Structural change and spatial pattern of intentional travel groups: A case study of metro riders in Hong Kong

Abstract

Amid the COVID-19 pandemic, face-to-face contacts decreased but still existed despite people's fear of virus infection and governments' social gathering restrictions. These interactions influenced virus transmission routes, if any and reflected people's essential social interactive demands in the city. In this article, we identified people who intentionally travel as groups (ITGs) to characterize social interactions before and amid COVID-19. To systematically understand ITGs' mobility patterns, an ITG structure was defined and measured in multiple dimensions, including composition, function, size, intensity, quality, and spatiotemporal distribution. Based on a longitudinal smartcard dataset in Hong Kong spanning the year of 2020, we operationalized the ITG structure in the local metro system and examined whether and to what degree the structure changed during the pandemic. We found that ITGs' activities fluctuated as the pandemic progressed and their changes differed across different ITG groups. The long-distance ITGs saw the most significant change. The spatial distribution of persistent ITG trips before and amid the pandemic became spatiotemporally more concentrated. Stations with similar ITG indices clustered in proximity, and features of station areas like residents' education level and quantity of commercial facilities could well predict stations' ITG indices. In other words, inequal distribution of essential facilities and opportunities could notably influence ITGs, social contacts, and socioeconomic benefits brought about by them amid COVID-19. The findings provide insights concerning both resilience management amid the crisis and the long-term planning of essential facilities and services that facilitate group-based outgoings and activities.

1. Introduction

The COVID-19 virus is primarily transmitted through physical contacts and proximity among people (Liu et al., 2020; WHO, 2020). Thus, we must understand people's daily social interactive networks to reduce the risk of COVID-19 contagion. To fight against large-scale COVID-19 contagion, governments across the globe have implemented restrictions on social interactions and gatherings. These countermeasures largely suppressed physical social interactions and thus had mitigated the spread of COVID-19 (Chang et al., 2021; Flaxman et al., 2020; Zhou et al., 2021). Nevertheless, they have also imposed exogenous shock on societies; for instance, economic stagnation due to decrease d human activities, financial pressures on the poor, and increased depression among people (Bonaccorsi et al., 2020; Lazarus et al., 2020; Yang & Xiang, 2021). To tackle these problems whereas containing the virus spread, people have increasingly called for profound understanding of the (emerging) mechanisms governing human activities and heterogeneous distribution of opportunities and health risks amid and post COVID-19. The understanding could help us devise more tailor-made rather than universal social restrictions in our fight against or co-existence with the pandemic (Benzell et al., 2020).

The existing scholarship has examined the heterogeneity in human activities during the pandemic, including individuals' mobility behaviors and their collective mobility patterns (Schlosser et al., 2020; Sullivan et al., 2021). However, to our best knowledge, little has been done on group activities involving two or more people. These activities inform us about how people collectively reacted to the pandemic. In previous studies, social relations and contacts have been explored in different disciplines such as Sociology, Anthropology, Epidemiology, and Psychology. Using data from traditional sources, e.g., surveys, observations, and interviews, scholars have examined how human interactions were associated with the physical environment and individuals' behavioral choices, health, and well-being (Antonucci et al., 2014; Milgram, 1972). However, related studies often covered only a small sample, which could hardly depict the citywide social interactions given the widespread and lasting impacts of the pandemic on all walks of life. In disciplines such as Geography and Urban Studies, (big) data from emerging sources, e.g., Wi-Fi data, social media data, and smartcard data, have been used to broaden and deepen the above studies. A wider range of social interactions in both virtual and physical worlds have been explored subsequently (Gross & Acquisti, 2005; Liang et al., 2016; Sun et al., 2013; Zhou et al., 2020). Notably, Calabrese et al. (2011) investigated social relations using cellphone call records. Zhu et al. (2021) and Zhou et al. (2022) centered on group travel behaviors based on metro smartcard data. Besides strong ties such as friends and family, more weak or potential social ties have been identified, e.g., encounters and familiar strangers (Sun et al., 2013; Zhou et al., 2020). However, in the existing literature that we have collected and reviewed, little have been done on the group activities’ characteristics and their measurement.

In this article, we probe into potential group activities by identifying intentional travel groups (ITGs) among metro riders. ITGs are identified within the metro network in Hong Kong, a city internationally known for its high transit reliance. Since mobility is a typical prerequisite of physical group activities, we consider the metro and related facilities like carriages and platforms where people frequently meet. The travel of ITGs and their structural characteristics could help us fathom into the group behaviors and patterns on the city level. This is especially the case in transit-reliant cities, such as Hong Kong, Tokyo, and London, where a large proportion of population go about using transit. We trace and characterize how people traveled and met as groups before and amid the pandemic. We regard ITGs’ travel during the pandemic is more than the fact that people knowing one another hang out together. We treat it also as a subset of social relations and essential interactive activities. By looking at where ITGs occur, we also know better venues that people value amid the pandemic. In short, ITGs amid the pandemic offer us new lenses to know better how different people reacted to the pandemic, how they forewent or valued certain venues amid the pandemic, and whether and to what degree, their relative spatial locations influenced their group-based outgoings.

