Introduction

The relationship between travel and subjective well-being (SWB) has recently attracted increasing interest in transport studies (Ferdman 2021; Mokhtarian 2019; Morris 2015; Morris and Guerra 2015; Reardon and Abdallah 2013; De Vos et al. 2013). This is arguably because happiness and well-being are considered the most important goals in human history (Compton 2005; Mokhtarian 2019), and it is desirable for travel to contribute to this goal (Ettema et al. 2010). Stanley and Stanley (2007) suggested that social and transport policies should consider improved well-being as a goal instead of improved accessibility. Exploring this relationship can strengthen our understanding of multifaceted aspects of travel and can ultimately lead to improved SWB.

To date, most studies have explored this issue by relating travel to general SWB, including eudaimonic well-being, hedonic well-being, and life satisfaction. One approach involves travel utility, a measure of travel satisfaction (Ettema et al. 2010). This approach is based on the utility maximization principle (Ben-Akiva et al. 1999), a common and conventional assumption in travel behaviour analysis for exploring various travel aspects, such as mode choice and route choice. In Abou-Zeid and Ben-Akiva (2012), for example, well-being was considered an indicator of utility. However, there is also concern that utility in a travel choice context is not a good proxy for travel satisfaction because choices made in this situation are often constrained by various factors (Bergstad et al. 2011). In addition, some theoretical models have been introduced to systematically analyse the effect of travel on SWB. De Vos et al. (2013) suggested five ways that travel affects eudaimonic and hedonic well-being, including experiences during travelling (e.g., stress induced by car use, physical exercises by walking and cycling, travel time), activities performed while travelling (e.g., reading a book or enjoying entertainment), activities at the destination (e.g., eating out or leisure activities), travel as an activity (e.g., recreational walking), and motility (i.e., the potential to travel). Another theoretical model was introduced by Ettema et al. (2010) in which travel contributes to affective and cognitive SWB through activity participation, affective factors (e.g., stress, comfort) and instrumental factors (e.g., travel time, cost). Reardon and Abdallah (2013) and Delbosc (2012) also linked travel to SWB but in a more general framework with a focus on the impacts of macro factors such as policy, infrastructure, and mobility. Chatterjee et al. (2020) examined the effect of commuting on SWB by suggesting that well-being can be generated during travel (commute satisfaction), after travel (satisfaction with other life domains), and in the long term (overall SWB). The literature reviewed in the study then provided evidence in support of the effects of commuting on SWB during and after travel. However, the long-term effect has not been confirmed.

Another line of research focuses on domain-specific SWB rather than SWB in general. Motivated by the fact that SWB can be defined in specific domains, the Satisfaction with Travel Scale (STS) was initially developed by Bergstad et al. (2011). Regression analyses revealed that STS influenced affective and cognitive SWB both directly and indirectly via satisfaction with activities. An extended version that balances cognitive and affective elements of STS was proposed by Ettema et al. (2011), which was confirmed to be able to measure travel satisfaction with different modes in different urban contexts (Ettema et al. 2016). Using STS, Olsson et al. (2013) found that satisfaction with work commutes contributed to overall happiness. Relatedly, De Vos and Witlox (2017) noted that STS can be applied to both specific trips or daily travel in general and suggested a bidirectional relationship between trip satisfaction, satisfaction with daily travel, and life satisfaction. In addition to STS, the concept of travel liking, or affinity for travel, originally introduced by Mokhtarian and Salomon (2001), has been mentioned as a source of travel-related SWB (Singleton and Clifton 2021). Mokhtarian and Salomon (2001) listed three sources of travel liking, including activities conducted during travel, activities conducted at the destination, and the activity of travelling itself. The last item, i.e., the activity of travelling itself, was described as ‘intrinsic aspects of travel’ that include the sensation of speed and the feeling of exposure to the environment and attractions. Ory and Mokhtarian (2005) additionally discussed a variety of sources of travel liking, such as independence, status, and conquest. Whereas travel liking refers to travel-related SWB that people gain from travelling, its counterpart in the opposite direction, i.e., travel disliking, has received quite limited attention (see De Paepe et al. (2018) for an example). Other SWB-like concepts in the travel context include travel affect and travel eudaimonia (Singleton 2019; Singleton and Clifton 2021). In Singleton and Clifton (2021), a step-by-step development of scales for these measures was introduced. Whereas the measurement for travel affect was adapted from the literature on affective SWB, the measurement for travel eudaimonia was based on the literature on both psychology and travel behaviour analysis. Notably, four elements of travel eudaimonia were identified, including health, autonomy, competence, and security. An empirical evaluation of the effects of walking and cycling on travel affect and travel eudaimonia was provided in Singleton (2019). Recently, some studies have focused on the potential effects of specific aspects of travel on trip satisfaction. For example, Mao et al. (2016) found a U-shaped impact of perceived modal flexibility on trip satisfaction. Similarly, Ye et al. (2020) explored the effect of the dissonance between the expected commute time and the experienced commute time and found that trip satisfaction was highest when this dissonance was minimized.

Overall, previous studies suggest that travel affects SWB in both a direct way (e.g., travel experiences) and an indirect way (e.g., activity engagement). Furthermore, Mokhtarian (2019) noted that previous studies used simple analysis methods and many confounders were not accounted for.

The target of this study is the role of travel as an enabler/facilitator for out-of-home activities and its effect on multiple dimensions of general SWB. In some recent reviews (Delbosc 2012; Ettema et al. 2010; Mokhtarian 2019; De Vos et al. 2013), travel was consistently found to have an indirect effect on SWB by enabling/facilitating activity engagements, and this was considered the most important contribution of travel to SWB (Mokhtarian 2019). In this regard, however, there are at least two issues that remain unexplored.

First, Ettema et al. (2010) noted that we know little about how the travel context, such as travel mode use, influences SWB. The authors argued that understanding this relationship can be a helpful source for evaluating transport policies. Accordingly, a general theoretical framework was provided for this purpose. However, this framework has yet to be operationalized. Although there is a rich body of literature that studies the relationship between travel and SWB (see Chatterjee et al. (2020) for a review), we found no empirical evidence of how multiple dimensions of general SWB are affected by a measure of travel context such as travel mode use.

Second, the term ‘indirect’ has been used with two implied causal relationships: travel enables/facilitates activities, and satisfaction with these activities is a direct source of SWB (Mokhtarian 2019). Unfortunately, no further details were provided regarding ‘how’ travel enables/facilitates activities. We argue that this may be a key point in our understanding of the relationship between travel and SWB via activities. To illustrate this, let us consider a person who can use a car only for eating-out trips. In this case, only car use enables this activity. If the person can also use a bus, then although both car use and bus use enable this activity, we may expect that car use facilitates the activity better than bus use due to its higher flexibility (e.g., easier to shift to another restaurant if busy) and other attributes (e.g., allowing for the whole family to eat together). As a result, satisfaction with the eating-out trip may differ depending on which mode is used. A similar situation can occur for other activities, such as leisure activities. Therefore, travel mode use can provide additional insights that are helpful for understanding the indirect effects of travel on SWB through activities.

