Intentional Travel Groups and Social Networks during the COVID-19 Pandemic

doi:10.21203/rs.3.rs-1548702/v2

Download PDF

Research Article

Intentional Travel Groups and Social Networks during the COVID-19 Pandemic

https://doi.org/10.21203/rs.3.rs-1548702/v2

This work is licensed under a CC BY 4.0 License

Version 2

posted

You are reading this latest preprint version

Although face-to-face social contact decreased during the COVID-19 pandemic, some people remained active in traveling and physically meeting as a group to gain benefits like sustaining intimacy and increasing productivity. These group-based activities indicate the necessity of some essential physical meetings despite high risks of virus infection from physical contact amid the pandemic. To balance the benefits of these meetings with the risks and consequences of infection, more should be done to understand the characteristics of physical social contacts and interactions before and amid the pandemic. Using multiday continuous smartcard data, we identify patterns of intentional group travel in the Hong Kong Metro system. These patterns serve as our proxies for physical (visible) interactions and social (invisible) contact networks among people who intentionally travel as a group (ITG). We measure the spatial centrality of ITGs and persistent group riders (PGRs), a subset of ITGs remaining active amid the pandemic, to infer social interactions among metro riders across different locales. By examining the social network formed by the ITG members across time, we found that its size and interconnections varied during the pandemic, and PGR members might be influential vertices in maintaining the networks’ topological properties. The findings could facilitate transit-usage-and-virus-spread modeling and the formulation of more effective pandemic countermeasures in transit-reliant cities.

COVID-19

intentional travel groups

social networks

scaling laws

metro system

The COVID-19 virus often spreads within social interactive networks (WHO, 2021; Block et al., 2020). In response, face-to-face meetings decreased, as a consequence of both people’s voluntary reductions of physical contacts and governments’ anti-pandemic measures to restrict group activities, e.g., bans on gatherings and closure of non-essential businesses (Block et al., 2020; Benzell et al., 2020; Chou et al., 2020). Hence, COVID-19 greatly suppressed physical social interactions, part of which were increasingly integrated into digital (invisible) interactive activities (Marinucci et al., 2022; McQuire, 2017; Carlsen et al., 2021), especially online working and learning (Carrillo and Flores, 2020; Nguyen, 2021; Lee et al., 2022), and others were probably forgone—temporally or permanently. A new social interaction landscape has emerged. However, there remain in-person social activities that encapsulate those resilient, meaningful and enduring social bonds such as friends, relatives and families in our cities (Florida et al., 2021). For many people, they are beacons in the darkest days caused by the unprecedented pandemic.

Exploring social interactions among people from the field of transportation have been many. Public transportations and its supporting facilities, like platforms, comprise important public spaces where different social interactions such as encounters, familiar strangers, and group travel behaviors frequently occur (Sun et al., 2013; Zhou et al., 2020; Zhu et al., 2021). A series of studies simulated how joint decisions through social contacts influence travel choices (Dubernet and Axhausen, 2015; Gliebe and Koppelman, 2002). Meanwhile, various factors were found to impact joint travel behaviors and local social ties, e.g., personal characteristics, facilities in the neighborhood, and working time (Tao and He, 2021; van den Berg et al., 2015). In this context, social networks and activity-travel patterns tend to be associated and co-evolve over time (Arentze and Timmermans, 2008; Sharmeen et al., 2015). This might especially be the case during the pandemic, when both people’s social interactive activities and travel patterns changed.

During the pandemic, studies hitherto have considered mobility as a prerequisite of physical meetings. A wide range of individual mobility and their potential offline social interactions were investigated based on various mobility data, like cellular signal data (Chang et al., 2021) and smartcard data (Zhang et al., 2021). Data fusion in different formats and from different sources, e.g., social media data and cellular network data, were used to examine human interactions in both virtual and real worlds (Chatterjee et al., 2022; Eagle et al., 2009). Against the backdrop of the pandemic, it has been investigated whether and to what extent online social interactions substitute in-person activities, promote social and psychological supports, and help predict the transmission process (Marinucci et al., 2022; Carlsen et al., 2021; Feng and Kirkley, 2021). Yet, to our knowledge, little research has concerned how realized physical social interactions and corresponding social networks reacted to the pandemic. Some other studies particularly examined how individuals’ sociability and behaviors changed given the fear of infection based on empirical data, e.g., interviews, surveys, and time-use diaries (Chou et al., 2020; Amram et al., 2022; Sullivan et al., 2021). Physical human interactions were still found essential during the pandemic, which could sustain and even improve people’s mental health and working efficiency (Lee et al., 2022; Glaeser, 2020). However, little has been done on the emerging citywide characteristics of social interactions amid the pandemic.

In this article, we explore citywide social interactions before and amid the pandemic from the field of transportation. Specifically, we probe intentional travel groups (ITGs) within the Hong Kong metro system based on one year’s local transit smartcard data. The data spans the year 2020 and represents a notable percentage of physical social interactions in the city, where Metro accounts for 41% of the local passenger trips (HK Transport Department, 2022). We assume that each ITG consists of two riders who know each other and intentionally travel together. ITGs are a subset of inter-relationship among riders, and they form an invisible but meaningful “social network” (“ITG network”). In this network, individual ITG riders can be regarded as nodes and the group travel interactions between individual ITG riders, can be regarded as links. Based on the ITG riders’ records in the metro system, we can identify their spatial and temporal distribution across the city.