Our contributions lie in three aspects. First, we propose the concept of ITG structure. This concept provides an analytical framework to characterize the regularity of group travel activities (and the lack of it) amid the pandemic. Second, we relate the spatiotemporal patterns of ITGs to the probable virus transmission routes and locations with higher infection risks. Third, we examine the correlation between stations’ ITG structural changes and station-level sociodemographic and built-environment attributes. This allows us to better understand if group travel varied across locations and social groups. The findings are expected to inspire more relevant anti-pandemic countermeasures that balance the control of pandemic spread and upkeep of social well-being.

The remainder of the article is organized as follows. Section 2 defines the ITG structure, and highlights the ITG's significance, which can be more salient amid the pandemic. Section 3 introduces the data and methods for an empirical study of ITGs in the Hong Kong metro system. It describes how to identify ITGs, measure the ITG structure, and explore the ITG's structural changes as well as geographical distribution. Section 4 presents the empirical results and Section 5 concludes.

2. ITG structure and significance

2.1. Defining ITG structure

In this study, we propose “the ITG structure” for us to systematically explain and quantify intentional group travel patterns. Specifically, the structure consists of the following characteristics: composition, function, size, intensity, quality, and spatiotemporal distribution.

Composition: Composition describes ITG members' attributes and the relationship among them. On the one hand, differences in sociodemographic attributes indicate the disparity in social relations and connections (Ajrouch et al., 2005; Antonucci et al., 2014; McPherson et al., 1992). Attributes like age, sex, and race shape dynamic social interactions among individuals. On the other hand, the question of “with whom” is an essential social dimension in travel behavior (Carrasco et al., 2008). Members in (strong-tie) social relations typically are spouses, children, siblings, other families, close friends, etc. (Antonucci et al., 2014). There are, however, weak-tie social relationships, namely, familiar stranger, in-role, friend, and stranger (Liang et al., 2016). The composition of a person's social network, e.g., weak ties vs. strong ties, influences her/his well-being and behaviors. In this article, the ITG structure considers both ITG members' sociodemographic attributes and the relationships between the ITG members.

Function: The functions of ITGs involve two aspects: (1) direct purposes fulfilled because of an ITG trip; (2) indirect benefits ITG members could get from the ITG trip. The functions explain the underlying motives and significance of intentional group travel, which presumably sustains intimacy, improves productivity, and influences individuals’ health and well-being (Dimmock et al., 2022; Hynes, 2016). For instance, why do people travel as a group? Zhu et al. (2021) found people travel together for the purposes of commuting, business, and leisure. What kind of benefits can ITG members achieve? We can get some inspiration from the existing studies. Notably, Antonucci et al. (2014) regarded functions of social relations as aid, affect, and affirmation exchanges. Harrington and Emanuel (2020) found that office workers have more opportunities to learn from their fellows and improve the quality of work through office meetings. In light of these, we speculate that ITGs help members achieve some purpose together, consitute a subset of social relations, and promote mutual learning.

Size: Size is the number of members or subgroups within the same ITG. It affects the magnitude and formats of possible social interactions because of the presence of ITGs. Both data from traditional and emerging sources can help us estimate the size. Based on surveys and interviews, some studies have identified social network sizes by asking individuals in a given social network to enumerate people who are important to them (Cornwell et al., 2009). Using smartcard data, Zhou et al. (2020) calculated and visualized the number of familiar strangers by station in the metro system of Beijing.

Intensity: ITG intensity is the ratio of the ITG trips to all the other trips that the ITG members made. It shows the relative importance of physical group travel interactions, which represent a subset of daily social contacts that riders make. Contact frequencies are typically impacted by interactive modes, relationships, and geographic distance (Calastri et al., 2017; Van den Berg et al., 2012). In existing studies, weak-tie social contacts have been paid more attention to. Notably, Milgram (1972) calculated the ratio of familiar strangers to railway users from a particular station in a New York's suburb using survey data. In a similar vein, Zhou et al. (2020) calculated the ratios of familiar strangers to railway users by metro station in Beijing using smartcard data. Liang et al. (2016) found that familiar strangers were socioeconomically similar and encountered one another more frequently, and enjoyed more stable network than strangers.

Quality: The ITG quality measures to what degree ITG members can affect one another or the attractiveness of a location where ITG trips occur. In existing studies, the quality, characterized by positive or negative social interactions, is considered predictive of and influential on people's long-lasting life experiences, e.g., physical and mental health (Birditt et al., 2009). Relationship quality might not be positively associated with contact frequency but can contribute to effects of interactions, e.g., closeness and satisfaction (Stavrova & Ren, 2021). Leisure activities like meeting “friends” or “relatives”, for instance, entail the longest distances traveled (Schlich et al., 2004). In this article, we use individuals or locations' attributes as proxies of the ITG quality.

Spatiotemporal distribution: Spatiotemporal distribution characterizes how ITG trips vary across space and time. Spatiotemporal patterns are essential characteristics when understanding dynamic social interactions. Ajrouch et al. (2005) regarded geographical proximity as a measurement of social relations and found that professional occupations had the least geographically proximal social networks. Zhou et al. (2020) identified familiar strangers through their co-presences in time and locations. More generally, the probabilities and intensities of social contacts, including ITG trips, are impacted by urban forms (especially the distribution of jobs and homes), one's daily time budget and activity schedules (Carrasco et al., 2008). Farber et al. (2013), for instance, examined such probabilities using empirical data from Salt Lake City, Utah.