Accordingly, this study aims to explore ‘how’ travel mode use can be linked to SWB through its role of enabling/facilitating out-of-home activities. In addition, we explore multiple dimensions of general SWB that are measured by the Subjective Well-Being Inventory (SUBI) scale (Sell 1994). As noted by De Vos et al. (2017), there is still limited literature on how travel affects the multidimensionality of SWB. The study by De Vos et al. (2017) partly filled this gap by exploring two specific dimensions of eudaimonic well-being and life satisfaction. However, the 11 dimensions of SWB in the SUBI scale were not systematically considered in any previous transportation studies. In addition, this scale offers an alternative conceptualization of SWB compared to conventional measures of SWB. Specifically, it considers the interrelationship between a variety of concerns that have been studied unsystematically in the literature (e.g., social adjustment, loneliness, and perceived health) and the relationship between these concerns and overall SWB, such as general well-being and positive/negative affect. We describe our conceptual model in Sect. 2. The data for operationalizing the conceptual model and the analysis method are presented in Sect. 3 and Sect. 4, respectively. The results and discussion are presented in Sect. 5 and Sect. 6, respectively, followed by the conclusions in Sect. 7.

Indirect effect of travel mode use on SWB: the conceptual model

In this section, we first present a brief overview of SWB conceptualizations. We then discuss the theoretical basis and the related literature for the link between travel, out-of-home activities, and general SWB. Our conceptual model that allows for examination of the potential effects of travel mode use on SWB follows.

Subjective well-being: an overview

The concept of well-being refers to people’s evaluations of their own lives (Diener 2000, 2006). A person with high well-being, for example, is a person whose life is going well (Raibley 2012). Because it is expected that people must believe that they are living well to have a good life, this concept is called ‘subjective well-being’ (Diener 2000). Structurally, SWB consists of three components: the cognitive component of life satisfaction and two affective components of positive affect and negative affect (Diener 2000; Diener et al. 1985; Morris 2015). These elements can be further categorized as short-term (i.e., affective components) versus long-term (i.e., cognitive component) and domain-specific versus general assessment (Mokhtarian 2019). The literature distinguishes between two SWB types: hedonic well-being and eudaimonic well-being (De Vos et al. 2013, 2017). The hedonic view often associates SWB with pleasure and happiness (Kahneman et al. 1999; Mokhtarian 2019; Ryan and Deci 2001), whereas the eudaimonic view, a concept named by Waterman (1993), focuses on ‘more than just happiness’ (Ryan and Deci 2001), such as self-realization (Waterman 1993), human potential (Ryan and Deci 2001), and purpose in life (Mokhtarian 2019). Studies have shown that these conceptualizations of SWB are distinguishable (Waterman 1993).

In the literature, happiness is a concept closely related to SWB and is, in fact, often used interchangeably with SWB (Deci and Ryan 2008; Delle Fave et al. 2011; Easterlin 2005; Medvedev and Landhuis 2018). Whereas well-being is essentially related to benefits and harms (Raibley 2012), happiness refers to a ‘general positive mood’ (Diener 2006), or one’s appraisal of his or her overall quality of life (Shin and Johnson 1978). In common language, Davitz (1970), cited by Veenhoven (2013), reported that the word happy is mainly associated with ‘pleasant mood-states’. In some studies, happiness is considered an element of well-being (Delle Fave et al. 2011; Feldman 2010; Raibley 2012). Diener et al. (2009), for example, argued that one component of a good life is happiness; that is, a happy person reacts positively to his or her life. Similarly, Diener et al. (2009) and Diener et al. (1999) posited that happiness consists of pleasant emotion, which is an element of SWB.

The relationship between travel, out-of-home activities, and SWB

In the economist view, travel is largely considered ‘derived demand’, i.e., people travel not for its own sake but to facilitate a variety of activities (Small et al. 2007)Footnote 1. This ‘derived’ aspect of travel has commonly been studied with an activity-based approach where ‘any understanding of travel behaviour is secondary to a fundamental understanding of activity behaviour’ and travel is considered simply an attribute of activity (McNally and Rindt 2007). Because travel’s main function is for out-of-home activities, the change in activity demands can result in a change in travel behaviour. For example, Metz (2000) argued that older adults travel less frequently and for shorter times than younger people partly because they have fewer travel demands, e.g., for work and business. The economist view thus emphasizes the activity end and regards travel as a step in attaining this end. For example, going to a supermarket to buy something is no different from shopping online if one solely considers the final goal of buying the thing.

The economist view is based strongly on the notion of utility maximization and the focus is on the activity utility side rather than the travel utility side, which has not received particular attention. However, the influence of travel on subsequent activity, often known as the spill-over effect, may be significant in certain circumstances. For leisure activities, De Vos et al. (2017) found that mood during travel and the evaluation of travel directly affected activity satisfaction. Mattisson et al. (2015) found that commuting by car was associated with less social participation and lower general trust than commuting by active modes. However, a similar comparison between commuting by public transport and by active modes showed no significant statistics. In a systematic review, Chatterjee et al. (2020) found that longer commuting time was associated with less time spent on social and leisure activities, and a spill-over effect exists for commuting duration on positive/negative moods at work and job satisfaction. More importantly, certain activities cannot be done without certain travel modes (e.g., an amenity that can be reached by car only). In these cases, travel plays the role of an enabler for out-of-home activities (Mokhtarian 2019; De Vos et al. 2013).

This role of travel as a facilitator, and sometimes an enabler, for out-of-home activities suggests the idea of an indirect effect of travel on SWB via activities. First, the connection between activity engagement and SWB has been well established in the literature (Abou-Zeid and Ben-Akiva 2012; Ettema et al. 2010, 2011). Theoretically, activities can contribute to both affective SWB through experienced positive/negative affect and cognitive SWB through making progress towards one’s goals and potentials (Ettema et al. 2010), particularly for valued activities (Cantor and Sanderson 1999; Myers and Diener 1995). There is also empirical evidence in support of this connection (Bergstad et al. 2011; Pychyl and Little 1998; De Vos et al. 2017; Waterman et al. 2008). These facts, together with the close relationship between travel and activities discussed above, create a theoretical basis for linking travel mode use with SWB via activities. Limited travel options, for example, may be a barrier to fulfilling one’s demands for activities, which ultimately results in lower SWB. In other words, we posit that travel exerts a role by facilitating/enabling activities to affect SWB. In a theoretical framework, Ettema et al. (2010) suggested that both available travel options and the quality of travel options can affect SWB through activity engagement. Similarly, Bergstad et al. (2011) argued that travel options can have an effect on SWB by influencing the ease with which activities are performed. Empirical evidence for this relationship is also available. For example, Bergstad et al. (2011) found that the effect of satisfaction with travel on SWB decreased significantly when the variable of satisfaction with activities was entered into the regression model. This result was interpreted as the mediating role of satisfaction with activities on the relationship between travel satisfaction and SWB. In a similar vein, De Vos et al. (2017) suggested a spill-over effect of travel on activity engagements, and in other studies, De Vos and Witlox (2017) and De Vos (2019) further suggested an interrelationship between the three factors of travel satisfaction, activity satisfaction, and life satisfaction.

