; Hung Hom; Tsim Sha Tsui; East Tsim Sha Tsui
1	01/23/2020	Hong Kong West Kowloon Station	Austin; Kowloon; Jordan
2	01/23/2020	Heng On Estate, Ma On Shan, Hong Kong	Ma On Shan
…	…	…	…
43	02/11/2020	Jubilant Grace Methodist Church, Siu Sai Wan	None
…	…	…	…
95	02/29/2020	Bauhinia Garden (Estate), Tseung Kwan O	Tseung Kwan O
…	…	…	…

3.3. Elasticity of metro travel to the local pandemic

Amid COVID-19, metro travel behaviors were affected by the pandemic progress, which could be later translated into individuals' perceptions on infection risks and responses to the abrupt changes. By making each metro station as the analytical unit, we used Eq. (1) to measure the elasticity (e) of metro travel behaviors to the perceived inflection risks by station.

$$e=\%\mathrm{\Delta F}/\%\mathrm{\Delta R}$$ (1)

$\%$ΔF is the change in metro travel behaviors;

$\%$ΔR is the change in perceived infection risk.

Eq. (1) describes how riders choose and travel to certain stations as the destinations. A place would have higher risks if it involves more case footprints (Kan et al., 2021). A primary assumption here is that metro riders, especially long-distance riders, would particularly decrease their trips to a certain station (destination) if they perceived a higher infection risk at and around the station. In this article, we measured %ΔR by considering the number of case footprints within 800 m of metro stations, the travel distance between one metro station and the other, and the mean latency period of a confirmed case. %ΔF was measured by the total incoming trip length ex ante the pandemic and that ex post. Specifically, the equations used to calculate $\mathrm{\Delta F}$ and $\mathrm{\Delta R}$ are as follows.

Total incoming trip length of Station i ($A_i$), which reflects to what degree a station, as a destination, could attract a large quantity of long-distance metro trips.

$$A_i=\sum _{j=1}^{n-1}\left(t_{\mathit{ji}}\ast l_{\mathit{ji}}\right)$$ (2)

where, $t_{\mathit{ji}}$ is the number of incoming trips from Station j to the measured Station i, j<>i and n is the total number of stops/stations in the local transit/metro system; $l_{\mathit{ji}}$ is the network distance between Station j and Station i.

Perceived infection risk at Station i ($R_i$), which reflects to what degree riders could perceived infection risks at and around stations (destinations). It is calculated as:

$$R_i=\left(c_i+{\sum }_{j=1}^{n-1}\frac{c_j}{l_{\mathit{ji}}}\right)/\mathrm{\delta }$$ (3)

where, $c_i$ is the number of case footprints within 800 m of Station i per day between the baseline and the comparative week of each surge, j<>i and n is the total number of stations in the local transit/metro system; $c_j$ is the number of case footprints within 800 m of Station j per day. Hence, $c_j$/$l_{\mathit{ji}}$ describes to what degree case footprints at Station j would impact the perceived infection risks at Station i. The constant δ represents the mean latency period, which might be inversely proportional to the perceived infection risk. We estimated δ = 96 h based on previous studies (Chang et al., 2021b). Roughly speaking, Eq. (3) indicates that a centrally located station with lots of recent case footprints would have higher perceived infection risks.

Elasticity of Station i ($E_i$). The sets of elasticity in the empirical study reflect how total incoming trip length would change by station in relation to the perceived infection risk at the destination. $A_i^0$ is the total incoming trip length in Station i ex post of a certain case surge, and $A_i^1$ is the total incoming trip length in Station i ex ante. $E_i$ would be calculated as follows:

$$E_i=\left(1/R_i\right)\ast \left(A_i^0-A_i^1\right)/A_i^1$$ (4)

To better understand the role of different stations amid the pandemic, we classified them into three groups based on their respective elasticity values: the “luxury”, “necessary”, and “inferior” stations. The idea is based on the typical classification of goods in economics by examining how the demand elasticity of the number of goods (or services) consumed by an individual would change with income change (Kemp, 1998). Similar classifications have also been applied in many other studies to explain elasticity between variables (Freeman, 2003; Melo et al., 2019; Vargas-Lopez et al., 2022). In this article, the “inferior” stations are defined as those with an elasticity value larger than 0, which means the increase in perceived infection risk amid the pandemic is accompanied by increase in trip length to this station. For those with elasticity smaller than 0, we conduct the k-means clustering to identify the relative “luxury” stations and the relative “necessary” stations. The clustering method could divide the measured stations into two groups by minimizing their within-cluster variances in elasticity values. Hence, the relative “necessary” station will be those with absolute elasticity values smaller than the relative “luxury” stations. That is to say, one unit increase in perceived infection risk of the relative “luxury” stations will lead to more decrease in metro travel than that of the “necessary” stations.