2.2. Significance of ITG structure

As a special form of social interaction or a prerequisite for other social activities, intentional group travel involves meaningful coordination, collaborations, and innovations among members. Within the context of COVID-19, understanding these group activities, characterized by an ITG structure, could generate insights on both ad-hoc anti-pandemic countermeasures and long-term public policies.

On the one hand, as the COVID-19 virus is primarily transmitted through physical proximity and contacts among people, the ITG structure measures the regularity and distribution of an important subset of face-to-face interactions in cities. Other things being equal, it is possible that members in recurrent ITGs get infected and/or to transmit virus more easily. Should we know where they go most often and when, we could better predict and monitor the COVID-19's infection risk (Benzell et al., 2020; Chang et al., 2021). In other words, the ITG structure can help us identify locations possibly contributing more to the spread of COVID-19 virus.

On the other hand, the spatiotemporal distributions of ITG members or their physical presences at different destinations epitomize varying demands for social interactions and related locales and facilities pre- and amid the COVID-19 pandemic. Impacted by the fear of infection and restrictions on social gatherings, the persistent group travel activities despite new surges of COVID-19 reflect the significance of certain social interactions and related locales and facilities. Whether and how the members adjust their travel purposes, frequencies of ITG trips, or the composition of their ITGs are partially associated with people's abidance of related anti-pandemic policies. Furthermore, variations in the ITG structure during the pandemic can be related to the distribution of essential resources, facilities, and opportunities in cities. These variations could therefore have equity and justice implications. For instance, it is likely that ITG members who made recurrent trips with one another all had good access to venues such as shopping malls and pedestrian-only streets.

3. Data and methods

3.1. The site and data

In this article, we empirically examined ITGs among metro riders in Hong Kong and measured their structural changes during the COVID-19 pandemic. The Hong Kong MTR Corporation Limited (called “MTR” locally) operated the system with 96 stations as of 2020. Considering data constraints and temporary service suspensions of several stations during the pandemic, only 88 stations were used for analysis. The data included each metro trip of smartcard holders between two stations with the swiping-in/out records in sequence. All records had been anonymized to protect privacy. We used metro records of all the weekends in 2020, with a total number of 104 Saturdays/Sundays (52 weekends), to examine group travel behaviors in different periods of the pandemic. The data recorded the following variables: hashed smartcard ID, date of the day, entry and exit time, station ID of each transaction, and card type. In this article, four card types were used to identify ITGs, including adult, child (3–11 years old), student (full-time day course students of 12–25 years old), and senior (65 years old and above). The trips with the same entry and exit stations were excluded as we considered that they did not truly occur.

Throughout the year of 2020, Hong Kong suffered from four surges of locally confirmed COVID-19 caseloads since the first local case emerged on Jan 23, 2020 (SI Fig. 1). There were approximately 70, 886, 3661 infected cases and 3174 related deaths in the four surges, respectively. In response to these surges, a series of anti-pandemic measures were implemented, e.g., the working-from-home mandate, bans on gatherings in public spaces, and closure of selected “non-essential” businesses such as recreational facilities. Impacted by these countermeasures and fear of infection, social interactions in the city changed.

3.2. The Research Framework

The overall research framework is presented in Fig. 1 . Specifically, we first identified ITGs in the metro system. Then, to explore their structural changes during the pandemic, we measured different characteristics of the ITG structure across all the metro stations and by individual stations and station areas. After those, we examined whether and how the local ITG structure varied and clustered across space and time, which reflects the geographically unequal distribution of ITG performances in the city. Lastly, given the pattern in ITGs' distribution and its temporal changes during the pandemic, we investigated probable influencing factors of the pattern and its changes.

Fig. 1.

The research framework.

3.3. ITG structure identification

3.3.1. Identifying ITGs in the metro system

We identified ITGs based on Zhu et al. (2021) and Zhou et al. (2022) and our own experiences following the steps shown in Fig. 2 . First, we established a dataset for potential group trips (PGTs). The dataset included trips by any two metro riders who swiped into the same station, and then swiped out of the same station within 1 min. Only weekend data was used since there were fewer concurrent commuting trips and more leisure and maintenance trips on weekends.

Fig. 2.

A flow chart for the ITG identification.

Then, considering commuters who worked on both weekdays and weekends might lead to false group travel identification, we identified these cohorts and removed their probable commuting trips from the PGTs dataset. Specifically, we identified regular daily commuters as adult riders who (a) had consecutive round trips with at least 6 h’ stay as the probable working duration in a day, and (b) made such trips at least eight times per month. The potential commuting trips during the weekends were then removed from the PGTs dataset. After this, we obtained a non-commuting PGTs dataset.

After that, to filter out riders traveling together by chance, we categorized the PGTs according to their group trip frequency (β) by month in the non-commuting PGTs dataset. We assumed that two riders were more likely to know each other if they had more PGTs. Learned from a KS-test designed by Zhu et al. (2021), we examined the distribution of the exit time difference (0–60 s) between riders’ PGTs to single out ITGs. A primary assumption for the test is that the exit time difference of PGTs between two riders in an ITG would generally be smaller than that between riders traveling together by coincidence. The details of the KS-test are shown in Supporting Information. Finally, we considered two riders with at least four PGTs in a month as our most probable ITGs.

