Impacts of a multi-scale built environment and its corresponding moderating effects on commute duration in China

Abstract

Chinese

Many studies have found that the built environment affects commute duration. However, they have paid little attention to the moderating effects of the surrounding built environment, and few have focused on the built environment at different levels simultaneously. Based on a sample of 3453 individuals across China in 2014, our study examined the impacts of the built environment at both the neighbourhood and city levels on commute duration as well as the moderating effects of the city-level built environment on the neighbourhood-level built environment using a multilevel linear regression model. The results show that built environment elements at both levels affect commute duration: city population size, population density at both levels, and the quadratic term for population density at the city level are positively associated with commute duration, while jobs–housing balance and short distance to business centres and transit stations are negatively related. City population size can strengthen the time-shortening effects of the jobs–housing balance and of proximity to transit stations on commute duration. City population density decreases the time-shortening effect of business centre proximity. This study has important implications for future research and policies regarding reducing traffic congestion and commute duration in Chinese cities.

Keywords

built environment China commute time heterogeneity effect multilevel model

Introduction

Auto-oriented suburbanisation and urban expansion have led to an increase in traffic congestion as well as commute duration, because high congestion forces people to spend more time in their vehicles (Barnes, 2001). For example, the average commute duration in Beijing increased from 38 minutes in 2005 to 43.6 minutes in 2010 (Meng et al., 2011). Lengthened commute duration not only wastes people’s time but also lowers their subjective wellbeing and health (Stutzer and Frey, 2008). Therefore, reducing traffic congestion and commute duration has become a common goal for transportation researchers and urban planners.

Much of the existing literature has identified that a built environment with higher density, diversity, walkability, and destination and transit accessibility could reduce commute duration (van Acker and Witlox, 2011; Zhao, 2013). However, few have considered the moderating effects of built environment factors on commute duration, although Heres and Niemeier (2017) pointed out that the built environment affects travel behaviour not only directly but also indirectly, by influencing other attributes. Moreover, few studies on the relationship between the built environment and commute duration have considered the built environment at both neighbourhood and city levels (Schwanen et al., 2004; Zhu et al., 2017). The commute destinations of workers usually are far beyond their neighbourhood areas. Thus, the built environment at both the neighbourhood and city levels affects their commute durations.

This study contributes to the literature by examining the moderating effects among built environment elements while considering both neighbourhood-scale and city-scale built environments. The findings provide evidence that supports the idea that built environment elements at both the city and neighbourhood levels matter for commute duration, and city-level built environment elements strengthen or weaken the association between neighbourhood built environment elements and commute duration.

In the next section, we review literature regarding the relationships between the built environment and commute duration, identify research gaps, and then present our theoretical framework. The subsequent section describes the data and methodology, followed by results and discussions. We summarise the key findings and discuss their implications in the last section.

Literature review and theoretical framework

Built environmental impacts on commute duration

During the past 30 years, a growing body of literature has examined the relationship between the built environment and commute duration. We describe built environment elements in terms of the ‘five Ds’– the variables of population size and density; land use diversity and jobs–housing balance; road design; destination accessibility; and distance to transit – following previous studies (Cervero and Kockelman, 1997; Ewing and Cervero, 2001, 2010; Handy et al., 2002), and review the link between them and commute duration.

Population size and density

Several studies support the positive relationship between city size and population, and commute duration (Izraeli and McCarthy, 1985; Lee et al., 2009; Schwanen et al., 2004), controlling for all other related built environment variables including density. The primary reason is, as cities grow, people’s choice of workplace is not limited by what is available nearby but by what is available in the whole city, thus commuting distances and duration increase (Yang et al., 2012).

The literature on the relationship between population density and commute duration is vast but messy. On the one hand, higher population density could increase residents’ likelihood of proximity to jobs, resulting in reduced commute distance and duration (van Acker and Witlox, 2011; Zhao, 2013). On the other hand, higher population density may contribute to crowding and congestion (Melia et al., 2011; Sarzynski et al., 2006) and thus, a longer commute duration (Dai et al., 2016; Schwanen et al., 2003). Some studies have even found a non-linear (U-type) relationship between population density and commute duration, suggesting that residents living in both low-density and high-density zones have a longer commute duration (Antipova et al., 2011; Levinson and Kumar, 1997). Yang et al. (2012) have also found that the effect of population density on commute duration is insignificant because the positive and negative effects cancel each other out.

Land use diversity and jobs–housing balance

Some studies have found that mixed land use contributes to reducing commute time because it may allow residents to live closer to their destinations (Antipova et al., 2011; Cervero and Duncan, 2006). Jobs–housing balance also shows a negative association with commute duration because the commute distance and time are relatively shorter if the residential and work locations are close (Dai et al., 2016; Wang and Chai, 2009; Zhao, 2013).

In the aggregated studies, many scholars also found the imbalanced jobs–housing spatial relationship leads to ‘wasteful (excess) commuting’ (Hamilton, 1982; Kanaroglou et al., 2015; Ma and Banister, 2006; Small and Song, 1992; White, 1988). Some studies also found housing–job proximity is one of the most important elements to reduce congestion (Cervero, 1989, 1996) and commute duration (Sarzynski et al., 2006). However, other aggregated studies found the jobs–housing balance has a limited effect on commuting, because the reasons for residential location choice are various and most people do not treat the work location as an important factor (Cervero, 1996; Giuliano, 1991; Giuliano and Small, 1993).

