The impacts of mediating the work environment on the mode choice in work trips

The impacts of work characteristics on travel mode choice behavior has been studied for a long time, focusing on the work type, income, duration, and working time. However, there are no comprehensive studies on the influence of travel behavior. Therefore, this study examines the influence of work environment as a mediator of socio-economic variables, trip characteristics, transportation infrastructure and services, the environment and choice of transportation mode on work trips. The mode of transportation consists of three variables, including public transportation (bus rapid transit and mass rapid transit), private vehicles (cars and motorbikes), and online transportation (online taxis and motorbike taxis online). Multivariate analysis using the partial least squares-structural equation modeling method was used to explain the relationship between variables in the model. According to the results, the mediating impact of work environment is significant on transportation choices only for environmental variables. The mediating mode choice effect is negative for public transportation and complimentary for private vehicles and online transportation. Other variables directly affect mode choice, including the influence of work environment.


Introduction
To forecast transportation demand, a thorough understanding of travel behavior is necessary. Transportation experts and psychologists have broad study experiences on travel behavior, including mode choices during work trips. The general characteristics of work trips include routines on and from trips (commuting), though mode choice is one of the essential stages in transportation planning [1]. In general, mode choice directly affects the transportation policy because it can predict behavior and influencing factors [2]. The decision to travel reflects the alternative modes available, and therefore, it is essential to plan urban transportation systems, especially forecasting travel demand [3,4]. The pressure due to private vehicle ownership has changed lifestyles and increased outdoor activities, pollution, and congestion [5].
From various mode choice studies, several models have been developed. However, there debates on the right model as choices relates to the inconsistency of people's behavior [6]. Seemingly, various studies reviewed travel demand management to reduce private vehicles' use for satisfaction since 2000 [7]. However, travel model development is mainly used to overcome transportation problems, such as traffic jams, accidents, and pollution [8].
During the last decade, studies related to mode choice on work trips have increased tremendously. Most of them have emphasized on factors influencing workers to use modes and policy solutions in controlling mode choice behavior and realized workers' smart mobility. Frequently studied factors relates to socio-demographic aspects and individual/household characteristics, such as gender, age, number of family members, vehicle ownership, and income [9][10][11][12][13][14][15]. Other completed studies focused on travel characteristics, including time, cost and travel distance. Time and cost studies focused on travel time value and its relationship to costs [16]. Additionally, waiting time in modeling travel behavior, leads to satisfaction [17], while travel time is an essential aspect of activity-based travel modeling [18]. Travel distance relates to the existing road network structure and affects time and transportation mode [19].
In the transportation system analysis context, travel occurs through the interaction of transportation systems (modes and infrastructure of transportation) and the activity system [20]. A study on transportation system influence on work trips showed that well-connected road networks reduce travel time to work [19]. Infrastructure development, based on intelligent systems, increases public transportation use [21]. Also, good network layout and connectivity increases public transportation use [22]. According to Irfan et al. [23], improving transit services and implementing congestion charges encourage transit use compared to private cars. Additionally, limiting the availability of free parking reduces the use of private cars [24]. According to Huang [25], good accessibility reduces travel time to job location.
Several latent variables that relates to attitudes, behaviors, and lifestyles showed an increased model in explaining modes of choice behavior [26]. Donkor et al. [27] established that attitude as a latent variable and public transportation perceptions affect transportation mode choice for the round trip. However, attitudes towards public transportation mode cannot influence a person's decision to use it in case they own a car [28]. People living in residential areas that use more private vehicles are dissatisfied with their trips than those walking to work [29]. When using the integration choice and latent variable approach, which assumes aspects of security, comfort, safety, public transportation services, and feelings of waiting as latent variables, there is high accuracy in explaining behavior than the traditional logit model [30].
Several studies have examined environmental variables, including land use, as factors influencing mode choice. The built environment and its integration with the road network changed modes' selection in working trips [31,32]. Land use patterns that produce a certain distance between home and workplace also allow the choice of transportation modes [33].
Some studies analyze aspects related to work because travel is done in the framework of work. Examples of such aspects include the duration of work and start time of work, which affected the choice of transportation mode [34]. Moreover, regulating work duration and start time through management policies (office) leads to flexible work travel [35]. Another challenge to changing car users' behavior to public transportation is the need of incentives stimulation for employees to public transit. However, providing incentives does not make employees walk or use public transportation [24]. Conclusively, factors influencing the mode of choice by people on work trips including the availability of modes, trip, network and work (work time and duration) characters, individual characteristics (socio-demographic, household, ownership vehicle) and attitude or perception related variables [36].
Seemingly, studies related to 'work characteristics' have not comprehensively focused on their effects on mode choice behavior. Efforts to explain office management factors' influence are more related to policy on employee's aspects. Contrastingly, the full work environment has not been fully studied hence has opportunities for policies that can directly lead to realization of smart mobility. It is based on considerations that the work environment is a setting, situation, and condition showing employees' existence. There is also a relationship between employees, employers, and the work environment, including technical, human, and organizational climate [37]. Work environment manifestation occurs through working hours, security and safety, relations with coworkers, esteem needs, and top management [38]. Moreover, the work environment can be viewed based on psychological, physical, and social environment approaches [39,40]. Some studies show a strong relationship between work environment and employee behavior, which relates to office policy, aspects of the work atmosphere and psychology [41,42].
This study examines the effect of latent variables work environment (WE), socio-economic (SE), trip characteristics (TC), transportation infrastructure and services (TIS) and the environment (Env) on the choice of transportation modes. Also, the study places work environment as mediation between other variables and transportation modes choice. Consequently, study raised questions are (1) how much influence each of the SE, TC, TIS, Env, and WE variables has against the choice of transportation mode and (2) whether the variables SE, TC, TIS, and Env have an indirect effect on the choice of transportation mode through WE mediation variables. Figure 1 shows the general concept of model development, where SE, TC, TIS, and Env are independent variables and modes of transportation is a dependent. The WE is mediated between the independent and dependent variables. It is an endogenous variable when viewed from a direct relationship with the SE, TC, TIS, and Env. However, it is an exogenous variable when viewed from a direct link with the transportation mode. To answer the study questions, the partial least squaresstructural equation modeling (PLS-SEM) was used. SEM is a multivariate analysis technique with the most accurate estimation for simultaneously analyzing multiple regression equations [43]. Comparably, PLS is an SEM-based approach on covariance analysis where data are assumed to be abnormally distributed and uses different settings to determine the factor scores [44,45]. Moreover, PLS-SEM analyzes complex constructs, indicators, and structural paths [46]. This technique has widely been used in transportation studies to evaluate public transit and road safety [47,48], traffic accidents [49], public transportation [50,51], and taxi services [52].