3.4. A binary regression model

As mentioned above, the changes in the trip lengths and perceived infection risks by station amid COVID-19 would be jointly affected by multiple sets of influencing factors. When examining the elasticity of metro travel to perceived infection risks in this, it is essential to investigate the influencing factors of stations' elasticity to understand why the elasticity varied in time and space. In this article, we initially investigated whether community characteristics could vary by station-level classification based on elasticity value. We categorized stations by the value of their probable influencing factors and fitted binary logistic regression models for station elasticity classification in each of the four surges and across surges.

Various local publicly available data were used to quantify probable influencing factors by station. Table 2 shows descriptive statistics of all the variables. In light of the literature reviewed and local data availability, three sets of variables were formulated to measure the factors that probably affect the station-level elasticity: (1) Socio-demographic characteristics of the local Tertiary Planning Unit (TPU) that a station is located at. The social and economic characteristics of metro riders would be associated with their travel patterns when perceived infection risk increased. (2) Built environment within 800 m by station. We hypothesize that people still need to utilize some facilities in some venues around stations despite the pandemic. (3) The centrality degree of each MTR station based on the January 2020 Octopus (smartcard) data and the MTR network. MTR stations' centrality measures the level of metro-based accessibility to different facilities and opportunities (Zhou et al., 2021).

Table 2.

Descriptive statistics of the variables for community characteristics.

Types	Variables	Description	Mean (st. dev)
Socio-demographic characteristics	Income	Income median (HK dollars)	14,539.55 (5736.34)
	Age	Age median (years old)	41.98 (2.57)
	Education level	Percentage of population with postsecondary degrees (%)	30.20 (12.52)
	Employment	Percentage of working population (%)	51.51 (8.75)
	Transport usage	Percentage of population using metro for commuting (%)	45.17 (19.82)
	Household size	Average number of household members (persons)	2.90 (0.32)
	Occupation	Managers and administrators	14.53 (9.93)
		Professionals/associate professionals	28.69 (6.36)
		Clerical support, service, and sales workers	27.62 (10.54)
		Craft and related workers	9.38 (5.43)
Built environment	POIs	Number of POIs within 800 m	2,682.19 (2,240.53)
	Retails	Number of retail facilities within 800 m	48.72 (117.11)
	Intersections	Number of intersections within 800 m	582.49 (345.89)
	Bus stops	Number of Bus stops within 800 m	98.17 (47.56)
	Parking space	Number of parking spaces within 800 m	49.76 (37.38)
Centrality degree of a station	Average travel time to other stations	Average travel time to any other station in the metro network (min)	39.42 (7.67)
	Population within 15 mins' travel	Population coverage within 15 min in the metro network	41,606.45 (15,129.42)

4. Results

4.1. Elasticity comparison in four surges

Based on the abovementioned data and methods, the elasticity is measured by station in each of the four pandemic surges. Table 3 shows the descriptive statistical results. Decreasing incoming trip lengths can be observed in each surge. Some riders did reduce or avoid metro travel after a surge occurred. The perceived infection risks increased as the pandemic progressed as a result of increasing case-related footprints in a later surge than an earlier one. The absolute values of stations' elasticity in a later local case surge were smaller than that in an earlier surge. This indicates that due to people's adaptation and increasing countermeasures implemented as the pandemic progressed, metro riders had been less sensitive to the occurrence of a new surge, even though the perceived infection risk increased.

Table 3.

Descriptive statistics of indicators of stations in four surges.

Surge	R	$A^0$	$A^1$	E (Abs.)
Surge 1	0.01 (0.01)	174, 062 (141,786)	346, 583 (294, 573)	109.01 (72.44)
Surge 2	0.02 (0.01)	182,396 (146,890)	197,379 (161,369)	5.80 (4.69)
Surge 3	0.11 (0.05)	170,791 (133,476)	270,531 (222,314)	4.18 (2.61)
Surge 4	0.33 (0.20)	271,060 (219,494)	295,210 (234,617)	0.37 (0.30)

To measure the distribution of elasticity across metro stations in each surge, the Kernel Density Estimate (KDE) was applied. For each surge, the elasticity values were standardized to control the variations in elasticity across surges. Fig. 3 shows that the distributions of elasticity among metro stations were the most dispersed in the second surge (Surge 2) and then in the third surge (Surge 3). A larger proportion of stations in these two surges had a relatively low elasticity value. A probable explanation is that after the local pandemic outbroke, a series of countermeasures had been implemented since then, e.g., bans on traveling and gatherings, suspension of face-to-face school classes, and closure of some non-essential businesses. These countermeasures had a citywide influence on various stations.

