Analysis of travel time and mode choice shift for non-work stops in commuting: case study of Beijing, China

Previous studies have identified the influential factors on travel time for non-work activities. Personal and household attributes such as gender, age, income, household (worker) member numbers have been found to influence travel time (Golob et al. 1995; Schwanen et al. 2002; Tarigan et al. 2012). Regarding trip attributes, while shopping travel time decreases as their accessibility to vehicles increases in Canadian cities (Farber et al. 2011), in Dutch cities that discourage car use, drivers takes a relatively long time (Schwanen et al. 2002), Among the built environment factors, residential context has especially been examined (Schwanen et al. 2002; Cao et al. 2009).

Another factor influencing travel time is the duration of the activity at the destination. With respect to the direct relation between activity and travel time, some studies show that travel time increases with activity duration (Kitamura 2002; Kitamura et al. 1998; Levinson and Kumar 1995). However, individuals’ activity-travel behavior in urban space is constrained to the inseparability and scarcity of space and time (Dijst 1999; Hägerstrand 1970; Kondo and Kitamura 1987; Neutens et al. 2011). In other words, there is a trade-off between travel and activity time given the limited time budget of each individual (Dijst and Vidakovic 2000). TTR developed by Dijst and Vidakovic (2000) and the travel time price (TTP) by Chen and Mokhtarian (2006) are defined to to investigate this trade-off between travel time and activity time. TTR is defined as a ratio of travel time allocated to the total amount of travel time and activity time, whereas TTP is a ratio of travel time allocated to activity time, and the two concepts can be converted as TTP = TTR/1 − TTR. The formulation for TTP can be transferred as (Eq. 1):In this expression, Tj is the total travel time used for activity j, \(D_{j}\) is the duration of activity type j and \(\varphi_{j}\) is the TTP value.

This trade-off relation differs across different activity types, and this has been explored in various empirical studies. Dijst and Vidakovic (2000) first applied the TTR concept to non-work activities for a small sample of dual-earner families, and found that the ratio for maintenance activities ranged between 40 and 50% versus 20 to 30% for discretionary activities. Then, the travel time ratio for work activities was discussed and calculated to be 10.5% based on data from the 1998 Dutch National Travel Survey (Schwanen and Dijst 2002). Chen and Mokhtarian (2006) compared not only the TTP values for maintenance and discretionary activities but also their elasticity. Their findings indicated that travel time spent on maintenance activities was less elastic than that spent on discretionary activities.

With the exception of Dijst and Vidakovic (2000), the studies noted above examine the trade-off in single-purpose trips. Because trip chaining is now a common practice in our daily lives, the TTR and TTP concepts have been extended to explore the trade-off of activities in multi-purpose trips. Susilo and Dijst (2009, 2010) calculated TTR values for both mandatory and discretionary activities for multi-purpose trips. They first computed the travel time related to the secondary activity by deducting “the travel time needed for a direct trip between bases” from the “actual total travel time”. In the case of commute trips with more than one additional activity location, their solution was to distribute the TTR value based on the weighted durations of the activities. They employed multilevel regression models to explore the determinants of TTR on journey episode, individual, household and environmental levels. Their results show the efficiency of multi-purpose trips, as the number of activities chained has a negative influence on TTR. Soo et al. (2008) provided a more general approach for the TTP for multi-purpose trips, in which time spent in each activity contributes independently to the total travel time in the trip. Importantly, the impact of activity duration on travel time may in their formulation depend on socio-demographic or contextual factor, which is represented by interaction terms.