Area of study and transportation characteristic
This study was conducted in Jakarta, one of the largest metropolitan cities in Indonesia. Jakarta is divided into six administrative regions covering the North, South, East, West, and one archipelago island ( Figure 2) [53], totaling to 662.33 km 2 [54]. Currently, the modern operational modes of public transportation in Jakarta include bus rapid transit (BRT) and mass rapid transit (MRT), electric rail trains, and light rail transit (LRT). Also, online modes of transportation positively affect the decisions to travel. This study used Transjakarta BRT, Jakarta MRT, cars, motorbikes, online taxis, and motorbike taxis online to evaluate work trips. The Transjakarta Company has experienced a significant increase in bus fleets in the last five years [55]. The Consequently, Transjakarta bus passengers have increased by 136% during the same period [55]. The Jakarta MRT was launched in April 2019, and 79,114 people use its service per day [56]. Online transportation is overgrowing in all cities in Indonesia because of its convenience, web-based ordering ease [57], and availability of discounted programs and vouchers [58]. Comparably, passenger cars and two-wheeled vehicles have increased every year by an average of 6% for cars and 5% for motorcycles. The number of cars in 2018 reached 4,133,338, while motorbikes totaled to 16,155,961 units [59].
According to TomTom Traffic data, there was an 8% decline in Jakarta's traffic congestion from 61% to 53% in 2018 [60]. Although the congestion index remained 53% in 2019, Jakarta's position in the most congested cities worldwide fell from 7 to 10 [60]. This data shows that congestion reduction did not lower its index to below 50%. Compared to other countries, especially in Europe and America, cities have average congestion indexes below 35% [60]. This comparison shows that the applied strategies are not optimal in reducing private vehicle dependency. In 2018, 49.5 million trips occurred in Jakarta per day, 23.4 million from the city, while 20.02 million were private vehicle owners from Bogor-Depok-Tangerang-Bekasi (Bodetabek) [61].