The article is organized as follows. After the introduction section, we describe the data and methods used for operationalizing the ITG network using smartcard data in Hong Kong. Then, we examine whether and how the resulting ITGs and the identified ITG network varied amid the pandemic based on four hypotheses. After that, we conduct a disscussion on the work’s innovations, implication, and limitations, and lastly, conclude the work.

Metro Smartcard Data

The metro riders’ smartcard swipe records were obtained from the Hong Kong MTR Corporation Limited (MTR). MTR operates 96 metro stations as of 2020 (Supplementary Fig. S1). The records include metro usage information of smartcard holders, that is, a metro trip between two stations with the swipe-in and swipe-out data in sequence. All records were anonymized to protect privacy. A dataset covering all weekends 2020, with 104 Saturdays/Sundays, is used to examine group travel behaviors in different periods of the pandemic. The data consist of the following variables: hashed smartcard ID, date, entry and exit time, station ID of each transaction, and card type. In this article, data of all types are used to identify ITGs. However, four card types are particularly used to analyze the potential relationship between riders, 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 article excluded the trips whose entry and exit stations are the same.

Identify intentional travel groups (ITGs)

The premise of methods to identify the group travel in this article is that riders in ITGs have intentional and repeated social contacts (interactions) rather than those meeting coincidently. We filter ITG trips from encounters in the metro system based on their consistency in travel time, locations, and contact frequency. Firstly, an initial potential group travel dataset is constructed by setting time thresholds and fixing locations for the trips. Only when two riders, i and j, entering and existing the same stations within one minute are considered finishing a potential group trip (\({g}_{ij}\)). \({g}_{ij}\) is required to satisfy the following conditions:

\({o}_{i}\) = \({o}_{j}\)	(1)
\({d}_{i}\) = \({d}_{j}\)	(2)
\|\({ent}_{i}\) - \({ent}_{j}\)\| ≤ 1 min	(3)
\|\({ext}_{i}\) - \({ext}_{j}\)\| ≤ 1 min	(4)

where, \({o}_{i}\) and \({o}_{j}\) are the origin station ID of the rider i and j, respectively. \({d}_{i}\) and \({d}_{j}\) are the destination station ID of the rider i and j, respectively. \({ent}_{i}\) represents the time when rider i swiped in the origin station \({o}_{i}\) and \({ext}_{j}\) represents the time when rider i swiped out the destination station \({d}_{i}\). A similar explanation could be applied to \({ent}_{j}\) and \({ext}_{j}\).

Then, since the above strategy would include a volume of riders traveling together by chance, further work is needed to identify ITGs, that is, two riders in a group indeed know each other and conduct group trips together. We reason that two riders are more likely to be “friends” if they have more frequent group trips observed in the system.

Assuming that the exit time difference of group trips between two riders in an ITG would generally be smaller than that between those travelling together by coincidence (Zhu et al., 2021), we did KS-tests of the distribution of the exit time difference between group trips adopted for the dataset. The total number of group trips of rider i and j in a month is defined as \(\sum {g}_{ij}\). When \(\sum {g}_{ij} \ge\) \(\beta\), where \(\beta\) is a given threshold of group trip times, we consider the group with rider i and j as an ITG and it appears at least \(\beta\) times in the potential group travel pool in the certain month. The KS-tests of the distribution of the exit time difference between group trips were adopted for various potential \(\beta\). For an ideal \(\beta\), intentional travel groups occurring at least \(\beta\) times and those following a higher threshold should follow the same distribution in exit time difference of trips. Accordingly, the hypothesis for the tests is:

\({H}_{0}\): exit time difference of group trips in potential travel groups with group times more than \(\beta\) and more than \(\beta\) + 1 in a month follow the same distribution.

\({H}_{1}\): exit time difference of group trips in potential travel groups with group times more than \(\beta\) and more than \(\beta\) + 1 in a month follow the different distribution.

Considering the pandemic impacts, we did KS-tests for both \({g}_{ij}\) in January, as a sample month before the local pandemic outbroke, and February, as a sample month after. Supplementary Fig. S2 shows that for both months, the null hypothesis could be rejected before \(\beta\) = 4 with a p-value smaller than 0.05. Hence, the ITGs in this article would be defined as those with group trips at least 4 times within a month.

Non-commuting ITGs

Considering commuters who work on both weekdays and weekends might lead to coincident group trips, we tried identifying these groups of people and removed them from the ITG datasets when examining degree distribution. To be specific, we first select out regular daily commuters as who: (1) are adult riders and have at least one consecutive round trip in a day, with at least six hours between trips of the round trip; (2) have trips like (1) between the same pair of stations at least eight times per month. The “home” stations here are those where commuters start and end their round trips of a day, and the “work” stations are their intermidiate stations between their round trips. Hence, a dataset for local regular commuters is acquired by examining metro smartcard data for the first eight months of the year 2020.