Our purpose here is to examine the trade-off between travel and activity time for secondary activities when adding non-work stops to commuting, therefore, we focus on the extra travel time added to commuting trips. The travel time attributed to non-work stops (T_non-work) is computed by deducting the travel time for the direct commute trip (T_direct) from the total travel time for multi-purpose commute trips (T_total). As mentioned above, compared to travel time calculated on road network, actual travel time can be affected by more complicated traffic conditions (e.g. congestions) and also individuals’ choices (e.g. departure in the rush hour, switching to a different travel mode). When using actual travel time recorded for direct and indirect trips in practice, it is possible for this additional travel time to be negative values. We assume that T_non-work is related to the duration of the non-work stops as follows:

In this expression, \(D_{j}\) is the duration of activity type j, \(C\) is the choice set for activity types, and the variable \(X_{i}\) represents personal, trip and spatial characteristics. As travel time could be affected by the personal and household, trip and spatial attributes (e.g. Schwanen et al. 2002; Cao et al. 2009; Tarigan et al. 2012), this direct impact could also apply to the extra travel time in multi-purpose trips, and the direct impacts of these attributes (\(X_{i}\)) will be estimated in our analysis. In line with the TTR/TTP literature, we assume also that extra travel time may be influenced by the duration of the secondary activity of type j (\(D_{j}\)). Finally, we assume, in line with Soo et al. (2008) that the effect of duration (\(D_{j}\)) may be modified by personal, household or spatial attributes (Xi), expressed in the interaction term \(X_{i} *D_{j}\). Because for each multi-purpose commute trip we know the duration of the non-work stops \(D_{j}\), the model can be estimated as a regression model with \(X_{i}\), \(D_{j}\) and the interaction \(X_{i} D_{j}\) as explanatory variables. By including explanatory variables related to the recorded actual trips (i.e., timing, modal shift), this model structure is flexible to analyze T_non-work values whether they positive or negative.

Our model has some specific innovative characteristics. In previous studies concerning the trade-off between travel and activity time, spatial variables discussed include the density of inhabitants (Schwanen and Dijst 2002; Soo et al. 2008), density of retail or service facilities (Soo et al. 2008; Susilo and Dijst 2009), and location or urbanization level (Schwanen and Dijst 2002; Susilo and Dijst 2009). These studies show that spatial variation may offer a better explanation than socio-demographic variables for the TTR value of work activities; for example, higher density and residing in suburbs tend to result in a higher TTR. However, only spatial attributes in the vicinity of the place of residence have been included. Although Schwanen and Dijst (2002)’s descriptive analysis confirmed that TTR varies with the location of the workplace, this indicator has not been included in multivariable analyses. Because commute trips are anchored by two bases rather than the place of residence, the spatial attributes in this study will include both the anchors.

Another important addition to previous studies here is allowing for mode shifts between the direct trips and the trip chaining behavior. Thus far, mode choice has been regarded as an invariant decision in studies concerning the relation between mode choice and trip chaining behavior (Strathman et al. 1994; Walle and Steenberghen 2006; Ye et al. 2007). This is because their data are mostly based on one-day observation rather than multi-day, and the difference in mode choice has been explored mostly from an inter-personal rather than an intra-personal perspective. A few studies have shed light on the intra-personal variation of trip chaining behavior, such as Bhat (1999), in which the unobserved heterogeneity for non-work stop-making in evening commuting with multi-day data were accommodated. Despite the reorganization of the day-to-day variation of stop-making in this study, the work-related explanatory variables (work duration, departure time, travel time for direct trip, mode choice) were still regarded as pre-determined. Moreover, a very high fraction of individuals in their sample used the car mode and had little penetration for other modes; consequently, the mode choice showed no significant impact on the variation of stop-making. However, given their desired activities and time–space opportunities, some population categories have choice opportunities for their travel behavior (Dijst et al. 2002) or may need to negotiate the allocation of vehicles on a day-to-day basis. As a result, there is an opportunity to use different travel modes for a direct trip and a multi-purpose trip, which may significantly impact the TTP as expressed in Eq. (2). For example, by shifting from a slower (e.g., bicycle) to a faster mode (e.g., car), one may be able to minimize extra travel time to conduct the secondary activity or even save time compared with the direct trip. Additionally, by shifting (part of) the multi-purpose trip in time, travel time might be reduced. For example, by engaging in a secondary activity after work, the trip from the secondary activity site to the place of residence may be postponed until after the peak period, leading to an overall shorter travel time. Therefore, this paper aims to include these strategies into the analysis of the travel time invested in non-work stops in commuting to obtain greater insight into the trade-offs and outcomes of trip chaining behavior in such a rapidly developing city as Beijing.

The data come from the “Daily Activity and Travel Survey of Beijing 2012”. Since public transport dominates in the city of Beijing, on the one hand, the lack of flexible transport modes (i.e. private car) may make against the complex tour making behavior of residents (Yang et al. 2007). On the other hand, the traditionally high density, mixed land use pattern due to the heritage of Work-unit (Danwei) can encourage complex tour patterns. For example, Ma et al. (2014)’s study in Beijing found that higher density in residential neighborhood leads to more home-based tours with fewer stops, while mixed land use at workplace with higher density and accessibility leads to more stops within one work tour or a more complex tour pattern.

This survey was conducted from October to December 2012 in the Shangdi–Qinghe area, which is located northeast of Beijing’s 5th ring road. This suburban area serves as a residential center, with 240,000 residents, and also as a job sub-center, with more than 5000 companies and 160,000 jobs (Fig. 1). The survey first collected the socio-demographic characteristics of the respondents and then activity diaries filled out by the respondents for one week and GPS data from GPS loggers carried by the respondents. The complete dataset contains responses from 709 respondents. Based on the research aim, the sample for the current analysis comprises 275 multi-purpose commute trips conducted by 178 commuters, who reported both direct and multi-purpose commute trips in the diaries. Sixty-five commuters contributed more than one multi-purpose commute trip; thus, the number of multi-purpose trips is larger than the number of commuters and is not evenly distributed among the commuters.

Table 1 provides a list of the explanatory variables used in the regression model, their definitions, and associated descriptive statistics. For these non-work stops in multi-purpose commuting trips, we distinguished 4 types of secondary activities of the commuters including eating out, shopping, recreation/social activities, and family/personal/other activities as secondary activities. Of the multi-purpose trips, 52.9% are combined with eating out activity, 28.6% with leisure or social activity, 20.3% with shopping activity and 14.1% with family/personal/other activity. 15.6% of the trips (43 cases) have included stops for more than one activity types. Furthermore, the duration for each activity has been extracted from travelers’ activity dairies. Particularly, the standard deviation for family/personal and other activities is large at 62.7 min, which is likely related to its diversity because it includes family activities, such as picking up or dropping off family members, and personal affairs, such as seeing doctors and visiting hairdressers.

Of the multiple-purpose commute trips, 78.0% (215) are trips from work and the other 60 trips are to work. The mode share in the sample is as follows: 43.3% car, 42.9% public transit and 13.8% active modes (i.e., foot and bicycle). For all the commuting trips in the complete dataset, the largest proportion belongs to public transit trips (51.0%), and the proportion for car trips is much lower at only 21.8%. The difference in mode shares is consistent with the previous finding that multi-purpose trips are more car-oriented (Bhat 1997; Islam and Habib 2012).

Commute duration (for direct commute) as a constraint-related variable was subtracted from the diary data reported. For those commuters who had reported more than one direct commute trip, the value was computed by averaging the trip durations with the most frequently used mode. As for spatial factors, Points of Interest (POIs) data in Beijing were used to assess the density and mix level in 1 km around the two bases. In this paper, POIs density stands for the density level, and the mix level is indicated by the diversity of POIs as shown in Eq. (3) (Frank et al. 2004; Long and Liu 2013):

In the equation, \(p_{i}\) is the proportion of the POIs of type I; n denotes the number of POI types under consideration. The POI types (n = 4) here include commercial, public service, public recreation, and transport facilities.

This research focused on the non-work stops in commuting trips by using the dataset of the “Daily Activity and Travel Survey of Beijing 2012”. Compared with previous studies, which were mostly based on single-purpose trips, this study shows that chaining a stop to commuting trips reduces the travel time expenditure for non-work activities for commuters, which can benefit urban residents with tight time-budgets. The extra travel time for non-work stops and the trade-off between travel and activity time for non-work stops has been examined, and it varies among the different types of non-work stops.

For policy making, lowering the extra travel time for non-work stops can be a way to increase commuters’ utility and at the same time release the pressure imposed on the urban transport system by the extra travel time for those stops. Of the spatial factors that can be influenced by policies, mix level is more important than density, and the workplace plays a more important role than the residential environment for commuters according to the model estimation results. Therefore, a possible implication of this finding for planning practice is that raising the diversity rather than only the density of facilities will better fulfill commuters’ needs. The workplace merits particular attention in this respect.

Generally, longer work duration reduces the time budget for commuters, and consequently decreases their travel time for non-work stops, regardless of activity types. However, the impacts of socio-demographics and trip-related attributes vary across the activity types. For example, gender difference is found only for eating out, which suggests that male commuters are inclined to travel longer for the same amount of activity time. As a constraint-related variable, timing (i.e., chaining a non-work stop to a work-home journey) negatively influences the travel time attributed to shopping and family/personal/other activities as the larger time-window after work is used for a for longer activity duration. Commute duration (for direct trips) negatively influences extra travel time for eating-out, shopping and family/personal/other activities, because it results in a tighter time-budget for non-work activities.

Specifically, the descriptive analysis shows that almost 20% of the multi-purpose commute trips adopted a more motorized transport mode (compared with their direct counterparts) to adapt to the more complex agendas. Although the regression results show that active mode users and car users have lower travel time for each time unit of eating out and leisure activities, respectively, the variable of mode shift does not help reduce the proportional travel time for non-work activities. Thus, the hypothesis of shorter travel time due to reduced travel time and increased activity duration is rejected. Given a higher level of mobility and relaxed time constraints, the space–time prism for commuters is expanded with a larger choice set of non-work activities (Chen and Kwan 2012), and commuters may then relocate their non-work activity.

While our study is in line with previous studies on TTR (Soo et al. 2008; Susilo and Dijst 2009, 2010), it also shares with these studies the limitation that certain choice dimensions are regarded as exogenous, such as destination of the secondary destination, mode choice and the decision to engage in a secondary activity in the first place. While our study provides insight in the factors influencing the travel time in multi-purpose commute trips, it does not account for the fact that commuters may change mode, destination or activity participation in response to the explanatory factors used in this study. Adding these dimensions would be important if ones goal is to develop comprehensive forecasting models, and is seen as an important future activity.

It should be noted that this research has considered extra travel time rather than travel distance for non-work stops. Measuring travel expenditure in distance could be a useful addition to examine the assumption that individuals will locate their non-work stops farther when they have a higher mobility level and, consequently, a larger choice set. Another relevant future analysis may concern other possible choices for commuters, such as the adjustment of departure/arrival times, which has been proved to affect the stop-making propensity (Bhat 1999) and which may affect the travel times and travel time prices for chained activities.

Additionally, the influence of spatial factors in the current study focuses on the two anchors of commute trips, namely the place of residence and the workplace. However, a previous study based on a GPS dataset from 34 drivers in southeastern Michigan described the spatial configuration of non-work activities and found that non-work activities chained to home-to-work commutes are mainly located along commuting paths (Wang et al. 2013). Future explorations can take the attributes of home-to-work corridors into consideration rather than the two ends alone. Furthermore, the built environment in Chinese cities is deeply shaped by the institutional and economic transition. Variables such as community type, which has been widely discussed in previous studies concerning the built environment and travel behavior in China (Ma et al. 2014; Zhao and Chai 2013), could be taken into account in a future study. Lastly, additional research is required to examine whether the mode shift phenomenon is common for multi-purpose trips and whether its impact on travel time is similar in other contexts with different distributions of transport mode use.