Given the above backdrop, one ITG consists of two metro riders who (a) had PGTs, that is, they swiped into/out the same metro stations within 1 min, respectively; and (b) were detected four times or more non-commuting PGTs on weekends in at least one month of the year. As a result, throughout the year 2020, a total number of 1,316,073 non-commuting ITGs and 1,524,000 distinct ITG riders were identified using the above ITG definition. The ITG trips account for 2.4% of all the trips. Zhu et al. (2021), for instance, acquired a ratio of 5.7% for the group trips relative to all trips. We considered our ITG ratio could be lower since we focused on weekend trips and removed co-presenting commuting trips. Tao and He (2021) found that commuters’ participation to joint household travel (including discretionary and maintenance travel) would be about 1.4% in Hong Kong. The ratio in ITG trips might be higher since we involve more kinds of group trips from various types of people. Overall, we felt that the ITG ratio is comparable with that in previous studies which might help validate the methods on ITG identification.

3.3.2. Measuring ITG structure with empirical data

We utilized a smartcard dataset from Hong Kong to operationalize different indicators of the ITG structure of metro riders. Given that the city is highly transit reliant—41% of the daily trips were carried by MTR (Transport and Housing Bureau, 2017), we believe that intentional group travel of MTR riders can well represent that of the regular citizens in the city across space and time.

The ITG structure could be quantified at both the system level and the local level. Table 1 presents the characteristics of the ITG structure we considered at the system level. Due to data and budgetary constraints, we were unable to survey or verify the exact functions of different ITG trips that we identified. Hence, only five of the six characteristics of the ITG structure were considered in our empirical case.

Table 1.

The ITG structure and indicators at the system level.

Characteristics	Indicators in the Metro system
Composition	Group types identified by riders' card types (e.g., the adult, senior, student, and child)
Function	Not considered
Size	Number of ITGs
	Number of ITG riders
Intensity	Share of ITG riders relative to all the riders
	Share of ITG trips relative to all the trips
Quality	Daily trip distance of ITGs
Spatiotemporal distribution	Spatial Stickiness
	Temporal Stickiness

For composition, we used types of smartcards to define it. In our case, the card types include adult, elderly, student, and child. Each two-member ITG thus have one or two card type, and form compositions, such as two adults, or one adult and one elderly. We also defined each rider's probable “home” place as the station the rider mostly starts his/her trip of a day within a year. The compositions allow us to know some coarse attributes of the members in an ITG and their possible relationships, such as family members or friends. The one-adult-and-one-child ITG, for instance, with the same home station might involve a parent and child in a family.

For size, we operationalized it at both the system and station levels. It can be the number of ITGs/ITG riders for all the stations in the metro system or for a specific metro station.

For intensity, it was calculated as the share of ITG riders/trips to all the riders/trips. Again, it can be at the system or station levels.

For quality, we used the shared trip distance of ITG members as a proxy. We assumed that the longer an ITG trip, the higher quality of social interactions and closer contacts the members make.

For spatiotemporal distribution, we used the spatial and temporal “stickiness” indices by Zhou et al. (2021). The indices can measure to what degree ITG trips from few origins ($S_d$) (or few periods ($S_p$)) stick to Station i. S ranges from 0 to 1, where the larger S the fewer origins or time slots those incoming ITG trips of a station concentrate on. In this article, $S_d$ and $S_p$ for station i were calculated as follows:

$$S_{di}=\sum _{j=1}^n{\left(\frac{r_{ji}}{A_i}\right)}^2$$ (1)

$$S_{pi}=\sum _{t=1}^k{\left(\frac{r_{i,t}}{A_i}\right)}^2$$ (2)

where,

• $A_i$ is the total incoming ITG trips from other stations to Station i per day.

• $r_{ji}$ is the incoming ITG trips from Station j to Station i, j <>i and n is the total number of stations (origins) that have non-zero ITG trips to Station i;

• $r_{i,t}$ is the incoming ITG trips swiping out Station i at hour t, and k is the total hours of operation in the metro system.

3.4. ITGs’ geographical distribution

To explore ITG's geographical distribution, we grouped and measured the ITG structural characteristics by MTR station. Corresponding measurements served as our proxies for the geographical distribution of facilities and locations in or around MTR stations attractive to outgoings and health risks across these stations over time. Specifically, we calculated three station-level ITG indicators: the number of incoming ITGs (size), the share of incoming ITG riders (intensity), and total incoming ITG trip distance (quality). We then combined them into a composite index by utilizing the Information Entropy Weighting (IEW) method proposed by Shannon (1948). The IEW could help allocate weights among ITG indicators to identify informative indicators. The composite ITG index could be calculated by combining indicators according to their weights.

Based on this index, the Global Moran's I (Moran, 1950) has been used to examine the spatial autocorrelation of the station-level ITG characteristics in different local caseload surges. Moran's I statistics range from −1 to 1, where the closer the statistics -approach 1 or 1 the higher levels of clustering of similar or opposite values. Values around 0 denote complete spatial randomness. To combine the temporal variations of stations' ITG index with their spatial clustering, we applied k-means time-series clustering to categorize stations into a certain number of groups. Samples in each group share some similarity in one or more dimensions measured by different indicators or indices. In our case, the method is expected to classify stations with similar ITG structures across time into the same group. Specifically, we used K-means as the unsupervised clustering algorithm to split stations into k groups and Dynamic Time Warping (DTW) as the similarity measurement to calculate and compare the distance between different time series concerning the ITG structure.