Then, we identify group commuting trips during the weekends as those conducted by two metro commuters with the same “home” and “work” stations and travel together between these two stations during the weekends. These potential group commuting trips are removed from the dataset of potential group travel during the weekends. In this way, the non-commuting ITGs in this article would be defined as those with non-commuting group trips at least 4 times within a month. In 2020, 1,316,073 ITGs and 1,524,000 distinct non-commuting ITG riders were identified.

A Percolation Analysis

The spatial centrality of how ITG riders maintain their group travel is measured within the local metro system through a percolation analysis, which define a station’s central status based on its connectivity to other stations through ITG riders’ group travel flows (Sarkar et al., 2020).

Firstly, we construct an O-D matrix by counting the incoming ITG trip flows of each station, that is, the number of ITG trips from other metro stations to the given station. Then, the unique trip flows, which represent the unique values of a list of ITG trips between any two stations, are extracted. On the basis, the unique flows will be arranged in descending order and each of them will be used as a threshold for analysis. Only when the number of group trips is equal to or larger than the given threshold that these trips will be permitted.

\({T}_{p, \left(nm\right)}\) = \(\left\{\begin{array}{c}{T}_{nm}, if {T}_{nm} \ge {u}_{p}\\ 0, if {T}_{nm} < {u}_{p}\end{array}\right.\)(5)

where \({T}_{nm}\)is the number of incoming ITG trips from Station n to Station m. \({T}_{p}\) represents the O-D matrix that consists of all permitted ITG trip flows \({T}_{p, \left(nm\right)}\) given at the unique trip flow \({u}_{p}\). Accordingly, the sum (\({T}_{nm}\)), the total number of incoming ITG trips from other stations to the given Station m, is expected to decrease when the threshold value increase. Station m is considered disconnected with the system when sum (\({T}_{nm}\)) becomes zero, which means all incoming group flows of the station is smaller than the given threshold. Hence, the stations those are connected longer or with higher sum (\({T}_{nm}\)) at more thresholds could be considered as more central ones in the metro system.

A Big Picture

Throughout the year, there were 1,337,774 ITGs and 1,542,050 distinct ITG riders identified in the network. Daily ITG riders account for 10–16% of all daily riders. About half of all ITGs consist of two adult riders, and the proportion climbs to over 60% when we particularly examine ITGs with different home stations. We observe 23.5% of all ITGs include one adult and one child (Fig. 1a). In particular, we define riders’ “home” stations as those riders start their first metro trip of a day most frequently within a year. It could be found that 74% of all ITGs include two riders with the same “home” stations. These ITGs have an average of 10.96 ITG trips throughout the year, which is more than that of other ITGs with 8.24 trips on average. Considering “family” members are more likely to live together and have higher contact frequency, the ITGs including riders with the same home stations could at least partially reflect the share of “family” or “neighbors” in the dataset.

Fig 1b presents monthly ITG riders. Hong Kong suffered from four surges of locally confirmed COVID-19 caseloads during 2020 since the first such case emerged on Jan 23 (HK Gov., 2022). As expected, the numbers of both ITGs and ITG riders were significantly suppressed after COVID-19 hit the city. Specifically, the quantity of ITGs decreased from 313,441 in January to 129,858 in February. The size of ITG riders in February dropped to only 44% of that in January. In terms of persistent group riders (PGRs), who remain travelling as ITGs, 85,809 PGRs, 45% of all ITG riders in January, retained their intentional group travel in February after the local pandemic outbreak. Despite the presence of COVID-19, 9,461 PGRs remained in ITG for all the twelve months in 2020.

Destination Distribution Of Itg Riders

To explore how ITG riders’ destinations (metro stations) were impacted by the pandemic and how related potential transmission risks by destination changed across different surges, we compared the spatial distribution of those destinations before and after the outbreak of the pandemic based on a percolation analysis.

Fig 2 presents the spatial patterns of the top metro stations in terms of the most incoming ITG trips across different periods. Not surprisingly, before the local pandemic outbroke, the ITG riders centered around Kowloon South and Hong Kong Island North, where there is the highest concentration of local leisure, commercial, catering, and business facilities (Fig 2b). Specifically, stations serving Causeway Bay (Station 28), Tsim Sha Tsui (Station 3), and Mong Kok (Station 6) stood out among all the local metro stations.

After the local pandemic outbroke, only PGRs continued travelling as a group. The top metro stations that these riders visited varied notably from those for ITG riders pre the pandemic. First, Tesung Kwan O (Station 50) and its neighboring stations became one of the top 3 destinations. Second, Mong Kok was no longer within the top list. Third, Causeway Bay was still one of the most popular destinations whereas other stations such as Tai Wai emerged in the list of these destinations across the four surges of locally confirmed cases. Throughout the four surges, Tesung Kwan O was a destination that PGRs continued visiting. Given that on top of and around the station and its neighboring stations, it is one of the latest “Rail + Property” new towns of Hong Kong, one can say that the corresponding new town was quite self-contained for PGRs before and amid COVID-19, i.e., local facilities could satisfy most need of ITG riders (Cervero and Murakami, 2009) and residents have more opportunities for joint household activities when working at home and saving commuting time (Tao and He, 2021). Most if not all riders/residents in this new town only need a short metro ride to have their needs met, which exemplifies both the well-discussed neighborhood planning (Silver, 1985) and the emerging 15-minute city concept (Kissfazekas, 2022; Leung, 2021).

A Social Interactive Network Among Metro Riders During The Pandemic

Regarding ITG as a proxy for social interactions among metro riders, a metro “social network” (ITG network) is constructed. This network is facilitated by the presence and usage of the local metro system. The network’s nodes are individual ITG riders, and the links exist when any two ITG riders intentionally travel as a group.

Against the backdrop of the COVID-19 pandemic, we examine whether and how the network properties changed significantly amid the pandemic surges. Overall, the daily quantity of all riders, ITG riders, and ITGs decreased after the local pandemic outbroke, hit rock bottom in the first and second surge, and then rebounded gradually in the third and fourth surge (Fig. 3a). Particularly, Fig. 3b shows the average quantity of ITG riders by hour of a weekend day pre-pandemic and in each surge. In addition to an obvious decrease, the hourly ITG riders were fewer in the first and second surge than those in later two surges. During the surges, the pre-pandemic peak hour (5 pm) was flattened, which indicates some trips might shift from peak to off-peak hours. Despite the variations in quantity, we found that the average lag between any two ITG trips of all the ITG riders fluctuated amid the pandemic, with 82 hours during the pre-pandemic time and then 92,138,237 and 167 hours in the following four surges, respectively. It seems that the lag increased right after the local pandemic outbroke while decreased when people became aware that there could be multiple surges of locally confirmed cases.

To explore the variation among different periods of each surge, Fig. 3c depicts the COVID-19 case surges and their respective quiet (the time pre-COVID-19 or when case increase remained low), climbing (when daily confirmed cases were increasing), and declining periods (when daily confirmed cases were decreasing) in Hong Kong in 2020. The time windows of the periods were defined by referring to both the dates when local anti-pandemic measures eased or tightened (Supplementary Table S1), and the algorithm-based detection of optimal changepoints in numbers of daily locally confirmed cases (Killick et al., 2012) (Supplementary Fig. S3). All ITGs were allocated to the three different periods of each surge. Concerning ITGs across the three periods, we proposed and tested the first hypothesis:

Hypothesis 1

The size and interconnections in the ITG network varied significantly across the quiet, climbing, and declining periods of each pandemic surge.

Figure 3d shows the variation in the ratio of ITG riders to all the riders as the time approaches a surge’s zenith, when the number of daily locally confirmed cases reached the highest of the given surge. The ratio of ITG riders to all the riders in each surge fluctuated. Except for the fourth surge, a “smiling” curve in variation could be found for the first, second, and third surges. In the fourth surge, one half of the “smiling” curve could also be seen despite incomplete data. This means that there was a negative correlation between the share of ITG riders and the daily confirmed cases. Quantity of ITG riders decreased disproportionately strongly. Interestingly, even though there were a much larger number of locally confirmed cases in the third and fourth surges, the share of the daily ITG riders in these two surges were no smaller and the slope of the corresponding “smiling” curves were even flatter than their counterparts in the first and second surges. This implies that ITG riders became less sensitive to the locally confirmed cases as the fight against COVID-19 had become a long-lasting task (Petherick et al., 2021).

Student’s t-tests were performed to examine the variation in the ITG network’s interconnections, measured by the degree (k) of ITG riders, and size, measured by the average number (n) of ITG trips of each Metro rider per weekend day among the different periods of each surge. The degree (k) equals the number of other riders that a given ITG rider has ITG trips with. One hypothesis for the analysis is that k and n during the quiet period of a given surge would be lower than that within the surge.

Table 1

t-test Results for the ITG network
	Average Degree for the ITG Riders (k)
Surge	\({k}_{s}\) < \({k}_{q}\)	\({k}_{d}\) > \({k}_{c}\)	\({k}_{d}\) < \({k}_{c}\)	\({n}_{s}\) < \({n}_{q}\)	\({n}_{d}\) > \({n}_{c}\)	\({n}_{d}\) < \({n}_{c}\)
1st Surge	44.54	-	2.46***	10.81	-	37.17***
2nd Surge	14.17***	16.87***	-	11.83***	10.25***	-
3rd Surge	13.69***	9.68***	-	24.93***	-	6.89***
4th Surge	3.72	11.03***	-	6.71***	-	9.06***
t-statistic is reported. *p < 0.01, 0.01 < p < 0.05, *0.05 < p < 0.1. s denotes data during the given surge which consists of both the climbing and declining periods; q denotes data in the quiet period before the given surge; c denotes data during the climbing period and d denotes data during the declining period. Data is normalized into daily average.

Table 1 presents the t-test results, which were statistically significant for k or n models for the last three surges. This indicates that the pandemic situations reduced the ITG networks’ interconnectedness and size. Another hypothesis is that the two parameters would differ between two periods within a surge. Results of most n models show statistically significant lower values in declining periods compared with the climbing periods. This complies with the previous study that responses from Metro travel behaviors would lag behind the variation in COVID-19 transmission (Zhang et al., 2021). However, k of the last three surges is statistically significantly lower during the climbing periods than the declining periods, which implies lower interconnectedness of the network while newly confirmed cases are growing. Some analysis results incompatible with the above explanations might be impacted by implementation of local anti-pandemic measures during certain periods. For example, closure of cross-border check points occurred within the climbing period of the first surge, and bans on gatherings and international travel were implemented during the climbing periods of the second surge.