Fig. 3.

Kernel Density Estimate (KDE) of station elasticity in four surges.

4.2. The classified metro stations on elasticity

In terms of the k-means clustering methods for elasticity value, we categorized stations by elasticity classification (i.e., necessary vs. luxury) in each surge. Since all elasticity values were smaller than 0, no “inferior” station could be found. Fig. 4 geovisualizes the spatial variations of the classification, and the bubble size reflects the standardized elasticity value in each surge. The bubble is larger as the absolute value of elasticity is larger.

Fig. 4.

The classified metro stations in four surges.

The classified stations and elasticity values dispersed and varied across space and surges. In the first surge, the trip lengths of Sheung Shui Station (75), as one of the stations serving riders traveling to and from cross-border checking points, were suppressed the most relative to its perceived risk. This might be affected by the closure of checking points between mainland China and Hong Kong on Feb 4, 2020. Sheung Shui is the last metro station where riders can choose to go to either of the checking points. In addition, the stations located in one of city centers, Kwun Tong Station (15), and the well-known recreational center, Tsim Sha Tsui Station (3), were also luxury stations in this surge.

During the second surge, the luxury stations could be particularly found around areas serving tourists and local visitors, such as Tung Chung Station (43), the station near the airport, Sunny Bay Station (54) and Ocean Park Station (86), the two serving theme parks, as well as Tsim Sha Tsui Station (3), Admiralty (2) and Central (1), stations serving business and recreational centers. In this period, the local government issued restrictions on international travel and bans on gatherings in the public space, which reduced both international/regional tourists and local visitors traveling to some recreational places. It seems that the elasticity values of trip length of some stations were particularly affected by these countermeasures regardless of the case increase.

Overall, the distributions of the “luxury” stations in the third and fourth surges were similar despite a few exceptions. As adjunct local residential and recreational centers, Ma On Shan Station (102) and Heng On Station (101) had obviously large absolute elasticity values in the third surge. A notable decrease in trip length relative to perceived infection risks could also be found in the city or local centers like Tsim Sha Tsui Station (3), Kowloon Bay (13), and Kwun Tong (15). It is probable that when the daily new confirmed case reached the peak of the year in late July, people became more sensitive to perceived infection brought about by the case footprints and reduced more metro trips to areas where previous physical traveling and meetings were concentrated. The prohibition on “dine-in” and compulsory mandates on mask-wearing during this period might also play a role. In the fourth surge, despite similar distributions as the previous surges, the theme park at Ocean Park Station (86) temporarily suspended service during this period, which made the station particularly lose riders. Interestingly, the recreational center, Tsim Sha Tsui Station (3), was no longer the luxury despite the fact that even more confirmed cases and strict countermeasures were found in this surge.

To identify station classifications across surges, the time-series k-means clustering was used for analysis. Fig. 5 shows that 18 luxury stations kept relatively high absolute elasticity values among surges of the local pandemic, and 70 necessary stations remained stable in trip length relative to their perceived infection risk. The luxury stations spanning the four surges include those serving city or local centers, e.g., Kwun Tong (15) and Kowloon Bay (13), well-known recreational districts of the city, e.g., Tsim Sha Tsui Station (3) and Admiralty (2), theme parks, e.g., Ocean Park Station (86). In particular, we found some stations serving as local residential centers remain as the luxury, e.g., Sai Yin Pun (81), Kennedy Town (83), Fortress Hill (30), and Tiu Keng Leng (49). They are typical areas where a large number of commuters live in and are surrounded by self-reliant life facilities and services,

Fig. 5.

The time-series clustering for station classification

4.3. Modeling stations' elasticity classification and influencing factors

Table 4 shows the modeling results for how station classification would be associated with influencing factors. Overall, two factors can significantly predict the station classification for all the periods (p < 0,05). First, stations with a higher percentage of workers taking the metro for commuting were more likely to be necessary stations. This was consistent with Zhang et al. (2021), which found that changes in metro travel behaviors could lag behind the variations in COVID-19 transmission. Or one can say that metro riders' behaviors could be path dependent. Second, luxury stations would relatively have a higher degree of centrality in the local metro network, that is, more population covered within 15 min's travel. Amid COVID-19, metro riders perceived higher infection risks in areas with higher population density and reduced their trips to these areas subsequently.

Table 4.

Results of binary logistic regression models.

	Surge 1	Surge 2	Surge 3	Surge 4	Across surges
Socio-demographic
Income median	1.001	1.001	1.001	1.001	1.001
Age median	0.653	0.616⁎	0.726	0.992	0.675
Postsecondary degrees (%)	3.082E+32⁎⁎	4.148E+15⁎	2.123E+33⁎⁎	2.483E+35⁎⁎	4.162E+20⁎⁎
Working population (%)	0.000⁎	0.000	0.000⁎	0.070	0.000⁎
Metro for commuting (%)	0.912⁎⁎	0.931⁎⁎	0.925⁎⁎	0.957	0.940⁎⁎
Average household size	183.598	23.073	117.159⁎	515.746⁎⁎	15.958
Managers and administrators	6.731E+32⁎	3.798E+35⁎⁎	1.108E+30⁎⁎	0.351	1.825E+28⁎⁎
Professionals/associate professionals	0.000	227,781.016	0.000	0.000⁎	0.000
Clerical support, service and sales workers	6.903E+16	1.193E+14⁎	1.214E+20⁎	5.226E+10	3.092E+16⁎
Craft and related workers	4.415E+65⁎⁎	1.358E+51⁎⁎	8.177E+54⁎⁎	3.627E+30	4.647E+35⁎
Built environment
POIs	1.001	1.001	1.001	0.999	1.001
Intersections	0.995⁎	0.995⁎⁎	0.996⁎	0.997⁎	0.998
Retails	0.988⁎	0.985⁎⁎	0.987⁎⁎	0.990	0.987⁎⁎
Bus stops	0.981	0.993	0.980	0.988	0.993
Parking space	1.080⁎⁎	1.058⁎⁎	1.069⁎⁎	1.078⁎⁎	1.029
Centrality degree
Average travel time to other stations	1.358⁎	1.145	1.353⁎	1.430⁎⁎	1.250⁎
Population within 15 min's travel	1.001⁎⁎	1.001⁎⁎⁎	1.001⁎⁎	1.001⁎⁎	1.001⁎⁎
Omnibus test	0.001	0.000	0.002	0.003	0.006

Ref: necessary stations.

OR (Exp B) is reported.

Omnibus test is used to test the model fit. A significant value less than 0.05 could mean the model fits the data well.

^⁎⁎⁎p < 0.01.

^⁎⁎0.01 < p < 0.05.

^⁎0.05 < p < 0.1.

Several other factors' impacts were stable in at least three surges. As for socio-demographic characteristics, the results show that the luxury stations in this surge were located in communities with a higher percentage of the highly-educated population, managers/administrators, and craft workers. Working-from-home mandates decreased the number of metro commuters. It was these cohorts that have more capacity to shift to other travel modes, e.g., driving or telecommuting. Besides, it was more likely for stations with a larger average household size to become luxury stations in the third and fourth surges. Households with multiple members would include vulnerable populations like the young and old, who in principle were more sensitive to the same level of infection risk.

Concerning built environment attributes, stations surrounded by more parking spaces were more probable to be luxury stations. Such facilities provided more opportunities for people to drive. The more retail facilities there were at or around a station, the less likely that the station was “luxury”. This might point to the fact that trips to some retail facilities can be essential amid COVID-19.

5. Discussion

The outbreak of COVID-19 rekindled widespread interest in both urban resilience and resilience of subcomponents of cities, e.g., metro riders, transit stations, and their served communities; that is, how the urban and transport (sub)systems better adapt to and recover from the dynamic crises in cities. Given the unprecedented impacts of COVID-19 on travel behaviors, new theories, methods, and metrics are needed for us to better manage cities and transit systems to adapt to the “new normal” amid and after COVID-19. The article adds new academic knowledge to the existing literature by introducing elasticity and resilience to consider how metro travel changed relative to the perceived infection risks across space and time in the city. The findings from the empirical study in Hong Kong deliver some transferable insights for policy-making and urban/transport planning strategies.