4. Results

4.1. Overall structural change of ITGs

To analyze the overall structural changes of ITGs across different surges of the local COVID-19 pandemic, we examined the relative changes of ITG characteristics in the local metro system.

4.1.1. Compositions

ITGs did not decrease equally across group types. Fig. 3 shows the variations across ITG group types during the pandemic. Data was aggregated daily by calculating the average of each weekend. Throughout the year, 73% of the ITGs involved members with the same home stations ("same-home ITGs”), which were more probable to be family or neighboring groups, and the remaining 27% ("dif-home ITGs”) were more likely to be friend groups.

Fig. 3.

Variations in ITG Group Types. The group types are described by different card types, including the adult (ADL), senior (SEN), student (STD), and child (CHD).

Overall, except for those two-senior ITGs remained relatively stable across the local surges, most other ITGs fluctuated and followed similar trends. After the local pandemic outbroke, the most significant reduction could be found for the one-adult-and-one-child, one-senior-and-one-child, and two-children ITGs. These were followed by two-senior ITGs, which decreased up to 40% (for the same-home ITGs) or 25% (for the dif-home ITGs) of the pre-pandemic level and remained low at this level until August 2020. It seemed that some vulnerable cohorts, like the elderly and the young, reduced their group travel activities more than other cohorts during the pandemic. The fluctuation in children's ITG trips might be impacted by the local regulations on the resumption and suspension of face-to-face teaching. The seniors, however, still made few ITG trips but did not totally forgo such trips during the pandemic.

The dif-home ITGs shown in Fig. 3b had seen greater variations across group types than the same-home ITGs in Fig. 3a. Several types of dif-home ITGs, such as the two-adult, the two-student, and the one-adult-and-one-student, rebounded more quickly and significantly than other types of ITGs when the local COVID-19 caseloads remained low. A probable explanation for this phenomenon is that some outgoings activities of the same-home ITGs could done without using MTR whereas the friends or relatives from dif-home ITGs could not do so.

4.1.2. Size and Intensity

Fig. 4 shows the changes of daily ITG size, measured by numbers of ITGs and distinct ITG riders. During the baseline weekend, there were 189,224 ITGs and 277,327 distinct ITG riders daily on average. Both the numbers decreased significantly when a local COVID-19 caseload surge occurred. The size dropped more than 50% in the first three surges. However, the decline was followed by a subsequent hike to the nearly pre-pandemic level when the case increase remained low between surges. This might be joint consequences of the decreased fear of the pandemic and increased relaxation of local social-gathering restrictions. Interestingly, the decrease in the ITG size was less significant in the fourth surge relative to that of the first three surges, even though there were a larger number of cases confirmed within the last several months of the year. This implied that ITG riders became less sensitive to the pandemic as the fight against COVID-19 had become a long-lasting task.

Fig. 4.

Variations in ITG Size and Intensity. The size is measured by quantity of ITG riders and ITGs, and the intensity measured by share of ITG riders and ITG trips.

The intensity of ITGs was measured by the share of ITG riders/trips to all the riders/trips (Fig. 4 ), which were 13.5% and 13.1% respectively during the baseline week. These two indicators followed a similar fluctuation pattern amid the pandemic. A reciprocal relationship could be observed between the ITG intensity and caseloads. This indicates that group travel was more sensitive to the pandemic's severity than to the total number of local metro trips, i.e., the overall level of station crowdedness. Relative to all the riders/trips, ITG riders/trips decreased more when a surge came and rebounded more when the case increase reached a null.

4.1.3. Quality

The reduction in the numbers of ITG trips and riders did not necessarily influence the average benefits that produced by ITG travel. We calculated the daily ITG trip distance by the ITG to as a proxy for us measure the quality and closeness of physical interactions between ITG riders. Fig. 5 compares the complementary cumulative distribution functions (CCDF) of the daily trip distance by the ITG in different stages of the pandemic. The data was normalized into the daily average based on the three weekends pre-pandemic and those during each surge.

Fig. 5.

The complementary cumulative distribution functions (CCDF) for the daily trip distance by ITG.

It could be observed that the ITG trips longer than 20 km decreased more strongly than other trips after the local pandemic outbroke. Compared with the first surge, the daily trip distance was suppressed more in the second and third surges. It was during these two surges that various local countermeasures were implemented, e.g., restrictions of social gatherings and closure of non-essential businesses. These countermeasures might have largely decreased the quality of social interactive activities. In the fourth surge, the daily trip distance increased as compared to the second and third surges. But it was still lower than that before the pandemic. This implies that ITG riders might have gradually learned to co-exist with the pandemic over time.

4.1.4. Spatiotemporal distribution

The daily spatial stickiness ($S_d$) and temporal stickiness ($S_p$) for a station measure to what degree its incoming ITG trips concentrate in few origins and time slots, respectively. Fig. 6 shows the average $S_d$ and $S_p$ by stations. It could be observed that both the $S_d$ and $S_p$ fluctuated but were generally higher than the pre-pandemic level after the local pandemic outbroke. They both increased during the local caseload surges. Incoming ITG trips concentrated more in few origins and in certain hours after the local pandemic outbroke. This could mean demands for ITG trip destinations became more centralized and there were preferred hours of travel for ITG riders amid the pandemic.