A Scale-free Itg Network

Anti-pandemic measures and people’s fear of infection could reduce physical contacts in the city. Subsequently, the ITG network could collapse. However, most social interactive networks are found to be scale-free in distribution (Barabâsi et al., 2002) and scale-free networks are usually robust when faced with malfunctional nodes and exogenous attacks (Albert et al., 2000; Dou et al., 2010). That is, a scale-free network hierarchically includes dominant nodes with a large degree followed by less-dominant nodes. In the network, nodes with a small degree account for vast majority of the network. Therefore, the overall structure of the network usually would not be affected if we randomly remove a small fraction of nodes. We consider that this would also be true for the ITG network amid the pandemic. Such properties would increase metro travel anxiety of many people and notably reduce outgoings of many metro riders. Hence, the following two hypotheses are proposed:

Hypothesis 2

The structure of the ITG network remains robust amid the pandemic.

The topology of the network remains relatively robust as a whole despite the pandemic. Figure 4a compares the complementary cumulative distribution functions (CCDF) of ITGs network before the pandemic outbroke with those in different COVID-19 surges. The ITG networks perform statistically significant in power-law distribution spanning all periods with p-value of KS-tests smaller than 0.01. The α decreased after the pandemic outbroke and then fluctuated across surges around 3 to 5. This indicates that riders with a higher degree account for a larger proportion of ITG riders of a local surge compared to the pre-pandemic time.

Hypothesis 3

The ITG network increases in time and follows a scale-free distribution.

Shown in Fig. 4b, interconnections of the network can be measured by two indicators: average degree (k) and the relative size of the largest cluster (S). Despite the k mentioned above, the S is the ratio between the total number of riders in the largest cluster and that in the whole network. Both k and S increased in an approximately linear manner. The growth of k has three contributors. One, new ITG riders made ITG trips together and joined the existing ITG network. Two, these new riders travelled with existing ITG riders so new ITGs trips emerged. Three, new ITG trips emerged among existing ITG riders. A cluster of the ITG network represents a subset of ITG riders who became connected because of the presence of ITG trips. The ITG network of metro riders is embedded in the large social interactive network in the city. A larger S indicates that an increasingly dominant cluster in the ITG network and possibly in the large social interactive network had emerged.

As for the network structure, we used the whole year’s ITG dataset to examine distribution of frequency and degree of the ITG riders. As shown in Fig. 4c, the trip frequencies of most riders in the network typically increases as the degree increases. For all ITG riders, their trip frequencies’ mean and maximum are 28 and 1,132, respectively. Figure 4d presents the degree distribution in the logarithmic binning format. The distribution of all the ITG riders and that of non-commuting ITG riders remain similar. The average degree of all the ITG riders is 1.73 and the max value is 52, lower than 150, the suggested cognitive limit to the number of people with whom one could maintain stable social relationships (Dunbar, 1993). In particular, we found that the probability P(x) that a rider is connected with at least x other riders in the ITG network decays following a power law, which can be expressed as P(x) ~ \({x}^{-\alpha }\), where α = 5.67. The Kolmogorov-Smirnov test (KS-test) was performed to examine whether the network’s degree distribution statistically follows the power-law distribution. The test produces a p-value smaller than 0.01. This result demonstrates that the ITG network indeed resembles other social networks with a scale-free distribution in degree.

Who Maintains The Itg Network?

According to what we know about the structural robustness of scale-free networks (Albert et al., 2000), a probable explanation for the ITG network’s relative stability amid the pandemic is the continued presence of PGRs, who remained active in ITG travel despite of the surges in locally confirmed cases.

Table 2

t-Test Analysis for Persistent Group Riders (PGRs) in the ITG network
	Average Degree for PGRs (k)		Average ITG trips for PGRs (n)
Surge	\({k}_{s}\) < \({k}_{q}\)	\({k}_{np}\) < \({k}_{p}\)	\({n}_{s}\) < \({n}_{q}\)	\({n}_{np}\) < \({n}_{p}\)
1st Surge	19.82***	130.54***	7.90	52.18***
2nd Surge	67.70***	115.46***	52.17***	80.02***
3rd Surge	122.69***	142.99***	97.44***	60.09***
4th Surge	116.11***	141.15***	79.67***	20.69***
t-statistic is reported. *p < 0.01, 0.01 < p < 0.05, *0.05 < p < 0.1. s denotes data during the given surge which consists of both the climbing and declining periods; q denotes data in the quiet period before the given surge; p denotes data on PGRs occurring before the local pandemic outbroke and then remaining in the network within all four surges; np denotes data on non-PGRs within the given surge.

Table 2 presents the t-test results regarding the degree and number of the PGRs and non-PGRs. The results corroborate the hypothesis that the degree and number of PGRs would be significantly higher than those of non-PGRs in any given surge (p < 0.01). Not surprisingly, even the PGRs were impacted by the surge of the locally confirmed cases–their k and n in the four climbing periods were mostly smaller than those in the corresponding quiet periods. To further explore the role of PGRs in maintaining the ITG network’s robustness, Hypothesis 4 was proposed.

Hypothesis 4

PGRs are essential vertices those influenced the properties of the ITG network.

To test this hypothesis, we conduct an experiment to compare how removal of PGRs, who remain active in group travel amid the pandemic, and the same number of riders randomly drawn from the ITG dataset would affect the interconnectedness in the network. The interconnectedness can be measured by k and S. We validated that this is the case in Hong Kong. As for the experimental failures for measurement on k, we call removing PGRs who remain in the ITG network up to a certain period “Failure 1” and removing the same amount of randomly drawn riders from the network “Failure 2”. S indicates the largest number of riders that are potentially affected by an infected rider in the ITG network. Taking the “social network” of the local metro riders in the whole year as the subject, we separately remove the same fraction (number) of different subsets of riders for 50 times in total and up to 19% of the network size (the proportion of the largest cluster in the whole-year network.), to investigate how that would influence S. Our subsets of ITG riders include those that were randomly selected throughout the year (Subset 1) and PGRs presented in at least one (Subset 2), two (Subset 3), three (Subset 4), and four (Subset 5) surges throughout the year. Removing the different subsets allows us to simulate different failures of the network and to see related consequences.

Figure 5a shows k becomes significantly lower under the “Failure 1”, while the k’s for “Failure 2” are almost the same as those without removing any riders from the ITG network in the same period. Similarly, if we assume it is highly likely for ITGs and related interactions to transmit viruses when one ITG rider has been infected, S would indicate the largest number of riders potentially affected by an infected rider in the ITG network. Figure 5b shows that S apparently remains more stable when randomly removing ITG riders (Subset 1), compared with removing PGRs (Subset 2–5). For the PGRs in different subsets, the more frequently they presented in the four surges, the more influential they were in S. Removing those PGRs who presented in all the four surges (Subset 5) resulted in the smallest S for failures. These indicate that the PGRs played a more influential role in interconnections of the network than wandering riders who entered or left during the given period.

Cities serving as the magnet and mecca for billions of people is not by coincidence. They prompt more impactful social interactions among people, which are foundations for meaningful coordination, collaborations, and innovations. Over time, cities as mankind’s greatest “inventions”, have become increasingly richer, smarter, greener, healthier, and happier because of these collective activities (Glaeser, 2013). Group travel constitutes one special form of these activities and serves as prerequisites for others. By exploiting a year’s smartcard data in Hong Kong, we constructed an ITG network of metro riders, which generated the following insights and findings.

First, we demonstrate that the longitudinal Metro smartcard data can help us identify an ITG network of metro riders, which serves as a proxy for the general ITG network by the local residents. Compared with the existing scholarship, the work involves a much larger sample size with a one-year dataset, which enables us to conduct the study of citywide physical social interactions across more spatiotemporal units. These smartcard data processing analytics could be adapted to examine social contacts and interactions elsewhere, and facilitate cross-platform social network data mining and fusion.

Second, probable physical interactions and mobility patterns of the ITGs and PGRs amid the pandemic were systematically examined and tracked. We reconfirmed that physical social interactions largely decreased in the city since the pandemic’s outbreak (Chou et al., 2020; Lee et al., 2022). Using quantitative evidence, we found some general laws of physical group interactive behaviors amid the pandemic, which are consistent with those for virtual social networks (Barabâsi et al., 2002). The structure and dynamics of the ITG network among metro riders provides a snapshot of the large-scale physical social interactions in cities from the perspective of transportation, which we had knew little so far. With extra information such as infection cases among riders, our work can also be extended to simulate virus transmission routes via physical contacts in the metro system.

Third, we found that the interconnections of the ITG networks are determined by a subset of riders, that is, PGRs in our study. Given that virus spread often relies on a social interactive network, this argues for investigating whether PGRs are more likely to spread or succumb to the virus. Identifying PGRs thus can be useful in more targeted preventive measures and in effectively tracing important close contacts of the infected in the transport system.

Fourth, the changing and stable popular destinations among ITG riders and PGRs pre- and amid COVID-19 indicates that a portion of metro riders remain their ITG trips while others might not. The former was those mostly riders living and socializing around a small set of stations, e.g., the self-sustained new town stations. It was highly likely that certain stations and their surroundings had better supply of facilities and opportunities than others to satisfy the group traveling and meeting demands. How to provide a decent amount of those facilities and opportunities to have people group-based need met is thus an important issue amid pandemic. Our work has illustrated how to identify inequities in the distribution of essential facilities and opportunities (Jay et al., 2020). Given the above-mentioned benefits of physical social interactions, not addressing those inequities can further result in decreased quality of life and increased mental problems in certain communities amid the pandemic.

Limitations of this research should also be mentioned. Although the ITGs identified from the ,etro smartcard data could partially represent the local physical social interactions in transit-reliant cities like Hong Kong, they overlook interactions among people who use other modes of travel such as cycling, buses, private cars or walking. The smartcard data alone only allows us to identify probable ITGs. Without collecting more and travel information about metro riders, we still know little about how many ITGs there were in total, who the ITG riders are, why they are traveling together, and how much they actually benefited or lost from group travel.