Although the discussion above reveals that travel can, theoretically, have an indirect effect on SWB via activity engagements, the strength of this effect relative to the strength of the (more direct) effect of activity engagements on SWB may depend largely on the specific activity type in question. A highly valued activity that helps realize one’s goals in life may contribute to cognitive SWB substantially that makes the contribution of the associated travel mode insignificant. In other cases, however, the contributions of travel and activity to SWB may become less imbalanced. Another issue is that because SWB refers to well-being over one’s entire life and is not specific to a period, its predictors (i.e., travel and activities) should be measured in the same way. However, measuring travel and activities in long periods can be both costly and time-consuming, and arguably for this reason, most previous studies have used cross-sectional data (see Bergstad et al. (2011), De Vos et al. (2017), and Waterman et al. (2008) for some examples where cross-sectional data are used). Finally, this effect can vary depending on contextual factors that influence travel, activities, and SWB. These factors can be various elements of the transport infrastructure and the built environment, such as the quality of public transport and the spatial distribution of services, and can directly affect travel satisfaction, satisfaction with activities and, ultimately, SWB. In a conceptual model of the influence of travel on SWB, Delbosc (2012) suggested that infrastructure factors, such as pollution, noise, and land use, directly influence mobility, accessibility (to activities) and SWB.

A conceptual model for travel mode use, out-of-home activities, and SWB

Without a lack of generality, it can be assumed that an activity (e.g., eating out) consists of two parts: the travel part (e.g., using the car) followed by the activity part (e.g., eating at the restaurant). The overall satisfaction with the activity thus depends on both the travel mode used and the activity’s attributes. In the eating-out activity example above, travel by bus and by car may lead to different satisfaction with eating out. We therefore assume that the frequencies of the same activity performed with different travel modesFootnote 2 contribute differently and additively to the overall satisfaction with that activity, which then becomes a direct source of SWB. This idea is framed by the conceptual model depicted in Fig. 1.

Fig. 1
figure 1

The conceptual model for the effect of travel mode use on SWB via activities. Note: The ‘…’ symbol denotes other similar elements not shown; satisfaction measures (represented by dashed oval lines) are depicted only for illustration purposes and are unavailable in our final analytic models in Fig. 3

Full size image

A basic assumption of our model is that travel mode use causes activities. At first glance, this assumption may appear to be questionable under the economist view of travel as derived demand because this view focuses on the ‘activity side’, while the ‘travel side’ is secondary and of minor importance. However, there are at least three arguments that support this assumption. First, as discussed in the Introduction, travel mode use plays the role of an enabler/facilitator for activities. The fact that certain activities cannot be done without certain travel modes (e.g., a restaurant that can be accessed by car only) makes it possible to assume an effect of travel mode use on activities. Second, under limited time and resource budgets, activity decisions are directly influenced by travel time and travel cost. For example, the total time of 24 h a day (including time for in-home activities) implies that lower travel times (e.g., using a car instead of a bus) can allow for more (and/or longer) out-of-home activities per day. Rather than being considered separately, travel time and travel cost can also be seen as an integrated part of the activity (i.e., people may consider ‘total’ time/cost—the sums of time and cost of both travel and activity—rather than individual items). This is partly reflected in Small et al.‘s (2007) question, ‘Is the time required for a trip an attribute affecting demand, or is it part of the cost?’ In this case, the influence of travel time and travel cost on activity decisions becomes more evident. Third, in addition to travel time and travel cost, other mode attributes can affect activity decisions in different ways. For example, a person may take fewer eating-out trips just because of long waiting/walking times during bus use. Stressful trips caused by heavy traffic congestion or enjoyable trips with beautiful landscapes experienced while travelling may collectively form different travel experiences, which then become an important consideration in decisions about future trips. This is particularly relevant to leisure and pleasure activities and to people who are variety seeking or curious (Mokhtarian and Salomon 2001). In certain cases, these experiences can affect both the decision to engage in the future activity and the traveller’s satisfaction with the current activity. An example can be found in Mokhtarian and Salomon (2001), where the authors portray an image of a vacation lover who never expects ‘15 h in one or more crowded and noisy airplanes, the 6 h waiting in uncomfortable airports eating overpriced and unpalatable food, and the 3 h of ground access travel in peak-period urban traffic’. In this case, these unpleasant travel experiences can be a real barrier for this person to take this trip again. Similarly, De Vos et al. (2017) note, ‘A stressful trip, for instance, might disturb the execution of − and lower the satisfaction with − the upcoming activity and can therefore reduce the activity’s well-being enhancing effect’. These examples show various ways that mode attributes can influence activities and activity satisfaction. Thus, the basic assumption, although not in line with an extreme economist view, can be expected.

The causality directions in the basic assumption of our model reflect a time-sequential process, i.e., starting with travel mode use and finishing with activities. In this chronological order, the latter action is affected by the outcome of the former action. For example, a negative feeling of stress caused by car use can spill over to a subsequent activity of eating, resulting in less satisfaction with the eating activity. In this respect, our model shares the same principle as the approach in De Vos et al. (2017).

Another aspect of our model is that we consider different travel modes as distinct variables. This allows us to see how each of these modes contributes to SWB. Salomon and Mokhtarian (1998) note that different travel mode uses contribute differently to travellers’ SWB. If we find, for example, that one eating-out trip by car causes a greater increase in SWB than one by bus, this can be interpreted as car use contributing more to satisfaction with eating-out than bus use, all else being equal. This may be due to attributes of travel mode use (e.g., car travel time is more enjoyable) or the spill-over effect of these attributes on later activity (e.g., car use allows the family to eat together).

In summary, the conceptual model creates a theoretical basis for relating different travel modes with SWB through activities. The main idea in this model, i.e., different mode uses cause different satisfaction with activities that later contribute to SWB, serves as the basis for our analytic models in Sect. 4. In the next section, we describe the data used for evaluating our models.

Data

The data used in this study were extracted from the questionnaire data collected in the Japanese Smart Mobility Challenge project (hereinafter referred to as ‘the project’), co-operated by the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) and the Ministry of Economy, Trade and Industry (METI) (METI 2021). This project aimed to improve mobility in the country as an initiative to make transport contribute to efforts to deal with urgent issues, such as ageing, rurality, and regional revitalization. In 2021, an online survey was conducted with people living in 15 experimental regions (out of the total 16 experimental regions in METI’s section) to understand traveller mobility and related factors, including SWB. General information on the data can be found on METI’s website of the FY2020 Smart Mobility Challenge Project, https://www.meti.go.jp/english/press/2021/0402_002.html, and a description of the project can be found in Tran and Hashimoto (2022). Invitations to join the survey were sent randomly to all residents in the 15 regions who were aged 18 and above, and 13,000 answers were collected between 23 and 2021 and 1 February 2021. Data from all the participants were used in this study.

The online survey collected various forms of data. In addition to sociodemographic questions, each respondent reported in detail all the trips made during the latest weekday and the latest weekend day. On each day, the respondents listed all the trips following the time order in which each trip was made. Attributes of each trip were the performed activity at the destination, the travel mode used, and travel time spent in that mode. An important part of the survey was the collection of self-evaluation data of the respondents regarding some aspects of their lives, including indicators of styles of problem solving for depression (Matsumoto 2008), COVID-19 (coronavirus disease 2019) worry, and quality of life. Among the indicators of problem-solving style, the item ‘going out as a measure of problem solving’ followed by four possible answers, ‘Rare’, ‘Sometimes’, ‘Often’, and ‘Always’ was used as a covariate in our analytic models. The (latent) factor of worry about COVID-19 was measured by nine indicators asking about different feelings of worry about COVID-19, each with five possible answers of ‘Not at all’, ‘A little’, ‘A moderate amount’, ‘A lot’, and ‘A great deal’ (Barber and Kim 2021). This latent factor was modelled as an exogenous variable in our analytic models. Finally, the SUBI scale (Sell 1994) used in the survey provided a proxy for SWB. This scale includes 40 statements (see Sell (1994) for more details of the answer format), all translated into Japanese with three possible Likert-style answers of ‘Totally agree’, ‘Agree to some extent’, and ‘Mostly don’t agree’ for each statement. Some descriptive characteristics of the sample are presented in Tables 1 and Fig. 2, and the contents of the items in the measurement scales for COVID-19 worries and SWB are shown in Table 2.

Table 1 Summary of statistics of the analytic sample, N = 13,000
Full size table
Fig. 2
figure 2

The numbers of trips (the z-axis) by region (the x-axis) and activity type (the y-axis). Note: ‘N-leisure’ denotes ‘Nondaily leisure’. The name of each region and its percentage of participants in the total sample are as follows: Region 1: Kitahiroshima City, Hokkaido Prefecture (1.8%); Region 2: Namie Town, Minamisoma City, Futaba Town, Fukushima Prefecture (0.9%); Region 3: Hitachi City, Ibaraki Prefecture, Aizu area, Fukushima Prefecture (7.6%); Region 4: Machida City, Tokyo Capital (12.4%); Region 5: Niigata City, Niigata Prefecture (18.6%); Region 6: Shiojiri City, Nagano Prefecture (1.7%); Region 7: Shizuoka City, Shizuoka Prefecture (16.6%); Region 8: Kosai City, Shizuoka Prefecture (1.6%); Region 9: Hamamatsu City, Shizuoka Prefecture (19.8%); Region 10: Bisan District, Aichi Prefecture (11.1%); Region 11: Eiheiji Town, Fukui Prefecture (0.4%); Region 12: Yabu City, Hyogo Prefecture (0.4%); Region 13: Shobara City, Hiroshima Prefecture (0.5%); Region 14: Mitoyo City, Kagawa Prefecture (1.2%); Region 15: Tsukuba City, Ibaraki Prefecture (5.5%)

Full size image
Table 2 The results of PCA and CFA with the set of indicators for measuring SWB and COVID-19 worry
Full size table

Because the sample includes participants living in 15 specific cities/towns in Japan and due to the nature of an online survey, some important characteristics of the sample need to be mentioned. On average, the sample accounts for 0.31% of the population in each region, ranging from 0.17 to 0.56%, except for the case of Toyoake City in Aichi Prefecture, where 2.08% of its population is in the sample. The gender distribution (i.e., the percentage of males) of the sample, as a whole or stratified by regions, is similar to that in the populations. The greatest difference between the portion of males of a region in the sample and that of the population in the same region occurred in the case of Yabu City, Hyogo Prefecture, where the two statistics were 58.7% (in the sample) and 48.1% (in the population), respectively. Finally, the proportions of older people aged 65 and above in all regions in the sample were roughly three times lower than those in the populations of all the regions, arguably due to the survey method of online questionnaires. In other words, our sample was slightly skewed towards young populations in the regions, a fact that should be taken into consideration when interpreting the findings of our study.

In general, the travellers in our sample mostly had access to car use, with approximately 90% of travellers having a car driving licence and owning/sharing a car. On average, they made approximately 4 trips within two days, with shopping and work being the most popular activities. Many trips were made with car use as car driver or as passenger, and travel time by car was the greatest among travel times by different modes.

Analysis method

In this study, we employed structural equation modelling (SEM) to analyse the indirect effect of travel modes on SWB via activities due to its advantage of modelling latent variables (e.g., SWB). In SEM, the measurement errors and the ordered characteristic of the indicators, which are common in Likert-style scales and exist in our study, can be explicitly considered by a measurement model. Following the conventional approach in SEM, we started with the issue of model specifications. Then, an exploratory analysis was conducted to develop the measurement model. Finally, some issues of model estimation and evaluation are discussed.

Model specifications

In favour of a parsimonious approach, we first posited Model 1, where all possible activities (including ‘other’ activities) stratified by all possible travel modes (including ‘other’ modes) are assumed to simultaneously cause SWB. In other words, the target independent variables are modelled in the form of ‘Activity X by travel mode Y causes SWB’, where X and Y are any activity and travel mode, respectively. Here, ‘Activity X’ denotes the number of times the activity of type X (e.g., shopping) was performed by a person in two days (one weekday and one weekend day). We then compared this model with Model 2 of the form ‘Activity X causes SWB’. Model 2 departs from Model 1 by its implicit assumption of equality in parameters. Specifically, because travel mode use is ignored in Model 2, this is equivalent to assuming that the same activity X, when enabled by different travel modes, still contributes the same amount to SWB. This simple technique, however, allows us to see how the effect pattern changes when each activity is replaced by that activity stratified by different travel modes. Thus, the role of travel mode use on SWB can be revealed. In addition, we estimated Model 3 of ‘Travel time by mode Y causes SWB’ as a measure of the direct effect of travel mode use on SWB. This helped to determine whether there was any correspondence between the direct and indirect effects of travel mode use on SWB. Our hypothesis in this aspect was that each travel mode is characterized by some basic attributes, and these attributes cause some similarities between the direct and indirect effects of travel modes on SWB.

As mentioned earlier, any measure of travel modes and activities should represent one’s travel and activity patterns in general, i.e., not specific to a period in life. However, the three variables ‘Activity X by travel mode Y’, ‘Activity X’, and ‘Travel time by mode Y’ used in the three models were generated from activity data in two recent days and thus were specific to the survey time. As it is impossible to evaluate the representativeness of the dataset used, the possibility of parameters being biased cannot be ruled out. Therefore, interpretations from these parameters cannot be taken with complete certainty. However, it is not straightforward to collect data that accurately capture one’s general travel-activity pattern. In fact, these data are scarce in the published literature, whereas findings from the current study might still be plausible in the study context. We thus decided to use these three model specifications to explore the relationship between travel mode use, activities, and SWB.

To control for personal factors that could influence travel mode use, activities, and SWB (and potentially bias the parameters for the postulated effect of travel mode use on SWB), we included all possible covariates in all these models, and these confounders were allowed to correlate with our target independent variables. These covariates, which were exploited from the questionnaire data, clearly did not cover all possible covariates that mediate the examined relationship. For this reason, readers should note that certain covariates that should be controlled for may be omitted in our models. The list of independent variables thus included activity variables stratified by travel mode used (specific to Model 1), activity variables (specific to Model 2), travel time variables by travel mode used (specific to Model 3), and covariates (available to all models). The inclusion of these covariates (or the so-called confounders/controlling variables in regression analysis) is helpful in reducing omitted variable bias, although their coefficients are not always interpreted as causal effects (Bollen and Bauldry 2011). Examples of applications of this approach can be found in the literature (Allen et al. 2020; Ingvardson and Nielsen 2019; De Oña 2020). The modelled covariates (see the full list in Table 1) included (binarized) sociodemographic characteristicsFootnote 3 (e.g., sex, age), (binarized) car access variables, a (numeric) variable for the problem-solving style of ‘going-out’, an exogenous latent variable for COVID-19 worry, and territorial variables (14 dummies showing the cities/towns in which travellers live). Car access relates to both travel mode use and SWB as it serves as a measure of travel opportunities or motility. Similarly, people with a ‘going-out’ problem-solving style may go out more frequently (leading to more travel mode uses and activities) and potentially have higher SWB. For this reason, the ‘Problem-solving style: going out’ variable was included as a covariate and coded as a numeric variable with values taken from the corresponding Likert-scale scores (1 = ‘Rarely’; 2 = ‘Sometimes’; 3 = ‘Often’; 4 = ‘Always’). The ‘COVID-19 worry’ variable, an exogenous latent variable generated from 8 Likert-scale indicators (1 = ‘Not at all’; 2 = ‘A little’; 3 = ‘A moderate amount’; 4 = ‘A lot’; 5 = ‘A great deal’) showing different aspects of worries related to COVID-19 (see Table 2 and Exploratory analysis for developing the measurement model for more details), was included to account for the fact that the data were collected within the time when this pandemic was happening in the country, and this worry can affect travel mode use, activities, and SWB. The territorial variables were included to account for systematic differences in our sample caused by the differences in the regional characteristics of each city/town (e.g., some regions may have railway/bus systems while others do not, and some regions may have more amenities than others). The specifications of all models are presented in Fig. 3.

Fig. 3
figure 3

The analytic models for travel mode use, activities, and SWB. Note: the ‘…’ symbol denotes other similar variables in the same row that are not shown

Full size image

Exploratory analysis for developing the measurement model

SEM explicitly embeds a measurement model for measuring latent variables, where the selection of relevant indicators for each latent variable plays a crucial role. In this study, the data contain various types of indicators intended for measuring factors of interest, including COVID-19 worry and SWB. For COVID-19 worry indicators, a Japanese translation of the nine items for COVID-19 worry in Barber and Kim (2021) was employed. An examination of the eigenvalues of these items strongly suggested retaining one factor, with only one indicator excluded due to low factor loading. For SWB indicators, however, the set of 40 items in the SUBI scale was originally intended for measuring 11 distinct SWB dimensions (Sell 1994), which are shown in Table 3. In addition, this scale suggested only 7 dimensions when tested using a sample collected in Japan (TONAN 1995). Taking the nature of multiple dimensions of the SUBI scale into account and the possibility that its measurement performance may depend on the empirical dataset being used, we re-examined the dimensionality of the SUBI scale to derive a relevant set of analytic indicators following two steps. In the first step, we excluded all the items belonging to the dimensions that are irrelevant to the study context of this study. For example, the SWB dimension ‘Family group support’ (e.g., helped by family in illness) was considered to relate very slightly to the use of different travel modes and out-of-home activities and, thus, to lack the necessary theoretical basis. In the second step, we conducted a varimax-rotated principal component analysis (PCA) with the remaining indicators using the Psych package (Revelle 2021) in R language (R Core Team 2018) with the following conditions: (1) polychoric correlations were submitted to PCA to consider the ordered characteristic of the indicators; (2) the number of retained factors in PCA was decided by considering both the scree test and the intended number of factors in the original SUBI scale; and (3) any retained indicator must have communality greater than 0.5 (Hair 2010), a main factor loading greater than 0.4, and cross-loadings lower than at least 0.2 compared to the main factor loading (Stamper and Masterson 2002). In addition, the indicators of two factors, ‘general wellbeing negative affect’ and ‘inadequate mental mastery’, were merged to form a single factor in the PCA results. Thus, only one factor (the former one) was retained. The lists of retained factors and indicators for SWB are shown in Table 3. For the indicators of COVID-19 worry, we conducted an unrotated PCA because there was only one factor intended.

Table 3 The selection of analytic dimensions and indicators from the SUBI scale
Full size table

Finally, we conducted a confirmatory factor analysis (CFA) to evaluate the suitability of all the indicators for SWB and COVID-19 worry for use in SEM. The results of PCAs for SWB indicators and COVID-19 worry indicators and the results of CFA for both are presented in Table 2.

The Cronbach’s alpha coefficients for assessing the internal consistency reliability of indicators measuring the same latent variable were slightly lower than the minimum acceptable value of 0.6 (Hair 2010) for the cases of belongingness (0.54) and achievement (0.52). However, this fact can be ignored because Cronbach’s alpha is sensitive to the number of indicators (Hair 2010; Kline 2016), and there are only two indicators per factor in these two cases. The CFA’s model fits indicated a good measurement model as all the factor loadings were greater than 0.6 and all the goodness-of-fit values fell into acceptable ranges, except for the large chi-square statistics. However, the chi-square statistic is sensitive to sample size (Kline 2016), and this criterion can be safely ignored considering our large sample size of 13,000.

SEM estimation and evaluation method

Based on the result of the previous step, i.e., measurement model development, the full set of variables was submitted to SEM to evaluate the postulated relationships. All the variables of interest in Model 1 and Model 2, i.e., activities and activities stratified by travel mode use, were modelled by the numbers of corresponding trips, whereas travel time variables in Model 3 were modelled by travel times in minutes. In fact, some variables were not modelled because their values were mostly invariant. For example, there were only 16 travellers among 13,000 travellers (less than 0.5%) who used motorbikes to go to school. This variable ‘school commute by motorbike’ (and other similar variables) was thus excluded to ensure the stability of the estimation process. First, the model identification issue was checked using the two-step rule (Bollen 1989). With at least two indicators per latent variable and other latent variables with more than two indicators, all models were identified. For model estimation, a diagonally weighted least squares (DWLS) estimator was employed to consider the ordered characteristic of the indicators (Finney and DiStefano 2013). Estimations were conducted separately for each model using the Lavaan package (Rosseel 2012) in R language (R Core Team 2018) with the weighted least square mean and variance (WLSMV) estimation method, a robust version of DWLS. All model performances were assessed using some common criteria, including the chi-square statistic (χ2) (the lower, the better); degrees of freedom (df); normed chi-square χ2/df, with lower values (ideally less than 3) indicating better model fit(Hair 2010); RMSEA: root mean square error of approximation; GFI: goodness-of-fit statistic; CFI: comparative fit index; and SRMR: standardized root mean square residual. The acceptable range for RMSEA should be between 0.03 and 0.08 at a 95% level of confidence (Hair 2010), values for GFI and CFI should be greater than 0.9 (Bentler and Bonett 1980), and SRMR should be lower than 0.1 (Hair 2010).

Results

The effects of travel mode use (via activities) and travel time on SWB in Model 1, Model 2, and Model 3 are presented in Table 4. Because the effects of covariates on SWB are very similar between the three models, we present only this result from Model 1 in Table 5. The goodness-of-fit of all models, including the CFA model, is presented in Table 6. Generally, all models fit well to the data as all the examined fit indices were acceptable. In addition, Model 1 had the lowest normed chi-square, or χ2/df, compared to the corresponding statistics in Model 2 and Model 3, while all the remaining fit indices of these models were almost the same. Regarding the explanatory ability, the three models explained between 11% and 22% of the variance in SWB dimensions. This statistic is acceptable considering that sociodemographic variables can at best account for 15% of the variance in SWB (Diener 1984).

Table 4 The effects of travel mode use and travel time on SWB through out-of-home activities
Full size table
Table 5 The effects of covariates on SWB in Model 1
Full size table
Table 6 The model fit indices for the CFA model and SEMs
Full size table

In Model 1, the effect of travel mode use on SWB through activities varied greatly over different activity types. For working activity, only working by railway was associated with increased belongingness. For school activity, however, school commutes by bus were associated with increased negative affect, while school commutes by bus, car, and walking were associated with increases in both belongingness and achievement. Noticeably, school commutes by bus and walking showed higher effects on belongingness and achievement than school commutes by car use. A similar fact was found in the pattern of business activity’s effects on achievement as the coefficient for the effect of business trips by bus use on achievement (0.157) was nearly double the coefficient for car use as a passenger (0.083).

Shopping activities were found to have the most complex effect pattern on SWB. Whereas shopping trips by railway were associated with more negative affect, conversely, shopping trips by car as a passenger were associated with more positive affect. For the cognitive component of SWB, although shopping trips by most travel mode uses negatively influenced the three dimensions of belongingness, achievement, and confidence in coping, shopping trips by bus uses showed a positive effect on belongingness with a coefficient of 0.112.

As expected, daily leisure and nondaily leisure activities with most travel mode uses positively influenced all SWB dimensions, with only one exception occurring in the case of daily leisure trips by motorbike as these trips were associated with more negative affect. Most of the cases of statistically significant effects involved car use as driver or car use as a passenger. Interestingly, car use as a passenger showed a higher effect on satisfaction with these kinds of activities than car use as a driver, with higher coefficients observed in the cases of positive affect, belongingness, and confidence in coping. For clinic/hospital activities, more trips, even by car as a driver or passenger, were associated with more negative affect, as expected. However, trips by car as a passenger had a positive effect on confidence in coping. For the effects of other trips on SWB, although the types of activities were unknown in this case, these activities were found to positively affect the affective components of SWB.

A comparison between the estimates of Model 1 and Model 2 indicates that although the general trends of the effects of activities on SWB are similar, Model 1 had the advantage of showing how different travel mode uses contributed differently to SWB. Taking the case of shopping and belongingness as an example, whereas shopping by car had a negative effect (-0.018), shopping by bus had a positive effect on this factor (0.12).

In Model 3, we found that longer travel time by bus was associated with more positive affect and enhanced belongingness and achievement. Longer travel times by private motorized modes, i.e., car use as driver, car use as a passenger, and motorbike, were associated with more negative affect. However, travel time as a passenger positively influenced all other SWB dimensions, including positive affect and all cognitive components of SWB. Compared to other travel modes, travel time by bicycle benefited SWB the most because it positively influenced belongingness and achievement but was not associated with a negative affect. For the effects of travel time on belongingness and achievement, the coefficient for bus use was higher than the coefficient for car use as a passenger, a result that is in line with the effect pattern for school trips, business trips, and shopping trips in Model 1.

Many covariates were found to affect SWB, with most of the cases being in the expected directions. First, worry about COVID-19 affected most SWB dimensions in a negative way, except for the case of belongingness, which it positively affected. Being married or a student and having a ‘going-out’ style of problem solving were found to strongly and positively affect SWB in multiple dimensions. The result related to marital status is in line with the positive relationship between marriage and SWB that has consistently been established in the literature (Diener et al. 1999). Being male, however, was associated with less positive affect. Regarding this result, Bergstad et al. (2011) found that the effects of gender on SWB are mixed, with both women being happier than men and no gender difference in SWB. The effects of age on SWB were mixed for SWB’s affective component but consistent for SWB’s cognitive component as older people had greater belongingness and confidence in coping. Both measures of car access, i.e., owning/sharing a car, were found to positively affect SWB, with stronger effects for the case of owning a car.

Discussion

This study attempts to add insights to the indirect effect of travel mode use on SWB through the measure of out-of-home activities. We started with a conceptual model based on the idea that the same activity, if performed with different travel modes, will result in different levels of activity satisfaction, which ultimately contribute differently to SWB. We then estimated Model 1, where the number of trips stratified by travel mode use was assumed to cause multiple dimensions of SWB. We observed how different mode uses contributed differently to SWB via activities. The estimates from Model 2 and Model 3 confirmed the findings in Model 1. In this part, we discuss the findings of Model 1 and Model 3 in comparison with the findings in the previous literature. Any findings presented here should be taken with caution because the participants’ travels and activities were measured for two recent days; as such, they might not be a good representation of travel and activities in general.

Travel mode use and its heterogeneous contributions to SWB

First, public transport use, including bus use and railway use, was found to enhance the SWB dimension of belongingness through activities. This was evidenced by the results in cases of work trips, school trips, and shopping trips in Model 1 and was further supported by the positive effect of travel time by bus on belongingness in Model 3. Together, the results of the two models confirmed our hypothesis that there are some basic characteristics of the trave mode that influence both the effect patterns of travel mode use and travel time on SWB. In addition, we found some results in the literature that support this finding. Jones et al. (2013), for example, found that free public transport travel in London enhanced the sense of belongingness, i.e., ‘being a Londoner’, for both young and older people. Similarly, Stanley et al. (2010) compared some social exclusion indicators and found that those who never used public transport were more likely to be socially excluded. With regard to the new aspect that this study adds to the literature, although using more public transport leads to more social inclusion, both intuitively and by empirical evidence, we showed that this happened specifically for working trips, school trips, and shopping trips, with the greatest effect on belongingness observed in the case of school trips. This finding highlights the potential of public transport use for promoting social inclusion, which might be an interesting fact for both transport and social policy-makers who are interested in SWB and social inclusion in particular.

We also observed that public transport use negatively affected the affective components of SWB. Specifically, we found that school trips by bus and shopping trips by railway were associated with more negative affect. This finding implies that these public transport modes might have some joint attributes that cause a negative travel experience rather than a positive one. Some potential candidates for this might be the waiting time, number of transfers, or dependence on public transport schedules. While it seems that these issues may be due to the nature of public transport and thus sometimes impossible to change, the findings presented in this study at least add some arguments about the different aspects of public transport use that may be helpful for discussions of their role in society.

Despite school and business activities revealing that both bus use and car use (both as a driver and as a passenger) enhanced one’s achievement, the estimates of the effects of travel mode use on achievement showed that the effect in the case of bus use was nearly double the effect of car use. This finding was partly supported by the result in Model 3 that the estimate of the parameter for the effect of travel time by car as a passenger on achievement was lower (nearly half) than that for bus travel time. Thus, while car use generally offers more flexibility than bus use, this does not necessarily imply greater achievement in life.

For daily and nondaily leisure activities, we found a general pattern of positive effects of travel mode use on SWB. This result is also in line with previous studies; for example, satisfaction with leisure activity was found to increase SWB in De Vos et al. (2017). However, the effect of the use of a specific travel mode varied greatly. First, car use as both a driver and a passenger positively affected SWB in multiple dimensions, such as positive affect, belongingness, and confidence in coping. However, the effects of car use as a passenger were consistently larger than those of car use as a driver. In particular, daily leisure trips by motorbike were associated with more negative affect. Thus, the belief that more leisure activities enhance SWB seems to be challenged in this case. The estimates from our study illustrate that the effects of leisure activities on SWB can vary greatly over different travel modes, and they may even reduce SWB, such as in the case of motorbike use.

Some special discussion should be given to the effects of shopping activities by car use on SWB. In the estimation results, we found that shopping by car as a driver was associated with reduced belongingness and achievement. First, the main purpose of shopping activities seems to be to satisfy individual daily necessities rather than to facilitate social activities. As such, we do not expect this form of activity to increase feelings of belongingness. For the reduction in achievement caused by shopping activities, as this is a common research topic in economics, we introduce the adaptation theory when seeking possible explanations for this phenomenon. Richins (1987) explained the negative relation between materialism and happiness by citing the adaptation theory and argued that once an individual has obtained a desired status, the level of expectation for that status increases, resulting in ‘a gap between state and expectation’. It is this gap that causes feelings of dissatisfaction or nonachievement. Consequently, those who make more shopping trips may be subject to the feeling of nonachievement in their efforts to satisfy the short-term needs of having goods. There is also empirical evidence that supports this theory. The study by Kahneman et al. (2004) showed more positive affect and less negative affect of leisure and relaxing activities than shopping activities, and White and Dolan (2009) found a lower pleasure level of shopping activities compared to leisure activities. We argue that travel mode use plays little role in this case, as satisfaction with shopping might be heavily influenced by shopping activities. This reflects our previous discussion that the indirect effect of travel on SWB via activities might become insignificant when the effect of the activity itself on SWB is large enough. In fact, we only found a significant (negative) effect of shopping activities on achievement in the case of car use as a driver; thus, we were unable to compare the contributions of different travel mode uses on satisfaction with shopping. However, the results presented here suggest that car use, not other mode uses, is involved in this complex relationship.

The result in Model 3, i.e., between travel times by different modes and SWB, revealed some interesting facts about the (direct) relationship between travel mode use and SWB in addition to its role as a confirmation for the result in Model 1. First, travel times, whether by public transport, car use, motorbikes, or bicycles, were positively associated with SWB’s dimensions of belongingness and achievement. This finding is related to the observation in Stanley et al. (2010) that those at greater risk of social exclusion travelled less distance. Another finding is that car travel time, both as a driver and as a passenger, was associated with more negative affect. This result confirmed previous findings on the negative effects of car use on drivers’ satisfaction with travelling. For example, there is evidence that car use causes mental stress (Bellet et al. 1969) and is associated with a higher level of stress than train use (Wener and Evans 2011) and bus use (White and Rotton 1998). Similarly, traffic congestion can harm travellers’ moods (Morris 2015). Finally, travel time by bicycle did not have an effect on positive or negative affect but was associated with increased belongingness and achievement. Mokhtarian and Salomon (2001) argue that humans have an intrinsic desire to travel and be exposed to the environment. We believe that this might be an explanation for this finding.

How can transport policies be shaped towards well-being?

Ideally, understanding how travel mode use exerts its influences on activities to affect SWB would lead to suggestions to make SWB more beneficial with regard to travel and activity aspects. Considering this, we examined which specific combination of activity and travel mode contributed to a specific dimension of SWB in the results of Model 1. For SWB’s component of positive affect, among the activities investigated, only shopping and daily/nondaily leisure enabled by car use were found to be significant correlates. For SWB’s components of belongingness, working, schooling, and shopping activities enabled by public transport use and nondaily leisure activities enabled by car use were identified as significant correlates. Finally, significant correlates of SWB’s component of achievement were schooling and business activities, both enabled by public transport use and car use, and daily/nondaily leisure activities, enabled by car use. The identification of these correlates of positive affect, belongingness, and achievement serves as a basis for suggesting measures to shape transport policies towards enhancing SWB, which are presented in the following.