Fig. 6.

The average spatial stickiness ($S_d$) and temporal stickiness ($S_p$) for incoming ITG trips by station during the pandemic.

4.2. Station clustering and geographic inequality

Despite the above-mentioned overall changes, we found that the ITG structure varied across stations and over time. We acquired the station-level ITG indices across days by weighting and combining the size (the number of incoming ITGs), quality (total incoming ITG trip distance), and intensity (share of incoming ITG riders) of each station based on the IEW method. Then, the Global Moran's I statistics for stations' average ITG index in each local surge were examined (Table 2 ). It was found that it was more likely for stations in proximity to share similar the ITG indices during the first two surges, with Moran's I statistically significant at the 0.01 or 0.05 level. The distribution of ITGs, however, became more spatially random after the first surge. This could mean that ITG riders across stations had gradually adapted to the pandemic.

Table 2.

Station groups based on the ITG Index clustering in four local surges.

	Moran's I	Group size	ITG Indices
			Mean	Std	Max	Min
Surge 1	0.257***
Group #1		22	0.019	0.003	0.025	0.015
Group #2		43	0.011	0.002	0.015	0.008
Group #3		23	0.005	0.002	0.008	0.002
Surge 2	0.223**
Group #1		19	0.020	0.003	0.027	0.016
Group #2		38	0.012	0.002	0.015	0.009
Group #3		31	0.005	0.002	0.009	0.001
Surge 3	0.128
Group #1		17	0.020	0.004	0.031	0.016
Group #2		44	0.011	0.002	0.015	0.009
Group #3		27	0.006	0.002	0.008	0.001
Surge 4	0.105
Group #1		19	0.020	0.004	0.032	0.015
Group #2		50	0.011	0.002	0.015	0.008
Group #3		19	0.005	0.002	0.007	0.001

***p < 0.01, ** 0.01<p < 0.05, * 0.05<p < 0.1.

To explore how stations geographically clustered based on their ITG indices over time, we executed the K-means time-series clustering for stations’ relative ITG index in different local surges. Each dataset would be on 88 time series (stations) with the number of weeks in each surge as the length of time-series. Since the number of groups (k) needs to be specificized beforehand, we tried and ended up setting k = 3 after comparing the sum of squared distances for different k. This suggests stations would be divided into three groups for the ITG indices in our analysis. The groups are described as:

Group #1: Stations whose ITG indices were and/or remained higher than the median during a surge;

Group #2: Stations whose ITG indices were around the median during a surge;

Group #3: Stations whose ITG indices were and/or remained lower than the median during a surge.

Table 2 shows the station groups/clusters based the ITG index clustering in four local surges. Fig. 7 geovisualizes their geographical distribution across the four local caseload surges. Overall, stations in different groups scattered around the city. A small set of stations remained in Group #1, as the hot spot of ITGs, across four surges. The clustering of such stations could typically be observed in new towns, a type of self-reliant “satellite town” in Hong Kong. For instance, those stations clustered in the Tseung Kwan O New Town, e.g., Tseung Kwan O Station (50) and its neighbors, were all Group #1 members. Other new town stations on the same line in proximity with relatively high ITG indices were the Ma On Shan Station (102)-Wu Kai Sha Station (103), as well as the Tai Wai (67)-Sha Tin (68). In addition, stations located in areas well-known for high commercial and recreational densities could also remained in Group #1 across surges, such as Mong Kok (6), Tsim Sha Tsui (6), and Causeway Bay (28). It seems that these pre-pandemic leisure centers were still hot spots of local group travel activities during the pandemic.

Fig. 7.

Spatiotemporal distribution of station groups.

Several stations in Group #1 in the first surge became relatively cold spots in later surges. Wong Tai Sin (10) was a popular tourism attraction prior to COVID-19 in Hong Kong. However, people had largely reduced group travel and activities to this station after the pandemic. Additionally, Lam Tin (38) also lost its leading role in ITG index ranking after the first surge. A probable explanation fewer choice ITG riders would visit Lam Tin's shopping malls, recreational facilities, and schools, whose opening and services were largely restricted during the pandemic.

4.3. Association between case footprints and local ITG structure

To examine whether the distribution of COVID-19 cases was associated with the station-level ITG structure, we calculated the daily case footprints and ITG Indices by station during each case surge. The case footprints were measured by the number of venues visited by the infected within 800 m of stations. By ranking the stations based on their ITG indices in the different case surges, we classified them into four quantiles: Q1: the stations with the highest ITG indices of the surge, Q4: the stations with the lowest ITG indices, and Q2 and Q3: the stations with the ITG indices in between. Fig. 8 shows the percentage of case footprints of each quantile relative to the total number of case footprints size in different surges. The relative case footprints of Q1 and Q2 largely decreased in the second surge compared with the first surge, but rebounded in the third and fourth surges. The results might be because the government started implementing restrictions on social gatherings during the second surge, which decreased the impacts of group activities on local caseloads. However, some high-exposure stations, that is, stations with more historical case footprints, became popular destinations of group travel again as the pandemic progressed and other stations started seeing more case footprints.

Fig. 8.