In this work, we explored intentional travel groups (ITGs) in Hong Kong Metro before and amid the COVID-19 pandemic in 2020 using a smartcard dataset. We assessed the spatial dynamics of ITGs and persistent group riders (PGRs) in different periods of the local pandemic, as well as examined the properties and changes of the social interactive networks formed by these ITGs. The key findings related to our hypothesis include: (1) as for four local case surges in 2020, the size and interconnections of the ITG network were typically lower during the quiet period of a given surge than that within the surge. Interestingly, ITG riders became less sensitive to the caseload increase in later surges. (2) The topology of the network, that is, the distribution of its degree, remains relatively robust as a whole despite different surges the pandemic. (3) the network size and interconnections cumulatively increased in time, and the degree of the ITG network follows a scale-free distribution. (4) PGRs influenced the network’s interconnections more significantly than non-PGRs. The findings concerning metro riders, a subset of local residents help us to understand the properties of the citywide social interactive network and how they changed before and after the outbreak of COVID-19 and varied during the local pandemic.

At least two pieces of work could be done to extend the output presented in this work. One is to apply the ITG problematization and modelling into the transit-usage-and-virus-spread modelling, and to examine whether involving ITG could improve the accuracy of predicting transmission routes in the transit system. Two is to consider group travel consisting of three or more metro and even transit riders. In our current work, we only consider two-rider group travel in the metro system. This extension would allow us to see a fuller picture of social interactions facilitated by group travel among transit riders.

Ethical Approval

Not applicable.

Competing interests

The authors declare no competing interests.

Author contributions

J. Z. and M. Z. designed research; M. Z, J. Z., S. L., J. W., and H. M. performed research and analyzed the data and contributed new reagents and analytic tools; J. Z., M. Z. and D. L. wrote and reviewed the paper.

Funding

The research underlying this paper was financially supported by the Platform Technology Funding, University of Hong Kong (URC012530226) and General Research Fund of Hong Kong (Grant 17603220).

Availability of data and materials

The smartcard data that support the findings of this study are available from the Hong Kong MTR Corporation Limited (MTR) but are not publicly available. Restrictions apply to the availability of these data, which were used under license for the current study. Data are, however, available from the corresponding author upon reasonable request and with permission of the Hong Kong MTR.