First, the above findings suggest that promoting more private transport, such as car use, can lead to the outcome of positive affect (and achievement) for travellers, whereas greater priority for public transport, such as railway and bus use, may result in enhancing belongingness (and achievement). This multifaceted aspect of travel mode use must not be ignored when evaluating possible consequences from travel mode use to society. By explicitly considering these different SWB outcomes of each travel mode, one can eventually identify a mode-share pattern that benefits overall SWB or some specific dimensions of SWB. We particularly suggest this consideration when evaluating some emerging mobility prototypes, such as demand-responsive mobility and mobility-as-a-service (MaaS), in rural areas where services are often sparsely distributed, which can cause high dependence on car use. These mobility measures often mimic car use by making travel more individually served, e.g., in the case of on-demand taxis. As such, they can benefit cohort groups who do not have access to car use, such as elderly individuals. On the other hand, promoting public transport use is also the target of these measures. In this aspect, they benefit public transport systems in rural areas, which are in many cases underoperated (see Tran et al. (2020) for the case of local buses in Japan). As a result, promoting these new travel prototypes can benefit SWB in terms of positive affect, belongingness, and achievement while also having the potential to reduce car ownership. In urban areas where services are often more accessible and private transport may be discouraged, the above findings become less relevant. However, if ‘urban’ MaaS is aimed at reducing car ownership by providing a comparable multimodal travel mode, we expect that the complexity of MaaS, such as multimodality and subscription plans, may cause its effects on SWB via activities to become more complex. More attention should be given to this issue.

There is also potential for shaping transport policies towards SWB by collaboration between transport and other sectors related to out-of-home activities. In this aspect, there are at least two patterns that have been revealed in the results of Model 1: the combination of shopping and leisure activities with car use was associated with positive affect, and the combination of work, school, and business activities with public transport use was associated with belongingness and achievement. These patterns of effects imply that well-being can be promoted by stimulating demands for (1) shopping and leisure activities together with car use and (2) work, school, and business activities with public transport use. This suggests that travel is eventually a derived demand and that intervention should involve out-of-home activities. However, we argue that the transport sector can still play a role here by acting as a facilitator in realizing these activities and, therefore, being involved in nudging demands for out-of-home activities. This has been promoted in Japan in various forms of collaboration between transport operators and other service providers. For example, small-scale retailers and operators at entertainment facilities are collaborating with mobility service companies to provide incentives, e.g., a free taxi and discounts for shopping, to trigger more traveller demands for shopping and leisure activities. Similarly, in Shizuoka City, Japan, some companies provide their own shuttle buses for their workers to commute to work, which can be considered a form of collaboration between the transport sector and industrial sectors. These buses are now also open for all residents to access various facilities, such as shopping centres, through a collaboration platform in the Smart Mobility Challenge project described in the Data section while retaining their main task of transporting workers. In addition to these two patterns, interestingly, we found nondaily leisure trips combined with car use as both drivers and passengers to be associated with belongingness. As nondaily leisure trips in this study were mainly trips of long travel distances, this result implies that the feeling of belongingness can be enhanced through leisure trips with car use to distant destinations. This information might be interesting to policy-makers of social inclusion, tourism, and mobility services.

In brief, travel mode use contributes in different ways to SWB. Because of this, the first step in promoting SWB from a travelling perspective should be simply to explicitly consider this aspect. Travel mode use, although largely a means of performing activities, can also be utilized to nudge demands for these activities and thus enhance the ultimate outcomes of SWB.

Study limitations and some suggestions for future studies on this research trend

A limitation of this study was the discrepancy between the hypothesized relationship and the availability of the data that allow the relationship to be tested. Specifically, general SWB describes one’s general subjective assessment of well-being in life, and any measures intended to relate to this concept should be constructed in the same way. However, we had only data on travel-activity behaviours in the two recent days. Such data may fail to capture full variations or key patterns in the travel-activity behaviours of the respondents. Another limitation was that the number of modelled covariates might not be sufficient to cover the full list of variables that mediate the examined relationship. Compared to domain-specific SWB, general SWB may relate or vary according to more factors, including both personal and nonpersonal factors. This fact implies the need to collect a wide range of variables so that parameter biases can be minimized.

We have suggestions for future studies on this topic, which come from both the results of this study and our discussion with the reviewers of this paper. First, we have illustrated that the effect pattern of activities on SWB differs significantly with different travel modes. For example, although overall shopping activity (in Model 2) and shopping activities by car use (in Model 1) were associated with less belongingness, shopping with bus use positively affected this SWB dimension. Thus, the action of summing or averaging activity frequencies performed with different travel modes to form a general measure of activity frequency may result in less accuracy than a stratified approach, which was used in this study. We thus encourage the use of a stratified measure of activities, such as in Model 1 in this study, instead of a general measure. Second, except for some variables, our results generally support the fact that positive affect and negative affect are distinguishable and negatively correlated, and affect and satisfaction (i.e., cognitive component of SWB) are differently influenced by travel mode use and activities. The outcome variable of SWB thus should be considered in its multiple dimensions rather than as a single factor. There is also potential to obtain more insights into this topic by exploiting the information of activity durations, which was not available in this study. Travel mode and travel time can directly influence activity duration, a factor that potentially affects activity satisfaction (Schwanen and Wang 2014; De Vos 2019). Thus, possible indirect effects of travel mode and travel time on SWB can be revealed by the measure of activity durations. Similarly, longitudinal travel-activity data (Axhausen et al. 2002; Hanson and Huff 1988) have the potential to more accurately reflect the relationship between activities and SWB, which was unavailable in the current study based on cross-sectional data. Such data sources can help to explain the day-to-day variation in travel behaviours (Susilo and Axhausen 2014). However, a low participation rate can be a potential issue in longitudinal surveys. Finally, the conceptual model in this study can be generalized to consider the effects of any other aspects of travel behaviours and is not limited to the case of travel mode use in this study.

Conclusions

In this study, we explored how travel mode use indirectly affects SWB through the measure of out-of-home activities. We proposed a conceptual model relating various activities stratified by travel mode use with SWB’s five dimensions of positive affect, negative affect, belongingness, achievement, and confidence in coping. The conceptual model was operationalized using SEM and data extracted from the Smart Mobility Challenge project in Japan. The estimation results showed that the effects of activities on SWB varied over activity types, travel modes used, and SWB dimensions. In addition, we observed some general patterns with more details of these effects. Specifically, public transport use, despite enhancing SWB’s cognitive components, was also observed to have a negative effect on the affective components of SWB. Car use affected SWB in a more complex pattern: it promoted SWB by enabling daily leisure and nonleisure activities but also reduced SWB dimensions of belongingness and achievement through shopping activities. Active travel modes did not have a clear effect on SWB; for example, walking to school was associated with increased belongingness, whereas walking for shopping negatively influenced belongingness. For transport and social policies, our key message is that although the contributions of activities to SWB may depend strongly on activity attributes—for example, good weather conditions and the ability to view beautiful scenes may be important attributes that determine the satisfaction level of sightseeing activities, which then contribute to SWB’s dimension of positive affect—travel modes with the role of the enabler/facilitator of activities may also play a role in these contributions.