Percentage of ITG size among stations with different numbers of case footprints nearby. Stations were classified into four quantiles based on the ITG index. Q1 represents the top quantile; Q2 represents the second quantile; Q3 represents the third quantile, and Q4 represents the bottom quantile.

4.4. Effects of Station Features on local ITG structure

The station groups based on the time-series clustering method indicate the geographical pattern in group travel activities, which reflected varying health risks across MTR stations. The pattern can be related to the unequal geographical distribution of essential facilities and opportunities. We, therefore, hypothesized that some features of stations and their surroundings can predict the ITG structure. To test this hypothesis, we fitted multinomial logistic regression models for different local surges. The details are as follows.

4.4.1. Probable influencing factors

We assumed that there were three sets of probable influencing factors for the ITG structure. Each factor can be measured using one or two variables. Table 3 presents more detail about these variables.

Table 3.

Descriptive statistics of the probable influencing factors/variables.

Variables	Description	Mean (St.dev)
Centrality degree
Global Centrality	Average travel time to any other station in the Metro system (mins)	39.42 (7.67)
Regional Centrality	Population coverage in 15 min	41406.45 (26652.48)
Socioeconomics
Income	Income median (HK Dollars)	14539.55 (5736.34)
Age	Percentage of population over 65 years old (%)	13.05 (4.62)
Education Level	Percentage of population with postsecondary degrees (%)	30.20 (12.52)
Household Size	Average number of household members (persons)	2.90 (0.32)
Built environment
Points of Interest (POIs)	Number of POIs within 800m.	2682.18 (2240.53)
Simpson Index	Diversity index of POIs between 0 and 1	0.19 (0.05)
Commercial	Number of Commercial POIs within 800m	406.33 (430.69)
New Town	Whether the station is located in a new town developed in the third phase (Yes = 0 No = 1)	0.16 (0.37)

The first set is concerned with the centrality degree of a station in the metro system, which describes the level of metro-based accessibility to facilities and opportunities those people still need to reach despite the probable infection risks (Zhou et al., 2021). Specifically, the average time of a station to all other stations is used to measure the global centrality of this station in the metro system. The lower value indicates the higher centrality degree. Population residing in 15 min’ metro ride to a station depicts the regional centrality of this station. The centrality indicators were calculated because we assume that concentration of population is accompanied with more human activities and higher virus-spreading risks.

The second set includes the socioeconomic characteristics of station-served communities. The impacts of and the risks brought about by the COVID-19 pandemic were socioeconomically unequal in the city (Chang et al., 2021; Grekousis et al., 2021). Human mobility patterns also changed heterogeneously across locations with different socioeconomic attributes (Hu & Chen, 2021; Palm et al., 2021). We hypothesized that these attributes would be associated with people's ability and/or willingness to conduct group activities after the pandemic outbroke. Based on the 2011 Hong Kong Census Data, we formulated variables to capture sociodemographic characteristics by the local Tertiary Planning Unit (TPU). Each TPU with one or more metro stations was treated as a station-served community.

The third set involves variables characterizing the built environment of station-served communities. We hypothesized that people still need to physically meet friends and family in some venues, e.g., recreational and commercial facilities, despite the pandemic. Specifically, we used the points of interest (POIs) data acquired from Google Map API in 2017/2018 to formulate variables like the number of POIs, the POI mixture, and the number of commercial POIs (defined by Google). Inspired by Zhou et al. (2020), the Simpson Diversity index was used to quantify the POI mixture where smaller values indicate higher diversity. Besides, we used a dummy variable to decide whether a station is a third-phase new town in Hong Kong. This is because the third-phase new towns are internationally well-known for their “Rail + Property” development model, where metro stations are closely connected with residential and commercial development (Cervero & Murakami, 2009). This might influence how riders residing in these new towns use metro for group travel activities.

4.4.2. Modeling results

Table 4 shows the modeling results. The stations in Group #3, where the ITG indices were comparatively low during a surge, were chosen as the reference for each model. As for the centrality degree, compared with Group #3, stations with higher regional centrality, that is, more population residing in a 15 min’ metro ride, were associated with higher ITG indices. This was statistically significant at a p-value of 0.05 in the second and third surges.

Table 4.

Results of multinomial logistic regression models.

Group	Surge 1		Surge 2		Surge 3		Surge 4
	#1	#2	#1	#2	#1	#2	#1	#2
Centrality degree
Global Centrality	0.903	0.880	0.975	0.894	1.030	0.948	1.005	0.928
Regional Centrality	1.001	1.001	1.001**	1.001	1.001**	1.001	1.001*	1.001
Sociodemographic
Income median	1.001***	1.001*	1.001***	1.001**	1.001*	1.001**	1.001**	1.001
Age over 65	1.635E-11	5.892E-19***	34.740	1.134E-8	2.292	9.740E-10	1.254E-12	7.190E-24***
Postsecondary degree	3.083E-31***	1.426E-14***	3.650E-32***	8.032E-21***	4.087E-17**	1.158E-13**	2.812E-22***	3.451E-10**
Household Size	3685.746***	17.015*	22258.532***	770.820***	2222.017**	46.589**	300.657**	5.052
Built environment
POIs	0.998**	0.999**	0.996***	0.998***	0.997***	0.998***	0.998**	0.999*
Simpson Index	1.207E-9	3.872E-8	3.115E-5	0.001	0.001	7.761E-7	2.725E-9	1.359E-7
Commercial	1.022***	1.015***	1.034***	1.026***	1.031***	1.023***	1.023***	1.014**
New Town (ref = Yes)
No	7.011E-10***	1.057E-8	0.002***	0.027**	0.007***	0.132*	0.028**	0.560
Omnibus Test	0.000		0.000		0.000		0.000

Ref: Group #3.