Albert R, Jeong H and Barabási A-L.: Error and attack tolerance of complex networks. Nature 406(6794): 378-382 (2000)
Amram R, Ravreby I, Trainin N, et al.: Fear of being near: Fear supersedes sociability when interacting amid a pandemic. Personality and individual differences 187: 111404 (2022)
Arentze, T., Timmermans, H.: Social networks, social interactions, and activity-travel behavior: a framework for microsimulation. Environment and Planning B: Planning and Design 35, 1012-1027 (2008)
Barabâsi A-L, Jeong H, Néda Z, et al.: Evolution of the social network of scientific collaborations. Physica A: Statistical Mechanics and its Applications 311(3-4): 590-614 (2002)
Benzell SG, Collis A and Nicolaides C.: Rationing social contact during the COVID-19 pandemic: Transmission risk and social benefits of US locations. Proceedings of the National Academy of Sciences 117(26): 14642-14644 (2020)
Block P, Hoffman M, Raabe IJ, et al.: Social network-based distancing strategies to flatten the COVID-19 curve in a post-lockdown world. Nature Human Behaviour 4(6): 588-596 (2020)
Carlsen HB, Toubøl J and Brincker B.: On solidarity and volunteering during the COVID-19 crisis in Denmark: the impact of social networks and social media groups on the distribution of support. European Societies 23(sup1): S122-S140 (2021)
Carrillo C and Flores MA.: COVID-19 and teacher education: a literature review of online teaching and learning practices. European Journal of Teacher Education 43(4): 466-487 (2020)
Cervero R and Murakami J.: Rail and property development in Hong Kong: Experiences and extensions. Urban Studies 46(10): 2019-2043 (2009)
Chang S, Pierson E, Koh PW, et al.: Mobility network models of COVID-19 explain inequities and inform reopening. Nature 589(7840): 82-87 (2021)
Chatterjee S, Ghosh K, Banerjee A, et al.: Forecasting COVID-19 Outbreak Through Fusion of Internet Search, Social Media, and Air Quality Data: A Retrospective Study in Indian Context. IEEE Transactions on Computational Social Systems (2022)
Chou W-P, Wang P-W, Chen S-L, et al.: Voluntary reduction of social interaction during the COVID-19 pandemic in Taiwan: Related factors and association with perceived social support. International journal of environmental research and public health 17(21): 8039 (2020)
Dou B-L, Wang X-G and Zhang S-Y.: Robustness of networks against cascading failures. Physica A: Statistical Mechanics and its Applications 389(11): 2310-2317 (2010)
Dubernet, T., Axhausen, K.W.: Implementing a household joint activity-travel multi-agent simulation tool: first results. Transportation 42, 753-769 (2015)
Dunbar RI.: Coevolution of neocortical size, group size and language in humans. Behavioral and brain sciences 16(4): 681-694 (1993)
Eagle N, Pentland A and Lazer D.: Inferring friendship network structure by using mobile phone data. Proceedings of the National Academy of Sciences 106(36): 15274-15278 (2009)
Feng S and Kirkley A.: Integrating online and offline data for crisis management: Online geolocalized emotion, policy response, and local mobility during the COVID crisis. Scientific reports 11(1): 1-14 (2021)
Florida R, Rodríguez-Pose A and Storper M.: Cities in a post-COVID world. Urban Studies. 00420980211018072 (2021)
Glaeser EL.: Triumph of the city: How our greatest invention makes us richer, smarter, greener, healthier, and happier (an excerpt). Journal of Economic Sociology 14(4): 75-94 (2013)
Glaeser EL.: Urban resilience. Urban Studies 59(1): 3-35 (2022)
Gliebe, J.P., Koppelman, F.S.: A model of joint activity participation between household members. Transportation 29, 49-72 (2002)
HK Gov.: Coronavirus Disease (COVID-19) in HK (Geodatabase) (2022)Available at: https://chp-dashboard.geodata.gov.hk/covid-19/zh.html.
HK Transport Department.: The Government of the Hong Kong Special Administrative Region (2022) Available at: https://www.td.gov.hk/en/transport_in_hong_kong/public_transport/railways/index.html.
Jay J, Bor J, Nsoesie EO, et al.: Neighbourhood income and physical distancing during the COVID-19 pandemic in the United States. Nature Human Behaviour 4(12): 1294-1302 (2020)
Killick R, Fearnhead P and Eckley IA.: Optimal detection of changepoints with a linear computational cost. Journal of the American Statistical Association 107(500): 1590-1598 (2012)
Kissfazekas K.: Circle of paradigms? Or ‘15-minute’neighbourhoods from the 1950s. Cities 123: 103587 (2022)
Lee JH, Luchetti M, Aschwanden D, et al.: Perceived changes in social contact during COVID‐19 pandemic in the United States. Personal Relationships 29(1): 59-76 (2022)
Leung P.: How '15-minute cities' will change the way we socialize (2021) Available at: https://www.bbc.com/worklife/article/20201214-how-15-minute-cities-will-change-the-way-we-socialise.
Marinucci M, Pancani L, Aureli N, et al.: Online social connections as surrogates of face-to-face interactions: A longitudinal study under Covid-19 isolation. Computers in Human Behavior 128: 107102 (2022)
McQuire S.: Geomedia: Networked cities and the future of public space. John Wiley & Sons (2017)
Nguyen, M. H.: Factors influencing home-based telework in Hanoi (Vietnam) during and after the COVID-19 era. Transportation, 48(6), 3207-3238 (2021).
Petherick A, Goldszmidt R, Andrade EB, et al.: A worldwide assessment of changes in adherence to COVID-19 protective behaviours and hypothesized pandemic fatigue. Nature Human Behaviour 5(9): 1145-1160 (2021)
Sarkar S, Wu H and Levinson DM.: Measuring polycentricity via network flows, spatial interaction and percolation. Urban Studies 57(12): 2402-2422 (2020)
Sharmeen, F., Arentze, T., Timmermans, H.: Predicting the evolution of social networks with life cycle events. Transportation 42, 733-751 (2015)
Silver C.: Neighborhood planning in historical perspective. Journal of the American Planning Association 51(2): 161-174 (1985)
Sullivan O, Gershuny J, Sevilla A, et al.: Using time-use diaries to track changing behavior across successive stages of COVID-19 social restrictions. Proceedings of the National Academy of Sciences 118(35): e2101724118 (2021)
Sun L, Axhausen KW, Lee D-H, et al.: Understanding metropolitan patterns of daily encounters. Proceedings of the National Academy of Sciences 110(34): 13774-13779 (2013)
Tao, S., He, S.Y.: Job accessibility and joint household travel: a study of Hong Kong with a particular focus on new town residents. Transportation 48, 1379-1407 (2021)
van den Berg, P., Arentze, T., Timmermans, H.: A multilevel analysis of factors influencing local social interaction. Transportation 42, 807-826 (2015)
WHO.: Coronavirus disease (COVID-19): How is it transmitted? (2021) Available at: https://www.who.int/news-room/questions-and-answers/item/coronavirus-disease-covid-19-how-is-it-transmitted.
Zhang N, Jia W, Wang P, et al.: Changes in local travel behaviour before and during the COVID-19 pandemic in Hong Kong. Cities 112: 103139 (2021)
Zhou J, Yang Y, Ma H, et al.: “Familiar strangers” in the big data era: An exploratory study of Beijing metro encounters. Cities 97: 102495 (2020)
Zhu K, Yin H, Qu Y, et al.: Group travel behavior in metro system and its relationship with house price. Physica A: Statistical Mechanics and its Applications 573: 125957 (2021)

No competing interests reported.

SupplementaryInformation.docx

Download PDF

Version 2

posted

You are reading this latest preprint version

Intentional Travel Groups and Social Networks during the COVID-19 Pandemic

Status:

Version 2

Abstract

Figures

Introduction

Data And Methods

Metro Smartcard Data

Identify intentional travel groups (ITGs)

Non-commuting ITGs

A Percolation Analysis

Results

A Big Picture

Destination Distribution Of Itg Riders

A Social Interactive Network Among Metro Riders During The Pandemic

A Scale-free Itg Network

Who Maintains The Itg Network?

Discussion

Conclusions

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 2