Model design and hypothesis
In PLS-SEM, the relationship between variables is shown through a partial regression equation. Each endogenous construct is the dependent variable, and the exogenous construct represents the independent variable, while the indicator measures the construct [62]. Accordingly, there are two elements in the model, specifically the path structure model which is also known as the path structural and outer (measurement) model (reflective and formative indicator model) [63]. Table 1 shows the variable and indicator description.
The structural model is a direct relationship between the variables SE, TC, TIS, Env, and WE concerning the choice of transportation mode and the indirect connection of the variables SE, TC, TIS, Env through WE to transportation choice mode. Modes of transportation were categorized into 3 variables, including public transportation (PT), private vehicles (PV), and online transportation (OT). Consequently, there were 12 patterns of two indirect relationship segments through the WE variable. Also, there were 15 patterns of direct ties to the transportation mode variable with four direct relationships, between SE, TC, TIS, and Env vari- Trip characteristics (ξ 2 ) Travel cost Travel time Distance Travel time to work Travel time to home Transportation infrastructure and services (ξ 3 ) Availability of transportation infrastructure Availability of parking area Connectivity between modes Capacity and services of public transportation Access to public transportation Information technology The accessibility of workplace location The residential location Road conditions around the residential environment Weather and climate conditions Work environment (η 1 ) Work time flexibility Duration Incentives for users of public transportation Sanctions for violating work time discipline Travel during working hours Job satisfaction Public transportation (η 2 ) Transjakarta BRT, Jakarta MRT Online transportation (η 4 ) Online taxi Motorbike taxis online ables and WE. Figure 3 shows the pattern and relationships between variables.
After determining the structural model, the outer model was designed by identifying whether the reflective or formative indicator model formed the variable. Suppose the formative indicator models, changes in construct (variables) are caused by indicator and reflective indicator models if changes in the indicator are caused by construct [64]. The indicators showed a formative relationship in the SE variable because effects in X 1 to X 6 caused changes in the construct. X 7 through X 12 were formative indicators because trip characteristics changed based on them. Moreover, X 18 through X 21 were categorized as formative mod-els because environmental variables caused the indicators that shape them. However, transportation infrastructure and service variables were categorized as reflective because changes in these two factors affected indicators X 13 to X 17 .

Evaluation model parameters
Evaluation was conducted using the outer and structural model where by the outer model determined whether reflective and formative indicators meet validity and reliability requirements. Contrastingly, the structural model determines how strongly the situation is explained and the relationship between variables. Table 2 shows the parameters used to evaluate the structural and outer models [46,64,65]. R 2 and Q 2 parameters measured the influence of SE, TC, TIS, Env, and WE variables on transportation mode choice (to answer the first research question). Moreover, path coefficient relevance evaluated the mediation effects of direct and indirect relationships between variables (to answer the second research question).

Data collection and sample
Data was collected using a survey questionnaire designed to measure variables indicators, as shown in Table 1. Nominal scales analyzed TC variables, while the SE variables used ratio data. Furthermore, questions measuring TIS, Env, WE, PT, PV, and OT variables used a scale of 1-10, where one and ten presented strongly disagree to strongly agree respectively.
Formal sector workers with monthly income jobs in legal businesses, organizations, and government agencies were used as the study sample. Respondents were Jakarta residents who travelled from home to work every day following their respective jobs' characteristics.
Data was sampled through a simple random technique, a method widely used for mode choice surveys [66]. Questionnaires were distributed to activity centers such as private offices, government offices, business, and trade centers. The number of respondents required in the PLS-SEM analysis using the survey method was determined based on the formula [67] below:  [67]. In this study, 331 of the 350 survey questionnaires followed the appropriate criteria.

Outer model evaluation
The outer model test evaluated the validity and reliability requirements of reflective indicators because it has a reflective model. Contrastingly, formative indicators were tested with a different approach because the outer model cannot evaluate ordinary formative validity and reliability [51,65].
The evaluation of reflective models measured convergent and discriminant validity and composite and internal reliability. Measurement of convergent validity refers to loadings values > 0.7 [46,64] and p-values < 0.05 [65]. However, there are also more conservative opinions by still accepting loadings > 0.5 [68]. Moreover, other parameters were used to assess convergent validity based on the average variance extracted (AVE) value > 0.5 [46]. Therefore, some indicators in the reflective model (Table 3) must be excluded from the model if it refers to a value of loadings > 0.7. However, if it refers to a value of loadings > 0.5, all indicators remain included in the model. However, judging from the AVE value of 0.458 (Table 5), the TIS variable does not meet the convergent validity requirements if using all the model indicators.
Several studies have used Fornell-Larcker criteria to latent variables discriminant validity [69], while some apply cross-loadings methods [70,71]. The cross-loadings value in Table 3 shows that indicator X 13 has a lower loading value on its variable (TIS) than on PV. Consequently, the X 13 indicator had the lowest loadings (0.511) and was removed because it did not meet the discriminant validity requirements. Table 4 shows that each latent variable had a greater AVE square root than its correlation value. Moreover, the TIS reflective variable AVE square root of 0.677 is   The composite reliability > 0.7 and Cronbach's alpha > 0.7 parameters were used to evaluate the reflective model reliability [46,63,65,68]. Table 5 shows that the composite reliability and Cronbach's alpha values were > 0.7, hence the reflective indicator model met the reliability criteria.
The formative model was evaluated based on the significant value of p-value (< 0.05) and VIF to ensure formative latent variables' suitability. The same concept applies to multiple linear regression analyses, which, in many aspects, is parallel to the PLS-SEM [65]. VIF examination confirmed the presence or absence of multicollinearity in the model. The ideal VIF value is less than 3.3, but more conservative provisions allow VIF < 5 up to < 10 [65,72]. This study used the VIF limit < 5 as shown in Table 6 whereby the p-value of indicator X 1 is less than 0.05 and its VIF value > 5 hence excluded from the model. Similarly, the indicator X 11 was p-value > 0.05, hence it was excluded from the model even with VIF < 5. The indicator variables excluded from the model are age (X 1 ), travel time to home (X 11 ), and availability of parking space (X 13 ). From the second analysis, all reflective indicators' loadings were > 0.5, AVE from the reflective model was > 0.5 (0.513). It also had a composite reliability coefficient of 0.838 and Cronbach's alpha coefficient of 0.757. The cross-loadings and AVE roots in the first analysis showed no significant changes hence the reflective model in the second analysis met the validity and reliability requirements. Furthermore, all formative indicators of SE, TC, Env, WE, PT, PV, and OT variables were significant < 0.05 with VIF < 5 and proceeded to the structural model test.

Structural model evaluation
The structural model showed the influence of one variable on another. Secondly, coefficients R 2 and Q 2 were used to explain the strength of exogenous variables and each variable's predictive validity. R 2 is low if 0.25, moderate if 0.5, substantial if 0.75 and Q 2 > 0 is small, > 0.25 is middle, > 0.5 is high [46], hence the coefficient only applied to endogenous variables. However, a prior examination of collinearity was needed to ensure an unbiased regression analysis [46]. VIFs values of all latent variables were seen to meet the requirements < 5 [72], which showed no collinearity symptoms in the model (Table 7). Table 7 also shows an R 2 coefficient of 0.236, WE variable of 0.391 for the PT variable, 0.347 for the PT variable, and 0.341 for the OT variable. This shows the effect of SE, TC, TIS, and Env on the WE variable is relatively weak (< 0. 25) or was only 23.6% and other variables outside the model influenced the rest. Contrastingly, the effects of SE, TC, TIS, Env, and WE on transportation modes > 0.25, which means the effect did not reach moderate levels. The Q 2 coefficient for endogenous variables WE, PT, PT, and OT was > 0.25 hence classified in the medium category.
The evaluation of direct and indirect variables relations in the model based on p-value < 0.05 [46,63,65] is Figure 4 and Table 8. From the results, TIS, Env, and WE variables had a significant effect on all modes of transporta-

Discussion
The mediating effect of the WE variable and the influence of other dependent variables on the choice of transportation mode is known. Results showed that the mediating effect of WE variables on transportation mode choice was signifi-cant for the Env variable. The indirect relationship between the Env variable with PT was negative, with a path coefficient of -0.07 and a significant level <0.05 (Table 8). Consequently, growing environmental conditions change, and the working climate reduces public transportation modes. The indirect effect of Env on PV mode was significant with a path coefficient of 0.115, which is more than its effect on OT mode (0.072). These results showed that the increasing use of PV and OT modes is associated with increased environmental conditions through the work environment. Also, the Env variable positively affects PV and OT mode choice but is negative for PT mode. As in previous studies, the results proved that the integration of road networks in settlements with the built environment substantially influenced mode choices [32]. Also, residence and work locations have an impact on travel mode choices [73]. Workplace accessibility, residential locations, environmental road conditions, and increased climate change reduced public transportation use. Therefore, an integrated area arrangement with public transportation is essential and follows private vehicles' use restrictions.
The WE variable's direct influence had a significant effect on all choices of transportation modes. However, WE on PT's choice impact is significant with a negative path coefficient is −0.163, meaning that increasing changes in the WE variable effect decreased PT's use. Moreover, an increase in the flexibility indicator of working hours does not necessarily increase public transportation use. Contrastingly, previous studies show that some working hours settings can increase flexibility and encourage public transportation [35]. The finding was in line with several other studies stating that increasing work duration impacts the reduced use of PT mode [26,74]. Giving incentives to public transportation users decreases the use of that mode of transportation [24]. Increased work time violations, travel time during work hours, and job satisfaction decreased PT mode choice, hence policies can be used to control the use of private vehicles on work trips. The positive effect of WE were PV (0.270) and OT (0.169) modes, while opposite behavior increased flexibility of working hours, duration and provision of incentives, sanctions for disciplinary violations, travel during working hours, and job satisfaction.
Apart from work environment, TIS variable also affected all modes of transportation choices significantly. Other indicators, including transportation infrastructure, mode connectivity, public transportation capacity and service, access to public transportation, and information technology in transportation, significantly affected transportation choice. Therefore, the provision of good urban transportation infrastructure and accessibility provides ease in traveling, hence impacting the choice of mode of travel [19,21,22,75]. In the PV mode, the path coefficient was negative by (−0.280), which increased the TIS and reduced the use of PV. There was a strong relationship between increased transportation infrastructure and the PT mode use. PT path coefficient was 0.553, providing transportation infrastructure and improving services will continuously reduce the dependence on private vehicles.
The SE variable only affected the PV and OT modes compared to the TIS and Env variables, which significantly impacted all modes. Moreover, the SE variable is positively impacted PV with a path coefficient of 0.180 and was negative to OT mode (−0.096). These impacts show that increased family members, income, vehicle ownership, driving experience, and work experience increased PV use. To control this effect, ownership control policies, for example, limiting carbon dioxide emissions [76], production years vehicle sanctions, and other alternatives can be implemented.
Finally, the TC variable affected online transportation mode with a path coefficient of −0.129 such that increased costs, time, and distance and travel time to work decreased the use of OT. However, the OT is user simplicity [57], and service improvement [77] will sustain the online transportation network.

Conclusions
This study examined the indirect effect of independent variables on socio-economic, trip characteristics, transportation infrastructure and services, and the work environment as mediating variables. Results showed that the work environment variable's mediating effect occurs on environmental variables only, while other variables directly influence the mode choice. Moreover, the most significant influence was on private vehicle mode (11.5%), hence the need for good environmental and work policies to reduce its use.
Transportation infrastructure and services had the highest direct effect on mode choice by contributing 55.3% to public transportation. Furthermore, the work environment had the largest positive direct impact on private vehicles by 27%. In online transportation, the most significant positive influence from environmental variables was 32.2%. Contrastingly, public transportation was negatively impacted by the work environment by 16.3%. The private vehicle mode was negatively impacted by transportation infrastructure and services by 28%. The most significant negative effect on online transportation was from the trip characteristics at 12.9%.
Providing infrastructure and transportation services is not sufficient to encourage public transportation use on routine work trips. The government should implement infrastructure development with regional managements that increases mobility and high accessibility for public transportation. Moreover, transportation policies should be integrated with workplace policies. Some general transportation policies include efforts to reduce travel (avoid), using public transit (shift) and technological innovations related to environmentally friendly vehicles (improve). Therefore, integrating workplace policy with an avoid-shift-improve strategy is an excellent way to encourage workers to use public transportation.
This study enriched previous literature through the inclusion of work environment factors as variables that influence work trips. Comparably, previous studies examined work factors in terms of work duration and efforts to provide public transportation users incentives without analysing its effects. However, work-related aspects are comprehensive and cannot be explained by the duration and working time. The study grouped these variables into the work environment by adding indicators of sanctions for violations of work time discipline, travel during working hours, and job satisfaction, and the results showed significant effects. Some of the variable indicators to use as additional study materials include 1) Travel during working hours, can be used as study material to answer how these trips are carried out. 2) Time to travel during working hours can analyse congestion effects and levels in urban areas. Furthermore, further research can add other variables such as the nature of work and the Covid-19 effects on travel choices. For future studies, this model will be developed with a consumer psychology approach by assuming that employees are consumers to describe the mode choice behavior.