OR (Exp B) is reported.

***p < 0.01, ** 0.01<p < 0.05, * 0.05<p < 0.1.

Several station-level socioeconomic characteristics were correlated to the corresponding ITG indices by station. Income median and highly-educated population were statistically significantly associated with stations' ITG indices in all the surges. The income median was found to be positively related to the ITG indices, while stations with a higher percentage of high-educated people were less likely to have high ITG indices. Similarly, a statistically significant relationship could be found between stations’ household sizes and ITG indices across the surges. The household size could positively contribute to the ITG indices. It was probable that many families with multiple members would still maintain some household-based travel and activities despite the presence and persistence of the pandemic. In addition, stations with more elderly population over 65 years old resided saw lower ITG indices. It was less likely for these stations to be in Group #2 relative to Group #3 in the first and fourth surges. A probable explanation was that the elderly generally had fewer metro group activities during the weekends, or they decreased their intentional group travel more during certain surges.

Interestingly, the number of POIs around a station was negatively correlated to its ITG index in all surges, but commercial facilities could positively predict the ITGs’ distribution. It seems that intentional group travel activities were and remained active in certain kinds of facilities, like commercial and recreational facilities, so long as the local policies allowed them to operate. Besides, we found that stations located in the third-phase new towns were far more likely to belong to Group #1, which enjoyed higher ITG indices. These self-reliant communities seemed to bring about and sustain more group travel activities during the pandemic. Of course, these activities could pose higher virus infection risks for participants at the same time.

5. Discussion and conclusions

Close physical contacts amid COVID-19 surges could pose high virus transmission risks. However, we found based on empirical data from Hong Kong that people might still travel together and engage in face-to-face social interactions. These activities remained as indispensable parts of people's daily life amid the pandemic. In theory, they facilitate important social bonds and meaningful coordination, and help maintain and enhance people's well-being. By using metro riders as samples, we attempted to quantify and geovisualize those group-based activities, in particular, intentional group travel and their associated venues. We also examined the spatiotemporal evolution of these activities across the four different surges of locally confirmed cases.

In summary, we felt we advanced the existing scholarship in several aspects. First, we proposed the concept of ITG structure, which can be used to measure group travel activities in dimensions such as composition, function, size, intensity, quality, and spatiotemporal distribution. Second, we illustrated that existing proprietorial and publicly available data, such as smartcard and census data, can be used to operationalize indicators about the ITG structure and their probable correlates. Third, using those indicators, we can better quantify group travel activities, their spatiotemporal evolution, and possible influencing factors/variables. In the context of Hong Kong, we found that group travel activities' characteristics varied across space and time amid the pandemic. Notably, variations of ITGs in the fourth surge were smaller than in the first three surges, even though there was a presence of larger locally confirmed caseloads. ITG trips' origins and destinations clustered around several new town areas in all the four surges of local caseloads. The metro stations’ centrality, socioeconomic characteristics, and the built environment of their surrounding communities could well predict the ITG structure.

The resulting knowledge could facilitate more targeted and effective policies regarding both short-term anti-pandemic management, and the long-term supply of public/essential services or facilities in our cities. On the one hand, it is recommended that the local government implements more targeted resilience management and strict disinfection in stations where ITG riders frequent emerge. Other things being equal, group travel and related activities with close physical contacts could bring high virus transmission risks. On the other hand, there was heterogeneous ITG distribution in the city and the heterogeneity could vary across different surges of local caseloads. Some metro riders remained their ITG trips while others didn't or were unable to do so. These phenomena could have equity implications—if it was because of the supply and quality of essential facilities and services that prevent some people from making ITG trips. Because of our cities were not planned and built for unprecedented societal shocks like COVID-19, it is highly likely that certain stations and their surroundings can have better supply of facilities and services than others to satisfy the group travel and activities amid the pandemic. Our work on ITG provides some new angles for us to revisit the issue of the supply and quality of facilities and services across our cities hit by societal shocks.

Our research, nevertheless, suffers from some limitations. First, without conducting surveys among local riders/residents, we did not know if decreased group travel activities would reduce these people's social interactions and associated benefits. Second, the ITGs identified from the metro smartcard data only partially reflect local physical social interactions in transit-reliant cities like Hong Kong. To what degree metro riders and their group-based activities could represent the local population remains unknown. Third, the existing data we used did not cover all the dimensions of the ITG structure, most notably, the functions of an ITG trip. Future studies should well address all the above limitations to be more relevant in the scholarship and for policymaking.

Author contributions

J. Z. acquired the funding and data for the research; J. Z and M. Z. designed and operationlized the research; M. Z performed research, analyzed the data, and wrote the initial draft; J. Z. reviewed all the intermdiate outcomes, edited and revised mutiple versions of the draft.

Declaration of competing interest

The authors declare no competing interests.

Appendix A. Supplementary data

The following is the Supplementary data to this article: