key: cord-1041042-o93jdd8r
authors: Zhao, Pengjun; Gao, Yukun
title: Public transit travel choice in the post COVID-19 pandemic era: An application of the extended Theory of Planned Behavior
date: 2022-04-05
journal: Travel Behav Soc
DOI: 10.1016/j.tbs.2022.04.002
sha: f1c86ae4204e286f3e2396615ac91ae046207a62
doc_id: 1041042
cord_uid: o93jdd8r

It is widely reported that the COVID-19 pandemic has reduced ridership and brought severe challenges to urban public transit systems in many countries. The impact of the COVID-19 pandemic on individual people’s choice of public transit may continue for a while after the peak of the crisis. However, there is insufficient detailed knowledge of how individuals respond in the post-pandemic context and make choices on public transit travel. This paper contributes fresh evidence for this by looking at Beijing as a case. The theoretical framework of the Theory of Planned Behavior is used to model individuals’ public transit travel choice-making processes along with three additional constructs representing the impact of the pandemic and the nature of urban mobility behaviors, namely perceived knowledge of COVID-19, the psychological risks of COVID-19, and travel habits. Structural equation modeling is used in model estimation. We point out that there may be potential differences between the effects and meanings of model constructs in the post-pandemic context and in normal daily context. Interestingly, despite the higher psychological risk’s negative effects, higher perceived knowledge of COVID-19 has significantly positive effects on people’s decision-making processes. A strong pre-pandemic personal habit of traveling by public transit has significant and positive effects on post-pandemic intention and perceived behavioral control. Group comparisons show that “captive” transit users have higher psychological risk of COVID-19 than “choice” transit users, yet their transit use decisions are less influenced by it. Based on the modeling results, more behavioral experiments are needed to further inform efficient policy-making.

Public transit is essential to alleviating congestion and reducing carbon emissions in cities. As one of the most dynamic public facilities, it plays an important role in stimulating the urban economy and providing affordable mobility for inhabitants. However, the outbreak of the coronavirus pandemic has had severe negative effects on public transit systems worldwide.

Demand has decreased dramatically as risk perception has risen among passengers, while social distancing regulations and service providers' loss minimization strategies have imposed serious constraints on the supply side (Gkiotsalitis and Cats, 2020) . According to Google Mobility Reports data (Google, 2020) , the drop in activity levels for public transit hubs during lockdown periods was 50%-85% in the world's major economies. Moreover, evidence suggests that the impact of COVID-19 on travel behavior is ongoing in the postlockdown period (Gkiotsalitis and Cats, 2020) . Passengers are reluctant to take public transit, and they are more anxious about public hygiene issues than before the pandemic (Beck and Hensher, 2020; Labonté-LeMoyne et al., 2020) .

There are some studies about public transit systems in the pandemic moment. Most of the existing literature focuses on analyzing the characteristics of and modeling the dynamics of (Li, 2020; Liu et al., 2020) changes in the public transit system from an aggregated and sometimes top-down view. For example, Zhang et al. (2020) found that Hong Kong Mass Transit Railway ridership reduced by 52.0% during the pandemic, and that people's travel behavior changes reduced the close contact rate and slowed the spread of the COVID-19 virus. Liu et al. (2020) fitted US transit demand data in the early stage of the pandemic using a logistic anti-growth process. Combining epidemiological models with the transportation system, Li (2020) built a robust simulator for the spread of China that accounts for cross-infection in public transit systems. Furthermore, there is a growing literature on mass transit planning policy implications, such as a study by Bustamante et al. (2020) that estimated the maximum capacity of the busiest bus rapid transit station in Lima, Peru under COVID-19 security restrictions.

As many economies are now heading for recovery from this unprecedented crisis, studies on personal travel choice of public transit in the post COVID-19 pandemic era from a bottom-up perspective are becoming more urgent. Some studies have used individual or household measures to examine the influence of COVID-19 on public transit mode choice. Anke et al. (2021) investigated the changes in Germans' micro-mobility patterns via an online survey.

They found that the main reason people shifted away from public transit was risk perception, while the effect of lockdown on travel mode choice was minor. In a survey by Przybylowski et al. (2021) , the most frequently mentioned reasons for reducing transit use were change of job or remote working and risk perception. Dong et al. (2021) built a model to predict public transit passengers' satisfaction levels. An influence chain can be derived from their results: those who pay more attention to COVID-19-related information are psychologically closer to the pandemic and are more anxious, feel less safe, and are less satisfied with transit service.

Although some studies have explored public transit in the post-pandemic moment, one important research gap still needs to be filled: how do individual people decide whether to travel by public transit in the post-pandemic context, and what is the pandemic's influence on their decision-making processes? Few previous studies have tapped into the individual's choice-making mechanism regarding transit use, and among those that have, the relationships between attitudes and intentions or behaviors has not been thoroughly discussed. There is thus an urgent need for systematic, theory-guided behavioral studies to uncover the internal structure of passengers' decision-making processes (Gkiotsalitis and Cats, 2020; Liu et al., 2020) .

We aim to fill the research gap by using a behavioral model grounded in the Theory of Planned Behavior (TPB; Ajzen, 1991) to investigate passengers' intended and actual transit use in the post-pandemic era. TPB is one of the most commonly applied theoretical frameworks in the social and behavioral domain (Bosnjak et al., 2020) . It has also been widely tested in urban transport studies (Bamberg et al., 2003a; Lo et al., 2016; Schoenau and Müller, 2017; Borhan et al., 2019) . According to TPB, as Figure 1 shows, the volitional factors attitude (ATT) and subjective norm (SN) and the nonvolitional factor perceived behavioral control (PBC) are the proximal determinants of intention, while intention and PBC act as the direct predictors of actual behavior (Ajzen, 1991) . ATT indicates how positively or negatively the individual perceives the behavior in question (Zailani et al., 2016; Schoenau and Müller, 2017) . SN captures the individual's perceived social pressure to perform the behavior (Bosnjak et al., 2020; Han et al., 2020) . To be specific, SN could include injunctive norm and descriptive norm, referring respectively to individuals' beliefs that important others support the behavior or perform the behavior themselves (de Leeuw et al., 2015) . PBC relates to individuals' confidence in themselves to achieve a given behavior. It involves an evaluation of one's own strengths and limitations (Ajzen, 1991; Han et al., 2020) . Intention stands for the individual's motivation to perform a certain behavior. It reflects how hard one is willing to try to carry out the behavior (Ajzen, 1991) .

To analyze the pandemic's influence on people's decision-making process, we study the effects of two additional constructs on the original TPB framework: perceived knowledge of and psychological risk of COVID-19 (PRC). PRC and PKC are introduced into the model because concerns about infection and related negative feelings are the most relevant reasons for transit riders to change their travel behavior, while awareness of pandemic-related facts serves as a reference when people make decisions. Apart from the pandemic's influence, since daily travel behavior is regularly performed, it is necessary to study the effect of people's pre-pandemic transit use habits on their post-pandemic travel choices; thus, travel habit (TH) is also absorbed into the modeling framework.

Based on the above, the objectives of the present study are (1) to untangle the decisionmaking mechanism of public transit passengers in the post-pandemic context using the behavioral framework of TPB, (2) to examine the effects of two pandemic-related constructs (perceived knowledge of COVID-19 and psychological risk of COVID-19) on individuals' decision-making process, and (3) to analyze the role of one construct reflecting the intrinsic nature of urban mobility (travel habit). In addition, in the post-pandemic context, the decision-making process of those who travel by transit willingly ("choice" users) and those who are less willing but must travel by transit ("captive" users) may be different. Therefore, another objective of the present study is (4) to study the difference between the decisionmaking mechanisms of "choice" and "captive" transit users. This paper uses Beijing as a case. On the national level, China has entered the post-pandemic phase. New confirmed cases of COVID-19 dropped to less than 100 per day in early March 2020 and remained below 100 until early January 2021, when a minor wave of infection slightly raised the count. Zooming in, Beijing is a typical post-pandemic city. Many enterprises resumed production in March 2020. As of September 2020, all schools and higher education institutions returned to in-person instruction and the hospitality and entertainment industry operated at full capacity. Beijing has a large-scale public transit system consisting of bus and rail. It is usually crowded, especially during workdays. In 2019, the modal split rate for Beijing's public transit system was 31.8% (higher than the rate of 22.6% for private cars), making it an indispensable part of urban transportation. At the peak of the pandemic, Beijing's public transit system remained in operation with many preventive measures taken, but public transit ridership reduced to about 20% of the same period in 2019. Basically, public vehicles and transit stations were sterilized daily, strict flow restrictions and occupancy rate controls were enforced, passengers' body temperatures were taken, and they were required to wear masks. People only went out for essential trips, and many transit lines had very few passengers. As restrictions eased, ridership bounced back to around 80% in the post-lockdown period. To sum up, Beijing makes an ideal case for studying post-pandemic transit use behavior.

The remainder of this study is organized as follows: Section 2 presents a literature review and our research hypotheses. Section 3 offers a research design including a brief introduction of the case city, data collection, and measurements. Section 4 provides our modeling results and analysis. Section 5 discusses the study's key findings, practical implications, limitations, and future research prospects. Section 6 summarizes our conclusions. 

During the pandemic, public transit modes were among the most affected modes of urban transportation. People avoided public vehicles to keep social distance, causing transit ridership to drop dramatically (Google, 2020; Zhang et al., 2020) , and transport modal split to shift away from public transit. Anke et al. (2021) found that in Germany, regular passengers of public transit turned to walking, cycling, or using private cars. In China, the percentages of people using public transit, bicycles, and electric bicycles decreased during the pandemic, while the percentages traveling by walking or cars (either private cars or taxis) significantly increased (Jiang et al., 2020) .

Infection risk is frequently linked with people's reluctance to travel by transit. On the one hand, compared with other modes, there was a higher level of perceived risk associated with public transit during the pandemic. In a study of ten countries on six continents, Barbieri et al. (2021) found that airplanes and buses had the highest perceived risks of infection among all transportation modes. Beck and Hensher (2020) found perceptions that trains and buses were the least comfortable options for Australians during the pandemic. On the other hand, individuals more mentally susceptible to risk may be more reluctant to take public transit. As indirect evidence, Asselmann et al. (2020) found that students with less emotional stability tend to take public transit less often. While risk has been explored in some surveys conducted during the pandemic, an opposite concept, safety perception, has also been used in other studies (e.g., Dong et al., 2021; Przybylowski et al., 2021) . Risk and safety are two sides of the same coin.

In the post-pandemic era, although government restrictions have been lifted, passengers might still tend to trade time for less crowding (Gkiotsalitis and Cats, 2020) . A study in Korea compared transit users' crowding impedance levels before and after the pandemic and found that the crowding impedance in November 2020 was higher than in October 2018 (Cho and Park, 2021) . Another study in Australia found that in the early days of the post-lockdown period, people were still reluctant to travel by transit (Beck and Hensher, 2020) . It is possible that in the long run, the negative effect of infection risk will fade away. For example, a survey by Ehsani et al. (2021) found that people's expected post-pandemic travel by public transit did not significantly differ from their pre-pandemic travels. However, in the short run, the impact of COVID-19 on people's choice to use public transit still stands out.

Many existing literatures use traditional discrete choice models to study travel mode choice (Li and Zhao, 2015; Zhao and Wan, 2020) . These studies have identified factors that influence people's mode choice behavior. However, it is also important to analyze how these factors affect the internal mechanisms of individual decision-making processes (Zhao and Zhang, 2018) . In recent years, transportation researchers have begun to use behavioral variables to reveal the internal structure of the travel choice black box and to improve model explanatory power (Temme et al., 2008) . The present study extends TPB to study people's post-pandemic public transit mode choices.

TPB assumes that ATT, SN, and PBC directly influence intention, and that intention and PBC have direct effects on behavior (Ajzen, 1991; Liu et al., 2017; Han et al., 2020) . Many case studies exploring transit use behavior (Heath and Gifford, 2002; Nordfjaern et al., 2014a; Zailani et al., 2016) , switching intention from private vehicles to public transit (Chen and Chao, 2011; Lourdes and Dima, 2019) and the related topic of low-carbon transportation behavior (Eriksson and Forward, 2011; Liu et al., 2017) have shown the capacity of TPB in modeling individual mobility. Note here that much of the TPB-assisted literature, including many of the studies mentioned above, was interested only in explaining intentions. However, to get a broader picture of transit use in the post-pandemic context, it is necessary to bring the behavior construct into the model. Thus, the following hypothesis is adopted: H1. The relationships between attitude, subjective norm, perceived behavioral control, intention, and behavior in the original TPB framework still stand. This specifically suggests that (a) the more positive the attitudes are regarding using public transit; (b) the higher the subjective norm to use public transit is; (c) the stronger the perceived behavioral control is regarding using public transit, the stronger the intention is to travel by public transit in the post-pandemic era. Also, (d) the stronger the intention to travel by public transit; (e) the stronger the perceived behavioral control regarding using public transit, the more likely the behavior of traveling by public transit in the post-pandemic era is to occur.

Perceived knowledge is an important factor that can enhance the reasoning-based component of our theoretical framework. Perceived knowledge represents the individual's awareness of the given behavior and its consequences. In a study on museum visitors' pro-environmental behavior, Han and Hyun (2017) found that more environmental knowledge enables people to gain further access to the emotions they might have once they perform or avoid the behavior, helping them become surer of what to do. Sufficient awareness can enhance people's judgements and self-confidence when making decisions. This suggests that perceived knowledge is vital in determining people's intentions and behavior. Therefore, perceived knowledge of COVID-19 is included in the research model. It is the participants' awareness of COVID-19-related facts, represented by their self-rated knowledge level of COVID-19, and it is consistent with definitions in previous studies that also looked at post-COVID behaviors, such as Han et al. (2020) .

Under the TPB framework, perceived knowledge of COVID-19 influences ATT and PBC by encouraging objectivity and confidence in individual people. To be specific, those who are more aware of the facts about the pandemic are more likely to rule out the influence of irrational thoughts and make rational risk assessments; thus, they tend to hold more positive attitudes towards transit use. In the post-pandemic context, public transit stands out from other daily travel modes in that it is still most associated with infection risks and is subject to the strictest preventive measures; therefore, when considering how easy or difficult it is to take public transit, people tend to relate to the troubles or difficulties induced by control measures in addition to accessibility factors. Since people with higher PKC are more familiar with and more understanding towards suitable protection measures and more confident in dealing with additional trouble, they may find it easier to use public transit.

Many empirical studies in the fields of health-related behaviors (Meadowbrooke et al., 2014) , tourist behaviors (Han et al., 2020) , and pro-environmental behaviors (Han and Hyun, 2017; Zhu et al., 2020) have analyzed the role of perceived knowledge. For example, perceived knowledge's effects on ATT and PBC are found in Meadowbrooke et al. (2014) and Zhu et al. (2020) . Based on the above, the present study examines PKC's influence on ATT and PBC through the following hypothesis. The direction of PKC's effects is not presupposed because it still needs exploration:

H2. In the post-pandemic era, perceived knowledge of COVID-19 affects (a) attitudes regarding using public transit; and (b) perceived behavioral control regarding using public transit.

In the post-pandemic context, risk perception is another core factor influencing people's attitudes and perceptions regarding public transit, one that has been noted but not thoroughly discussed in the existing literature. The present study introduces a most relevant facet of perceived risk, psychological risk, into the reasoning-based part of the research model. In our study context, psychological risk of COVID-19 means feelings of mental discomfort or anxiety caused by traveling in a public vehicle (Garner, 1986; Labonté-LeMoyne et al., 2020) . It reflects the negative effect of the pandemic on the individual's peace of mind. The concept is derived from Perceived Risk Theory: a commonly accepted notion is that there are six facets of risk stemming from the general idea of perceived risk, namely social, financial, physical, performance, time, and psychological risk (Garner, 1986) . In public transportation, physical, psychological, and performance risks stand out among the other kinds of risks. A regression analysis by Labonté-LeMoyne et al. (2020) found that these three risks explain 34% of the variation in intended transit use among commuters in Canada. The present study mainly focuses on psychological risk because it is the most readily accessible facet of perceived risk for public transit passengers. After the peak of the pandemic, people's psychological vulnerability has become a major obstacle to restoring social and economic order, and quantifying its influence is of practical significance.

The results of existing research (Anke et al., 2021; Przybylowski et al., 2021) suggest that perceived risk is an important influencer of people's transit mode use decisions in the pandemic moment. To incorporate the psychological risk of COVID-19 into the TPB framework, we assume it can influence people's intentions and behavior through ATT.

People with higher levels of psychological risk tend to exaggerate the dangers of using public transit in the post-COVID context; thus, they hold less positive attitudes towards transit use.

By proposing the following hypothesis, we can test PRC's effects:

H3. In the post-pandemic era, the higher the psychological risk of COVID-19 is, the less positive attitudes are regarding using public transit.

As daily travel mode choices are usually performed repeatedly in stable contexts, they are considered to be habitual (Verplanken and Aarts, 1999; Gärling et al., 2001; Thogersen, 2006; Gardner, 2009 ). Therefore, we add travel habit into the research model. Habit is a nondeliberate behavioral response cued by a certain stable situation (Triandis, 1977; Wood and Rünger, 2016; Haggar et al., 2019) , and its features include a history of repetition, automaticity, and (for some kinds of behaviors) expressing identity (Verplanken and Orbell, 2003; Wood and Rünger, 2016) . In a broader sense, it is also referred to as inertia effect or psychological inertia (Goodwin, 1977; Gärling and Axhausen, 2003; González et al., 2017; Gao et al., 2020; Thorhauge et al., 2020) . Friedrichsmeier et al. (2013) categorized the concept of habit into two embranchments based on the level of generality: the association view and the script-based view. The "associationist" approach (eg. Wood et al, 2005; Neal et al., 2006) understands habit as a connection between specific context and behavior formed through repetitively performing behavior in context, while the "schema or script-based" approach (eg. Verplanken et al., 1994; Verplanken et al., 1997; Verplanken and Aarts, 1999; Fujii and Gärling, 2003) regard habit as a behavioral script linked to a behavioral goal, and this script can be built on past enactment or socialization, and can be effective in a wider range of situations.

Generally, in a stable environment, people with stronger habits of using a specific travel mode or weaker habits of using complementary modes are more inclined to take this travel mode in the future. Previous case studies and meta-analysis studies on individual mobility using TPB confirmed this hypothesis (Thøgersen, 2006; Nordfjaern et al., 2014a; Şimşekoğlu et al., 2015; Zailani et al., 2016; Zhao et al., 2016; Hoffmann et al., 2017) . However, the effect of habit is more complicated when decision context changes. According to the habit discontinuity hypothesis, a disturbance in decision context can weaken individuals' old habits and provide a window for them to consciously (re)consider their choice of action and even establish new habits (Verplanken et al., 2008; Walker et al., 2015; Verplanken and Roy, 2016) . Studies have found that old travel habits are weakened after context shifts such as work relocation (Walker et al., 2015) and home relocation and public transit-promoting interventions (Bamberg et al., 2003b; Bamberg, 2006; Eriksson et al., 2008; Verplanken et al., 2008) . Nevertheless, in many circumstances past habits can still last even if important contextual changes render the habitual choice less desirable , and new habits does not necessarily replace old ones (Wood and Rünger, 2016) . Many studies examining travel behavior after residential relocation (often combined with interventions aimed at changing travel behavior) (eg. Fatmi and Habib, 2017; Zarabi et al., 2019) , the introduction of a new travel mode option (eg. González et al., 2017) , transport reorganization (eg. Lattarulo et al., 2019) , or important life-course events (eg. Busch-Geertsema and Lanzendorf, 2017) have found that pre-change habits still influence post-change behaviors, and Gao et al. (2020) also found that this influence is more evident in a mild context change than in an overturning change.

The outbreak of the COVID-19 pandemic created a low-travel period in early 2020, separating pre-pandemic from post-pandemic periods, and the post-pandemic world is still under the influence of COVID cases and temporary lockdowns. This sudden interruption and consistent uncertainty could be viewed as a context change, pushing people to reexamine their travel goals and to invest more thought in their travel choices. Nevertheless, this change of environment in milder than that studied in previous research, thus pre-COVID travel habits may influence people's post-COVID intention and behavior. Also, those who were used to traveling by transit before the pandemic are familiar with the public transit system and its facilities, so it is easier for them to use transit and deal with the additional troubles caused by strict preventive measures. Based on the above, the following hypothesis is used to test the effects of TH in the TPB model:

H4. In the post-pandemic era, the stronger the habit is to use public transit, (a) the more likely the behavior of traveling by public transit in the post-pandemic era is to occur; (b) the stronger the intention is to travel by public transit; and (c) the stronger the perceived behavioral control is regarding using public transit. factors, we also introduced gender, age, educational level, employment status, monthly family income after tax, and access to a private car into the model. Beijing's spatial structure and transportation system are shown in Figure 3 . Like most cities around the world, Beijing's public transit system has been hit hard by the COVID-19 outbreak. Figure 4 shows the negative effect of the pandemic on both bus and rail transit ridership. On a national scale, the first confirmed case was reported in December 2019. In February, more cases were revealed as the result of large-scale tests. Since March, the number of new cases has decreased. Although there are still ups and downs, the incidence rate of COVID-19 has leveled off. As for Beijing, in line with national trends, the first wave of the pandemic occurred in January and February 2020. In the meantime, ridership of buses and trail transit decreased by 77.9% and 88.7%, respectively. Then, ridership increased despite a smaller wave in March due to increased imported cases from abroad. As of May, ridership rose to around half of that of the previous year, before being interrupted by a third wave in June due to a cluster of infections that emerged in Xin Fa Di Market. After that, Beijing entered a five-month low-risk period until a small number of confirmed cases was reported in late December 2020 and January 2021. Transit ridership continued to increase in August and September as social and economic orders gradually restored normalcy. Ridership then stabilized at around 20% below pre-pandemic levels until another drop in January 2021. 

The data were collected using a self-administered questionnaire. The questionnaire had two parts: (a) filter questions and basic sociodemographic questions, and (b) measures for the study constructs.

Participants were asked two filter questions: "Since the outbreak of the COVID-19 pandemic (since January 2020), have you been living in Beijing (apart from temporary departures for business trips, family visits, vacations etc.)" and "Before the outbreak of the COVID-19 pandemic, on average, how many times in a week did you travel by public transit (a single trip from the starting point to destination counts as one time, e.g., taking the bus and then transferring to subway to work is one time, returning by the same route after work is another one time)."

For sociodemographic attributes, we collected the gender, age, educational level, employment status, monthly family income after tax, and access to private car information.

The measurement items for study constructs were based on the existing literature and were customized to suit the local context and language habits. The meanings of important terms are repeatedly indicated in the questionnaire. In particular, "public transit" = buses (including trolleybuses) and rail transit (light rail and subway); "before the pandemic" = before January 2020; "the new post-pandemic normal" = from the time the educational and economic sectors resumed in-person activity (September 2020) to the time when COVID-19 vanishes or vaccines are widely available; "when the pandemic is over" = after COVID-19 vanishes or vaccines are widely available.

From November 27 to 30, 2020, an intensive online surveying campaign took place. A period including a weekend was chosen to increase the participation rate. This period was also ideal for reflecting people's attitudinal and behavioral states in the new post-COVID normal because it came after a four-month low-infection period and before the rise of confirmed cases in December (Figure 4) . Public transit ridership has largely recovered, and the pace of life has returned to normal, although COVID-19-preventive measures are still in place. The questionnaire was uploaded to and released via a professional online survey platform. It was then distributed through WeChat (one of the most popular social platforms in China), and a convenience sampling method was used.

During the data collection process, all questions required an answer, and questionnaires with one or more blank answers were not returned. A total of 1,360 completed questionnaires were returned. To control for quality, we dropped 152 questionnaires that were completed in less than 6 minutes (according to a small-scale test run, it takes a minimum of approximately 6 minutes to read and answer the questionnaire carefully). We then manually inspect the responses for illogical answers and dropped another 106 questionnaires. After that, we excluded 341 questionnaires with answers of "no" to the first filter question or "0" to the second filter question. Therefore, 761 effective responses were obtained, yielding an effective response rate of 56.0%.

The places of residence of most participants are within the boundary of the inner suburban area (Figure 3 ). This area consists of six districts, is densely populated with 52.2% of the city's total population (Beijing Statistical Yearbook, 2020), and has the best public transit coverage in Beijing.

The socio-demographic characteristics of our sample are in Table 1. The valid sample consists of 761 individuals, with the percentage of female participants (60.7%) being somewhat greater than that of male participants (39.3%). Some 73.6% of them are middleaged (30-49 years old), 75.9% have a bachelor's degree or higher, and 80.4% were employed at the time of inquiry. Family income is more spread out, as people from all income levels were surveyed. In addition, 58.7% stated that they or their families owned at least one private car, while the rest did not have access to private cars. The characteristics of our sample differs to some extent from the population statistics of Beijing; the impact of this on our understanding of the modeling results are further discussed in Section 5. 

A total number of 18 items capture intention, behavior, ATT, SN, PBC, PKC, PRC, and TH.

Another item was used to distinguish between "choice" and "captive" transit users. The original TPB constructs were measured following Ajzen's (2006) guidelines for questionnaire design and previous studies (Heath and Gifford, 2002; Han et al., 2020) . The measurements for TH were selected from the Self-Report Habit Index (Verplanken and Orbell, 2003; Nordfjaern et al., 2014b) , and the source of survey items for PKC and PRC was Han et al. (2020) . All items were measured using a five-point Likert-type scale.

To measure the public transit use behavior, we asked, "In the context of the new postpandemic normal, on average, how many times in a week do you travel by public transit?"

We converted the answer to a scale of 1-5 and used it as an indicator of behavior (B1).

The intention to use public transit in the post-lockdown period was captured by two items. I1:

"I plan to travel by public transit in the next 7 days" and I2: "I will try to travel by public transit in the next 7 days." We measured both on a scale of totally disagree (1)-totally agree (5).

Attitude was derived from three items: "In the context of the new post-pandemic normal, I think traveling by public transit is" ATT1: "bad (1)-good (5)," ATT2: "foolish (1)-wise Perceived behavioral control was measured by a single item, following a straightforward approach by Heath and Gifford (2002) . PBC1: "In the context of the new post-pandemic normal, for me traveling by public transit is easier than by other means of transportation:

totally disagree (1)-totally agree (5)." Note that according to Ajzen (2006) , the measure for PBC should cover meanings of capacity and autonomy, and many studies have used a twoitem measure including one question about how easy it is for the individual to travel by public transit, and one question about the individual's freedom of choice when it comes to using transit (eg. Bamberg et al., 2003b; Nordfjaern et al., 2014a; Zailani et al., 2016) .

However, the latter question is somewhat confusing and ambiguous in the study context, and that such statements could lead to invalidity and unreliability in participants' responses. Thus, the present study integrates the meaning of the second question into the first ("for me traveling by public transit is easy…") so that PBC1 is able to tap into both capacity and autonomy aspects; this is achieved by adding a comparative component ("… easier than by other means of transportation") which can cue respondents to think about alternative modes and the freedom of choice between these modes and public transit and can also make their thoughts more accessible.

Perceived knowledge of COVID-19 was rated by three items: PKC1: "compared with the average person," PKC2: "compared with people who are important to me (such as family and friends)," PKC3: "compared with other people who travel by public transit," "I know more about the facts of the COVID-19 pandemic (ways of transmission, protective methods etc.)."

All items were measured on a scale of totally disagree (1)-totally agree (5).

Psychological risk of COVID-19 was assessed using two items: "In the context of the new post-pandemic normal, the thought of traveling by public transit" PRC1: "makes me nervous," PRC2: "makes me mentally uncomfortable," PRC3: "makes me feel stressed." All items were measured on a scale of totally disagree (1)-totally agree (5).

To assess transit-using habit, we select two items from Verplanken's Self-Report Habit Index (SRHI) (Verplanken and Orbell, 2003; Nordfjaern et al., 2014b) . TH1: "Before the pandemic, I never (1)-always (5) traveled by public transit." TH2: "Before the pandemic, I often chose to travel by public transport without much thought: totally disagree (1)-totally agree (5)." Unlike the "habit strength" measure proposed by Wood et al. (2005) that belongs to the association view of habit and the Response Frequency Measure (RFM) proposed by Verplanken et al. (1994) that belongs to the script-based view, the SRHI cannot be classified into either association or script-based view. The SRHI is built on the idea that habit is not only past behavior, but also a psychological construct with three facets, namely repetition, automaticity, and identity (Verplanken and Orbell, 2003) . The present study uses TH1 to access the repetition facet and TH2 to access the automaticity facet; we did not include the identity facet because the identity facet is not always relevant (Verplanken and Orbell, 2003) , and in the context of the present study, habit concerning public transit travel choice is hardly connected with personal identity, thus a combination of TH1 and TH2 would be theoretically sufficient to capture people's public transit use habit strength.

Another item was used to assess the motivation (MO) for transit users. "Choice" users are those who take public transit willingly, and "captive" users are those who must use transit for practical reasons. Since these reasons are usually complicated, for example involving the inability to afford private vehicles, constraints caused by family members' travel plans and health conditions, etc., we asked a direct question to capture the motives. MO1: "In the context of the new post-pandemic normal, I am not willing to travel by public transit, but for various reasons (such as no private car and long travel distance), I have to travel by public transit: totally disagree (1)-totally agree (5)."

The background factors were derived from sociodemographic characteristics. Gender was recoded 1 for male and 0 for female. Age was recoded 15 for under 18 years, 25 for 18-29 years, 35 for 30-39 years, 45 for 40-49 years, 55 for 50-59 years, and 65 for 60 years and above. Education is the years of education corresponding to educational level: 6 years for primary school, 9 or 12 years for secondary school (9 years for middle school, 12 years for high school and secondary specialized school), 15 years for junior college, 16 years for university graduate, and 19 years for postgraduate. Employment was recoded 1 for employed (including full-time employed and studying on the job) and 0 for student, retired, and other (including full-time housewife/househusband, unemployed, and other). Income was recoded from monthly family income after tax, with 2.5 for under 5k RMB, 7.5 for 5-10k RMB, 12.5 for 10-15k RMB, 17.5 for 15-20k RMB, 25 for 20-30k RMB, 40 for 30-50k RMB, and 60 for above 50k RMB. Car was recoded from access to private car, with 1 for yes and 0 for no.

Based on the above study design, we first conducted descriptive statistics analysis on the collected data. Then, we followed the commonly used two-step approach (Anderson and Gerbing, 1988) to analyze the data:

Step 1 was to test the reliability and validity of the data using confirmative factor analysis (CFA), and

Step 2 was to test the research hypotheses and to conduct group comparisons using structural equation modeling (SEM). Covariance-based SEM was used. The software package Mplus 7.4 and its maximum likelihood estimation with robust standard errors (MLR) estimator were used to determine the models. MLR is a frequently used estimator and is robust to non-normal distribution of data.

The results of CFA should satisfy the following suggested requirements before SEM can be conducted. To satisfy scale reliability requirements, Cronbach's α should be greater than or equal to 0.7 (Nunally, 1978) . All standardized factor loadings (SFL) should be 0.5 or greater and statistically significant, the composite reliability (CR) of all model constructs should be greater than 0.7, and the average variance extracted (AVE) should exceed 0.5 to secure satisfactory convergent validity as suggested by Hair et al. (2006) and Kuo and Tang (2013) .

The squared correlation among the constructs should be less than the corresponding AVEs to ensure acceptable discriminant validity (Fornell and Larcker, 1981) .

The results of CFA and SEM should satisfy the following suggested rules to indicate good model fit. We used chi-square degrees of freedom ratio (χ 2 /df), root mean square error of approximation (RMSEA), comparative fit index (CFI) and Tacker-Lewis Index (TLI) to assess model fit. The recommended cut-off points for χ 2 /df and RMSEA are 0.5 and 0.06 respectively Bentler, 1998, 1999) . McDonald and Ho (2002) suggested that 0.08 should be taken as an acceptable threshold value for RMSEA, and a value less than 0.05 indicated a good fit. Moreover, models with CFI and TLI values greater than 0.9 are generally considered acceptable (Wang, 2014) .

Descriptive statistics for the research constructs and background factors are presented in Table 2 . On average, the participants travel by public transit 5.33 times per week (SD = 4.35).

In Table 2 , this is presented by M = 1.53 and SD = 0.44 because we rescaled the behavior construct to 1-5. The mean values of intention, ATT, SN, PBC, habit, and PKC are all greater than 3, the midpoint of their theoretical range, indicating that participants have positive perceptions of using public transit in the post-pandemic period. The average score for PRC is below 3, suggesting that participants have a low level of perceived risk of COVID-19; however, this does not mean that people's post-pandemic PRC is lower than the prepandemic level-according to our observation, residents in Beijing are generally more nervous using public transit after than before the pandemic.

Interestingly, the correlation between PKC and PRC is negative and the correlations between PKC and all other psychological constructs except MO are positive, meaning that people with more perceived knowledge are less concerned with the risk of infection and more willing to take public transit. This is quite unusual, because often, the more aware of the facts people are about a risky behavior, the more cautious they feel about performing it. One explanation is that since the infection rate is low in post-pandemic Beijing, those who have higher levels of perceived knowledge are less blindly afraid and more confident to reclaim daily mobility and its social and economic benefits by using public transit, especially when the local economy is starting to pick up and life is returning to normal. Further explanation follows in Section 5.

Looking at the differences between "choice" and "captive" transit users, MO is negatively correlated with behavior, intention, ATT, SN, and PKC, indicating that less willing users tend to have less perceived knowledge and more negative perceptions of using public transit in the post-pandemic period. MO is also negatively and weakly correlated with PBC, and it has little correlation with TH. It is worth noting that MO and PRC have a relatively large and significant correlation, meaning that less willing users have much higher psychological risk scores than more willing users. 

Confirmative factor analysis was conducted (Table 3) . The model-fit indices for the measurement model (χ 2 /df = 2.073, RMSEA = 0.038, CFI = 0.980, TLI = 0.971) meet all the requirements, indicating a good fit. Reliability and validity indices are shown in Table 3 .

Cronbach's α estimates, all above the threshold of 0.7, ranged from 0.747 to 0.898, which confirms reliability. Convergent validity was also achieved: SFLs all exceed 0.5, and the estimates for CR and AVE exceed their respective cut-off points of 0.7 and 0.5. The square roots of AVE of each model construct are greater than its correlation with other constructs ( 

Before building our proposed model, we conducted structural equation modeling to test the efficacy of the original TPB framework. As expected, TPB is capable of modeling participants' transit use intention and behavior in the post-lockdown phase (χ 2 /df = 2.898, RMSEA = 0.050, CFI = 0.980, TLI = 0.969). The explained variance in intention (62.6%) is greater than that in behavior (23.6%). This means the intention-behavior gap commonly found in previous studies (Ajzen, 2015; Nordfjaern et al., 2014a; de Leeuw et al., 2015) is also present in the current study. One reason for this is that the thoughts that occur to people when making decisions in a real situation may differ from those when responding to a survey (Ajzen, 2015) . After adding TH, the model still has good fit (χ 2 /df = 3.044, RMSEA = 0.052, Background factors also have some effects on the model and help to increase explanatory power. Older people and people who have access to a private car travel less by public transit in the post-pandemic era. The employed use public transit more than others; it is possible that employed people are generally younger and have more travel needs, so they need to resort to transit from time to time. Fig. 5 is inserted here. Source: the authors. Notes: * p < 0.05, ** p < 0.01, *** p < 0.001; to avoid overloading the diagram, correlations between the research constructs are not shown: they are 0.533*** between ATT and SN, 0.419*** between ATT and PBC, 0.273*** between SN and PBC, 0.308*** between TH and ATT, 0.546*** between TH and SN, 0.564*** between PKC and SN, 0.338*** between PKC and TH, -0.226*** between PKC and PRC, -0.398*** between PRC and SN, -0.171*** between PRC and TH, -0.144** between PRC and PBC. Effects from background factors to intention or behavior that are not significant are not shown.

The direct, indirect, and total effects of the key variables on public transit use behavior and intention are summarized in Table 5 . Original TPB constructs all have positive and significant total effects on intention and behavior. HT is obviously an important predictor of intention and behavior, and the total effects of PKC and PRC on intention and behavior are also significant though in opposite directions. The above analysis revealed the decision-making mechanism for the whole sample. Now we employ group comparison analysis to study the difference in this decision-making process between "choice" and "captive" transit users. We split the whole sample into two groups using the measurement item for MO: those with MO1 ≤ 3 were identified as "choice" or willing transit users (N = 449), and those with MO1 > 3 were identified as "captive" or notso-willing transit users (N = 312). We then conducted a series of invariance tests. The results are shown in Table 6 .

The baseline model (1) was estimated without constraints, allowing all model parameters to be different across groups. It has good fit, indicating that configural invariance holds. The equal loadings model (2) was estimated with equality constraints for the factor loadings of the two groups. The chi-square difference test result is not significant (p > 0.05), meaning that adding equal loading constraints does not significantly reduce model fit; therefore, measurement invariance holds.

After confirming measurement invariance, we can test structural invariance by adding equality constraints to all or some of the path coefficients and conducting chi-square difference tests. Model (3) was estimated with all path coefficients constrained to be equal.

The test result is significant (p < 0.05), meaning that the path estimates for "choice" and "captive" users are not exactly the same. Furthermore, we tested the cross-group invariance of single path coefficients. Model (4) was estimated with only one path coefficient (PRC→ ATT) constrained to be equal. The test result is significant, and the estimated coefficient for "choice" users is larger in absolute value than that for "captive" users, which suggests that the effect of psychological risk on attitude is greater for "choice" users than for "captive" users.

Besides, from PRC→ATT, we also tested all other paths in the research model and found nonsignificant results. Table 6 is inserted here.

As introduced, this paper investigates the determinants of personal travel mode choice of public transit in the context of the post-COVID era. In particular, the impacts of three key factors, transit travel habit, perceived knowledge of COVID-19 and psychological risk are examined by an updated TPB framework. The following important points are discussed according to the analysis results above.

First, although the effects of the original TPB constructs do not show any abnormities, some context-related issues should still be discussed.

The original TPB constructs attitude, subjective norm, and perceived behavioral control all have significant effects on public transit travel intention and behavior in the post-pandemic era, which proves (again) the effectiveness of the TPB; nevertheless, there may be potential differences between people's decision-making mechanisms in the post-pandemic context and in normal daily context. In the unstable post-pandemic environment, the role of reason-based factors may be different. For example, due to uncertainty, people might rely more on others' opinions or actions when making decisions; this phenomenon can be more prominent in China and some other countries where society promotes a collective culture that values social impressions and social outcomes. Also, the meanings of reason-based factors might have changed in the post-pandemic era. For instance, when making travel mode choices in the post-COVID context, safety has become a dominant component of personal attitude, and most residents in Beijing prioritize safety over other aspects they usually consider (accessibility, cost, speed, and the level of physical comfort). Perceived behavioral concern may also be laced with pandemic-related components such as the capacity to deal with troubles due to COVID prevention measures.

Second, a higher level of perceived knowledge of COVID-19 is a positive influence on people's public transit use intention in the post-pandemic context, while the higher psychological risk of COVID-19 is a negative influence.

Higher perceived knowledge of COVID-19 has significant and positive effects on transit use intention, mediated by attitude and perceived behavioral control. Many studies available to us (Meadowbrooke et al., 2014; Han and Hyun, 2017; Han et al., 2020; Zhu et al., 2020) convey the idea that higher levels of perceived knowledge mean higher levels of risk aversion. For example, Han et al. (2020) found that more perceived knowledge of COVID-19 positively influenced international travelers' intention to travel to a safer destination through its positive effects on attitude and subjective norm. However, the results of the present study suggest that higher levels of perceived knowledge mean stronger will to travel by public transit, i.e., higher levels of risk appetite. This contradiction can be explained when viewed from a rational decision perspective (instead of a risk avoidance one) combined with the research context. A risk behavior contains a choice between two options: taking or not taking the risk.

People with more perceived knowledge have a clearer picture of which side poses less expected loss and thus are more likely to make the rational decision. In most previous studies, participants with more perceived knowledge chose not to take the risk because this choice had less expected loss. For example, the survey by Han et al. (2020) was conducted in mid-April, when the pandemic was still spreading rapidly around the world; under this circumstance, the possible reduction in travel experience by changing to a safer destination appeared to cause much less harm than the possibility of catching a novel virus in a foreign country. However, our survey was conducted in a relatively low-risk post-pandemic environment in Beijing. Participants with more perceived knowledge are more aware of the effectiveness of prevention measures and how unlikely it is to get infected on a public vehicle. Therefore, they choose a small possibility of catching COVID-19 using public transit over reduced mobility and subsequent damage to their economic and social welfare caused by avoiding public transit. This is also a rational decision. In this sense, our results are in fact very consistent with previous research. The present study proves that the role of higher perceived knowledge level is to help people to react more rationally and constructively to risk rather than to promote risk aversion.

Higher psychological risk of COVID-19 has a significant and negative effect on transit use intention, mediated by attitude. There is little literature about psychological risk's influence on TPB constructs. Many past attempts have mainly focused on the general idea of perceived risk (Featherman and Pvalou, 2003; Pearson and Hamilton, 2014; Zhu et al., 2020) . Our results are consistent with the only available and relevant study by Seol et al. (2015) , in that more psychological risk has an indirect and negative effect on intention.

Both perceived knowledge and psychological risk of COVID-19 reflect individuals' perceptions of the risk of using public transit in the post-pandemic period, yet they affect transit use intention in opposite directions. Perceived knowledge represents a rational part of the individual, and psychological risk stands for an irrational part. A higher level of perceived knowledge means enough awareness or vigilance to respond to risk in an objective and constructive manner; in contrast, a higher level of psychological risk means excessive anxiety that can only damage people's peace of mind and prevent them from confronting the problem. It is recommended that future research continue to explore perceived knowledge and psychological risk's role in influencing people's decision-making processes.

Third, the habit of travelling by public transit has a long-term impact on people's public transit mode choice-making mechanism even when an outbreak event, such as COVID-19, intervenes in their intention and thus behavior.

Pre-pandemic public transit use habit has significant effects on people's post-pandemic transit use intention and behavior. The present study involves three different contexts: pre-COVID-19, during COVID-19, and post-COVID-19; since daily mobility was largely suppressed during the pandemic, participants' pre-pandemic travel habits were measured and used to predict their post-pandemic travel intentions. Our results are in line with many previous studies that have also found significant effects of old habits on travel behavior after context change (eg. Busch-Geertsema and Lanzendorf, 2017; Fatmi and Habib, 2017; González et al., 2017; Lattarulo et al., 2019; Zarabi et al., 2019) . Nevertheless, this does not mean it is unnecessary to study people's post-pandemic travel habits. The habit discontinuity hypothesis states that a disruption in decision context may lead to discontinuity in habit and eventually habit change (Verplanken et al., 2008; Walker et al., 2015; Verplanken and Roy, 2016) . In this process, old habits do not disappear suddenly but gradually fade away over time, and new habits also take time to build (Walker et al., 2015) . Moreover, new habits do not always replace old habits (Wood and Rünger, 2016) . Therefore, from the finding of prepandemic habit's influence on post-pandemic intention and behavior alone, it is hard to distinguish if old habits have survived the context change and will continue to be strong, or old habits have been gradually replaced by new ones but are still strong enough to make an impact, or old habits are coexisting with new ones. After the pandemic, some transit riders might turn to private cars to avoid crowding, and some others might favor cycling. It is important for future studies to provide more information on travel habit change before and after the pandemic, and to study ways to promote desirable new habits while suppressing unwanted ones.

Fourth, "captive" transit users' level of psychological risk is higher than that of "choice" transit users, yet their decision-making mechanism is less reactive to psychological risk than that of "choice" transit users.

The descriptive statistics show that those who are less willing to take public transit by choice in the post-pandemic era also tend to experience higher psychological risk of COVID-19 (the correlation between motivation and psychological risk of COVID-19 is 0.535, p < 0.001).

Naturally, this is consistent with our distinguishing criteria between "choice" and "captive" users. The results from the group comparison analysis indicate that motivation has moderating effects on the influence of psychological risk of COVID-19 on attitude, and that the influence of psychological risk is smaller for "captive" transit users than for "choice" users. Additional Wald chi-square tests also showed that psychological risk of COVID-19's total effects on intention and behavior for "choice" users is significantly larger than for "captive" users (total effect PRC→Intention: for "choice" users = -0.223, p < 0.05; for "captive" users = -0.089, p < 0.05, Wald test p < 0.05. Total effect PRC→Behavior: for "choice" users = -0.073, p < 0.05; for "captive" users = -0.033, p < 0.05, Wald test p < 0.05).

This means that despite the way concerns about infection risks could negatively influence their state of mind, "captive" transit users are to some extent still bound to make the same choices, making them underprivileged in the post-pandemic era.

The contributions of the present study are threefold: (1) It provides the needed behavioral knowledge of individual public transit travel choice-making after the COVID-19 pandemic.

The negative influence of COVID-19 on public transit systems is unprecedented, and the impression that public transit is no longer only a way of transportation, but also a route of transmission for potential health risks induced by people's pandemic experiences is likely to affect their transit travel choices throughout and beyond the post-pandemic era. In light of this, the present study provides an insight into people's decision-making mechanisms after the pandemic using the TPB framework, and it studies factors that can influence this mechanism, i.e., perceived knowledge of COVID-19 and psychological risk of COVID-19.

To the best of our knowledge, the present study is one of the first TPB-guided studies on transit use behavior in the post-pandemic world.

(2) It tests habit's effect in a post-pandemic environment, and found that pre-pandemic habits can influence post-pandemic intention and behavior. Many past studies have examined the influence of old habits on behavior after decision context change. However, the types of context change studied mostly belong to categories of home relocation (often combined with travel behavior intervention) (eg. Bamberg et al., 2003b; Bamberg, 2006; Verplanken et al., 2008; Fatmi and Habib, 2017; Zarabi et al., 2019) , job relocation (eg. Walker et al, 2015; Gao et al., 2020) , travel behavior intervention (eg. Bamberg et al., 2003a; Eriksson et al., 2008) , new travel mode option (eg. González et al., 2017; Gao et al., 2020) , transport reorganization (eg. Lattarulo et al., 2019) , or important life-course events (eg. Lanzendorf, 2010; Busch-Geertsema and Lanzendorf, 2017) , while context change due to a global pandemic is rarely discussed. Transport reorganization such as banning car use or roadblocks have similarities with the pandemic situation, as they both cause temporary shutdowns in certain travel modes; yet apart from lockdowns and negative effects on public transit ridership, the COVID-19 pandemic has also brought public health concerns and increased risk perception, therefore it is different from traffic control. The present study provides information of habit's effect in a distinctive kind of context change seldom studied by previous research. (3) It has potential implications that might be valuable for public transit promoting "nudge" interventions. For example, increasing perceived knowledge of COVID-19 and lowering psychological risks might be a promising direction; some measures may include informing the public about how small the possibility of infection on a post-pandemic public vehicle is, and adhering to effective protective measures such as mask-wearing in public spaces and regular disinfection of public vehicles. Another possible direction is to discourage unfavorable old habits and foster positive new ones during this period of contextual change. Also, the inequality between choice and captive transit users should not be neglected; ways to ease captive users' psychological stress might include offering special services such as subway by appointment and customized bus routes. However, we should be fully aware of the gap between psychological/behavioral studies and the efficacy of behavioral interventions derived from them. Respondents do not always genuinely show their thoughts and intentions thus selfreport surveys could be misleading, and the influence of nudges could vary across different categories of behaviors, not to mention in some situations more heavy-handed measures are needed to shake stubborn habits (Kristal and Whillans, 2020) . This calls for more comprehensive and solid behavioral experiments to pinpoint effective policies.

There are limitations to the present study. First, there are some methodological limitations.

We used a single item to measure perceived behavioral control and only two items to access travel habit. Although theoretically the items we used should be adequate to measure these constructs and also be suitable for the study context, it is still recommended that future studies use more items to improve robustness of the measures. Moreover, we also applied an alternative measure for habit based on the Response Frequency Measure (RFM) by Verplanken et al. (1994) to test the research model. The participants were asked to select their most frequently used travel mode before the pandemic from a list of all possible modes (bus, rail transit, private car, taxi, motorcycle, electric bicycle, bicycle owned by self, shared-bike, or other) for five different travel purposes: RFM1: going to work/school, RFM2: going shopping at supermarkets or markets, RFM3: going to shopping malls or other places of entertainment, RFM4: going to exercise or places of leisure/relaxation, and RFM5: going to the hospital or clinic); public transit travel habit strength was calculated as the number of times public transit (bus or rail transit) is chosen, and was rescaled to 1-5. Using this measure for habit, we re-estimated the research model and found that important findings discussed above remained consistent. Some differences exist, such as the RFM habit strength has smaller effects on perceived behavioral control and intention and larger and more significant effect on behavior; this might be because the two measures of habit were carried out in different psychological context of the respondents -the RFM questions (RFM1 -RFM5) were asked with objective questions like sociodemographic attributes, while the main habit questions (TH1 -TH2) were asked with other psychological items. Another difference is that along with motivation's moderating effect on psychological risk of COVID-19's influence on attitude, there is also a moderating effect of motivation on perceived knowledge of COVID-19's influence on perceived behavioral control, and captive transit users' perceived behavioral control is less influenced by their perceived knowledge of COVID-19; this further confirms that travel mode choices of captive users are less flexible and that they are rendered underprivileged in this sense. Second, we used the convenience sampling method and the sample collected may contain potential bias. Consequently, the generalization of the findings will be limited by our specific sample, and the modeling results should be interpreted with caution. Table 1 presents a comparison of sample characteristics and population statistics of Beijing. Generally, our sample reflects a highly educated young-to-middle-aged employees with average monthly family income and private car ownership that are slightly higher than the whole population. The sample also consists of a larger proportion of women than the whole population. However, since group comparison analysis for the research model (without gender) between male and female groups indicates measurement invariance and structural invariance, meaning that the model estimation is not significantly different for men and women, the imbalanced female/male ratio should have little impact on our interpretation of modeling results (note that in Table 4 and Figure 5 , gender makes a contribution to explaining behavior, but its effect is very small; thus, this is still in line with the group comparison analysis results). The 30s and 40s age groups, which are the backbone of labor market, consist of 73.6% of the sample but only 35.9% of the whole population; this could explain the dominance of employed participants (they consist of 80.4% in sample but 57.5% in the whole population). These employees could be regular commuters with stronger and more persistent public transit use habits than others, thus based on this sample the model could have over-estimated the effects of habit. However, existing research suggests that older people and people with less cognitive-control are less competent in seizing the opportunity of context disruption to change habits (de Wit et al., 2014; Otto et al., 2015; Wood and Rünger, 2016) , while young-to-middle-aged people can be more adaptive and fast-learning and may change their old habits faster after context change (Qin et al., 2019) ; therefore, it is also possible that the lack of juveniles and old people in the sample would cause underestimates of habits' effects. Our sample also has a high percentage of university graduates and postgraduates (75.9% in sample but only 32.4% in the whole population). Highly educated people tend to be more rational and assertive than average population, thus they might have smaller psychological risk and more perceived knowledge, and they might be less affected by these COVID-related factors. Therefore, we could have under-estimated the effects of psychological risk of COVID-19 and perceived knowledge of COVID-19's influence. High educational levels and large percentage of employed workers can explain the sample's higher average monthly family income (19.4kRMB compared with 16.2kRMB in the whole population) and private car ownership (58.7% compared with 53.0% in the whole population). This sample mainly reflected main-stream urban transit riders but did not include enough old/young people and those with lower educational levels and lower income. The travel choices of these groups are also important for understanding post-pandemic transit use.

Future studies could use a random sample to further generalize the results. Finally, public transit travel is considered as a whole, yet people's travel mode choice-making mechanisms could be different for trips with different purposes. Therefore, it is necessary for future research to examine the pandemic's influence on people's willingness to take public transit for commuting, shopping, recreational, and other purposes.

The present study builds a theoretical framework based on TPB to examine the mechanism behind the individual's choice to travel or not to travel by public transit in the post-pandemic era. We shed light on individual people's post-COVID public transit travel choice-making processes and examined perceived knowledge of COVID-19, psychological risk of COVID-19, and pre-COVID travel habit's influence on it. We analyzed the effects of the model constructs on transit use intention and behavior, discussed the role of vital constructs, and compared the decision-making process of "captive" and "choice" transit users. The key findings are as follows. Although modeling results confirms the effectiveness of TPB, the effects and meanings of original TPB constructs may be potentially different in the postpandemic context than in normal daily context; psychological risk has a negative effect on people's decision-making process, while perceived knowledge can offset this negative effect; people's pre-pandemic travel habit as an enduring effect on their post-pandemic transit use; "captive" transit users must use transit despite their higher psychological risk than "choice" users. This study calls for solid behavioral experiments to further examine potential implications on behavioral interventions.

Constructing a theory of planned behavior questionnaire

The theory of planned behavior

The theory of planned behaviour is alive and well, and not ready to retire: A commentary on Sniehotta, Presseau, and Araújo-Soares

Structural equation modeling in practice: A review and recommended two-step approach

Impact of SARS-CoV-2 on the mobility behaviour in Germany

The role of personality in the thoughts, feelings, and behaviors of students in Germany during the first weeks of the COVID-19 pandemic

Choice of travel mode in the theory of planned behavior: The roles of past behavior, habit, and reasoned action

Does habitual car use not lead to more resistance to change of travel mode?

Impact of COVID-19 pandemic on mobility in ten countries and associated perceived risk for all transport modes

Insights into the impact of COVID-19 on household travel and activities in Australia -the early days of easing restrictions

Beijing daily COVID-19 pandemic situation report

Beijing Statistical Yearbook editorial board and editorial staff

Extending the theory of planned behaviour to predict the intention to take the new high-speed rail for intercity travel in Libya: Assessment of the influence of novelty seeking, trust and external

The theory of planned behavior: Selected recent advances and applications

Optimization of passengers boarding in the BRT system based on the security protocols established by the Covid-19 pandemic

Habitual or reasoned? Using the theory of planned behavior, technology acceptance model, and habit to examine switching intentions toward public transit

Travel mode choice behavior model of public transit incorporating environmental concern and habit

China Statistical Yearbook 2020

Exploring the behaviour change of crowding impedance on public transit due to COVID-19 pandemic: Before and after comparison

Using the theory of planned behavior to identify key beliefs underlying pro-environmental behavior in high-school students: Implications for educational interventions

Understanding public transport satisfaction in post COVID-19 pandemic

Mobility patterns before, during, and anticipated after the COVID-19 pandemic: An opportunity to nurture bicycling

Is the intention to travel in a pro-environmental manner and the intention to use the car determined by different factors?

Predicting e-services adoption: A perceived risk facets perspective

Evaluating structural equation models with unobservable variables and measurement error

Development of script-based travel mode choice after forced change

Perceived risk and information sources in services purchasing

Public transport planning adaption under the COVID-19 pandemic crisis: Literature review of research needs and directions

COVID-19 Community Mobility Reports

Multivariate data analysis

The post-coronavirus world in the international tourism industry: Application of the theory of planned behavior to safer destination choices in the case of US outbound tourism

Fostering customers' pro-environmental behavior at a museum

Extending the theory of planned behavior: Predicting the use of public transportation

Fit indices in covariance structure modeling: Sensitivity to underparameterized model misspecification

Cutoff criteria for fit indexes in covariance structure analysis: Conventional criteria versus new alternatives

Transportation activity patterns of Chinese population during the COVID-19 epidemic

Relationships among service quality, corporate image, customer satisfaction, and behavioral intention for the elderly in high speed rail services

The unintended consequences of COVID-19 mitigation measures on mass transit and car use

The determinants of commuting mode choice among school children in Beijing

Simulating the spread of epidemics in China on multi-layer transportation networks: Beyond COVID-19 in Wuhan

The influence of social-psychological factors on the intention to choose low-carbon travel modes in Tianjin

The impacts of COVID-19 pandemic on public transit demand in the United States

Commuting travel mode choice among office workers: Comparing an extended theory of planned behavior model between regions and organizational sectors

The role of psychology and habit in the choice making process of car users: The case of Lebanon

Principles and practice in reporting structural equation analysis

Information behavior and HIV testing intentions among young men at risk for HIV/AIDS

Monthly report of passenger traffic volume in central cities of China

National Health Commission of the People's Republic of China

The role of deliberate planning, car habit and resistance to change in public transportation mode use

Perceived thresholds for transport mode change and tolerance of push measures in an urban Norwegian public

Psychometric Theory

Road network base map of Beijing

Investigating driver willingness to drive through flooded waterways

Mobility behaviour in view of the impact of the COVID-19 pandemic-public transport users in Gdansk case study

What affects our urban travel behavior? A GPS-based evaluation of internal and external determinants of sustainable mobility in Stuttgart (Germany)

Predictors of car use habit strength in an urban Norwegian public

Incorporating latent variables into discrete choice models -A simultaneous estimation approach using SEM software

Reflections on past behavior: A self-report index of habit strength

Habit, information acquisition, and the process of making travel mode choices

Latent variable modeling using Mplus

Is the intention to use public transport for different travel purposes determined by different factors?

Effects of human behaviour changes during the COVID-19 pandemic on influenza spread in Hong Kong

Examining the effects of neighbourhood design on walking in growing megacity

Travel behaviour and life course: Examining changes in car use after residential relocation in Beijing

Investigating the influence factors of public transit rider transfer intentions: A case study in Nanjing

Public risk perception and willingness to mitigate climate change: City smog as an example

Is a residential relocation a good opportunity to change people's travel behavior? Results from a theory-driven intervention study

From university to work life -Jumping behind the wheel? Explaining mode change of students making the transition to professional life

Impaired acquisition of goal-directed action in healthy aging

Interrupting habitual car use: The importance of car habit strength and moral motivation for personal car use reduction

Modelling mode switch associated with the change of residential location

Explaining stability in travel mode choice: An empirical comparison of two concepts of habit

Revealing psychological inertia in mode shift behavior and its quantitative influences on commuting trips

Modelling motivation and habit in stable travel mode contexts

Introduction: habitual travel choice

Empirical tests of a model of determinants of script based driving choice

Testing for inertia effect when a new tram is implemented

Habit and hysteresis in mode choice

Habit discontinuity and student travel mode choice

What cognitive mechanisms predict travel mode choice? A systematic review with meta-analysis

What we can learn from five naturalistic field experiments that failed to shift commuter behaviour

Key events and their effect on mobility biographies: The case of childbirth

Resistance to change: Car use and routines

Habits: A repeat performance

Research on China's Carbon Dioxide Emissions Efficiency from

Cognitive control predicts use of model-based reinforcement learning

Analysis on context change and repetitive travel mode choices based on a dynamic, computational model

The role of attitudes, transport priorities, and car use habit for travel mode use and intentions to use public transportation in an urban Norwegian public

Understanding repetitive travel mode choices in a stable context: A panel study approach

Understanding repetitive travel mode choices in a stable context: A panel study approach

The habit-driven life: Accounting for inertia in departure time choices for commuting trips

Interpersonal behavior

Habit, attitude, and planned behaviour: Is habit an empty construct or an interesting case of goal-directed automaticity?

Attitudes versus general habit: Antecedents of travel mode choice

Empowering interventions to promote sustainable lifestyles: Testing the habit discontinuity hypothesis in a field experiment

Context change and travel mode choice: Combining the habit discontinuity and self-activation hypotheses

Old habits die hard: Travel habit formation and decay during an office relocation

Habit in everyday life: thought, emotion, and action

Habits in everyday life: Thought, emotion, and action

Psychology of habit

Changing circumstances, disrupting habits

The impacts of residential relocation on commute habits: A qualitative perspective on households' mobility behaviors and strategies

Conceptualization, Supervision, Writing-Reviewing and Editing, Revision. Yukun Gao: Data analysis, Methodology, Writing-draft preparation