First, the elasticity of metro travel to perceived infection risks reflects which riders from which metro station changed more to perceived infection risks brought about by case footprints and how such relationships varied over time. Indeed, the elasticity varied across stations (space) and riders. This requires metro operators to launch customized rather than universal policies amid the pandemic. For instance, stations with a large trip length and remaining as “necessary” but high risks might need more frequent disinfection. In addition, peak-hour pricing could be introduced to stagger arrival trips to curb the virus transmission. Furthermore, the elasticity varied across local caseload surges. As the pandemic progressed, however, metro riders had been less sensitive to the same level of perceived infection risks in the later surges. This can mean that a new balance between the pandemic containment and normalization of metro operations must be sought after.

Second, we identified which characteristics of stations and their served communities could be associated with the stations' elasticity classification. Subsequently, we can consider how these findings can inspire urban/metro system resilience planning and management. Notably, the findings confirm that the disparity in stations' elasticities can be predicted by variations in riders' willingness and capacity to reduce their long-distance travel when perceiving increased infection risks. This is in line with what existing studies such as (Hu & Chen, 2021; Lee et al., 2021; Rahimi et al., 2021; Xiao et al., 2022) have found. The local government could consider these cohorts or certain communities with higher infection risk and provide more assistance like stringent social distancing rules and fast testing kits. Financial support could also be introduced to those station areas that saw larger decreased ridership and merchandise sales because of the pandemic.

On the other hand, the local businesses at and around luxury stations with a significant loss in trip length could particularly struggle with the crisis (Deslatte et al., 2020). It was more likely to for luxury stations have a higher population density. Businesses around these stations could have a stable customer base in pre-pandemic time but suffered more from case surges. It is suggested that the local government issue dedicated policies on business suspension, such as only closing businesses in high-risk areas for a given time. A flexible flow of funding for local governments is advocated to preserve essential local services and rebuild local economies in different areas of the city for long-term recovery and resilience.

Third, the association between stations' built-environment features and their elasticity classification provides insights into future resilience planning and development around metro stations. For instance, the luxury stations had to do with abundant parking space around. The results are in line with use (Bagdatli & Ipek, 2022; Rahimi et al., 2021), which found that people transferred from public transit to more private car amid the pandemic. Hence, it is meaningful to consider how multimodalism can help have people's essential travel and life demand met amid and post COVID-19 across different locales in our cities.

Limitations of this research should also be mentioned. First, we only examined the elasticity of metro travel to perceived infection risks. How travel in other transport modes and regions outside the station areas responded to the distribution of case footprints remains unknown. Second, we calculated the perceived infection risks based on the locally confirmed COVID-19 case footprints at and around stations. However, to what degree people truly perceived COVID-19 infection risks and changed their travel behaviors could be related to many other factors, for instance, the crowdedness of a travel route and people's occupational requirements. Third, the secondary data concerning station area features that we collected and used was not as current as the smartcard data we managed to get access to. In the future, all the above should be better considered and addressed.

6. Conclusion

Having identified that there are deficiencies in the existing scholarship concerning changes in travel behavior across subareas in a city, we collected empirical data in Hong Kong to first measure the elasticity of metro travel to metro riders' perceived infection risks by station. We then examined whether and how characteristics of stations and their served communities can predict the elasticity variations across station. Based on the above, we produced the following findings, some of which might be transferable to other cities:

(1) the elasticity of metro travel by station did vary across space and local case surges. (2) stations' elasticity can be predicted by the socio-demographic characteristics of station-served communities as well as the degree of centrality of a station and the built environment attributes of this station's adjacencies. The above together indicate that station-specific and group-based travel demand management should be adopted amid COVID-19. Plus, given the time-varying nature of metro travel amid COVID-19, public policies on metro travel management and metro operations should be revisited from time to time. This might also mean data measuring continuous and dynamic changes should be input for related policymaking. In our empirical study, we have utilized passively collected smartcard data to characterize metro travel behaviors and their continuous changes amid COVID-19. Compared to other data from sources such as censuses and travel surveys, smartcard data is continuously collected and updated and covers a much larger sample. This has enabled more longitudinal and (sub)group-based analyses of metro riders. Corresponding results can serve as better references for decision-making amid COVID-19.

CRediT authorship contribution statement

Mingzhi Zhou: Conceptualization, Methodology, Data curation, Formal analysis, Visualization, Writing – original draft, Writing – review & editing. Hanxi Ma: Methodology, Data curation, Formal analysis, Visualization. Jiangyue Wu: Methodology, Data curation, Formal analysis, Visualization. Jiangping Zhou: Conceptualization, Methodology, Formal analysis, Writing – original draft, Writing – review & editing.

Declaration of competing interest

The authors have no competing interests to declare.

Data availability

The authors do not have permission to share data.