Road design

Some studies have found denser and finer street layout was associated with higher congestion levels and longer commute duration (Ewing et al., 2003; Zhao, 2013). Zhu et al. (2017) also found that the per capita area of a city’s paved roads are negatively related to commute duration.

Destination accessibility

The literature shows that commute duration increases when residence is located far away from the city centre, leading to an obvious lengthened commute distance (Cervero and Day, 2008; Yang, 2006). Some studies have also found that job accessibility has a negative correlation with commuting duration (Dai et al., 2016; van Acker and Witlox, 2011).

Distance to transit

Shorter distance to transit facilities (or higher transit accessibility) has been found to be negatively related to commute duration (Dai et al., 2016; Zhao and Li, 2016). However, van Acker and Witlox (2011) found that the distance to railway stations could decrease commuting duration while the distance to bus stops could increase it. Zhao, Lü and de Roo (2011) also found that the effect of transit accessibility on commute duration is weaker than that of other factors.

The moderation effects of built environment

The impact of one built environment on travel behaviour extends beyond the direct impact from itself; other factors in the built environment might interact with it to create moderating effects on travel (Heres and Niemeier, 2017). The negative effect of density at the 1/4-mile radius on non-work auto travel frequency is increased by higher network load density (Chatman, 2008). The metro rapid stop with a dedicated-lane service would strengthen the increasing effect of total density (population and employment densities) on transit ridership (Cervero et al., 2010).

The multi-context built environment’s effect on commute

As well as the built environment in residential neighbourhoods, the built environment at destinations and around the route area also have effects on the commute. Compact workplaces promote active commute behaviour with slow speeds, such as walking, biking, transit, etc. (Ding et al., 2014; Zhang, 2004), and reduce driving (Chen et al., 2008; Zhang, 2004). However, Sun et al. (2017) found the workplace built environment is less influential on commute than the residence environment in Shanghai. They attributed it to the fact that the home locations of residents are more stable than their job locations, and residential location choice is highly correlated with built environment characteristics in Shanghai. Moreover, the built environment around the route area also affects travel behaviour (Kwan, 2012a). Some studies also found that large-scale built environment has significant effects on travel behaviour (Yang et al., 2012; Yin and Sun, 2018).

Research gaps and theoretical framework

In our review of the literature, we found that some topics concerning the relationship between commuting and urban environments have been neglected or could be expanded upon, which prompted us to conduct this research.

First, exploring the moderating effects between different built environment factors would promote a more comprehensive understanding of the impact of built environment on commute duration (Wang et al., 2011). Several studies examined ways that one built environment element might moderate the effect of another built environment element on travel behaviour, but few efforts have been made to disentangle the built environment’s moderating effect on commute duration, at least in developing countries such as China.

Second, many studies have identified the association between the built environment and commute duration, but far fewer have simultaneously considered the built environment at different levels, particularly hierarchical built environment elements. Social ecological models show people’s behaviours are affected by multi-level environmental factors (Sallis et al., 2008). Kwan (2012b) also pointed out that the uncertain geographic context problem (UGCoP), which refers to the spatial uncertainty in the actual areas that exert contextual influences on individuals, should not be ignored. Considering that commuters’ workplaces might be far beyond neighbourhood areas, the determinants of commute duration should not be limited to a single neighbourhood level (Sun and Yin, 2018). To some extent, the city-level built environment reflects the general features of the built environment at residences, workplaces and around route areas. Although many studies explored the effects of built environments at different scales on commute, they focused on modes choice or commute distance instead of commute duration. Moreover, most of them focused on built environment at both origins and destinations rather than the hierarchical scales. A very limited number of studies has examined the linkage between the built environment at different levels and commuting duration, but they have paid more attention to city size and density while ignoring the other ‘D’ variables. They have also paid less attention to the built environment’s non-linear relation with density (Schwanen et al., 2004; Zhu et al., 2017).

To fill these gaps, we propose a theoretical framework based on social ecological models, summarised in Figure 1, to guide this study, which explores the impacts of the built environment at both city and neighbourhood levels on commute duration and the moderating effects on the built environment at different levels. First, we hypothesise that the built environment at both city and neighbourhood levels affects commute duration (represented by the solid lines in Figure 1); their impacts are hierarchical and, specifically, residents live in neighbourhoods nested within cities. In theory, the built environment at the neighbourhood level should include both residence and workplace built environment elements. Moreover, the built environment at the city level could partly reflect the features of both residence and workplace built environment. Second, we hypothesise that the built environment affects commute duration not only directly but also through moderating effects of the city-level built environment on the link between the neighbourhood ones and commute duration (dotted lines).

Figure 1.

The theoretical framework of this study.

Data and methodology

Data source

The Center for Social Survey at Sun Yat-sen University has conducted labour force surveys since 2011. These surveys include individual socioeconomic attributes, commuting data and built environment elements at the neighbourhood level for residents in the mainland of China, including 29 provinces, autonomous regions and province-status city-regions. The most recent survey is the 2014 China Labor Force Dynamics Survey with a strict multi-stage cluster, stratified, PPS sampling method. A total of 3453 respondents from 148 neighbourhoods in 56 cities were included in our analysis, after excluding individuals that were of non-working age, unemployed or without a fixed workplace, or with missing values. The average sample size of neighbourhoods and cities is 24 and 62, respectively. As for the built environment elements at the city level, they were measured with data from the China City Statistical Yearbook and China Urban Construction Statistical Yearbook in 2014.

Model

Spatial autocorrelation occurs when observations at nearby locations tend to have similar characteristics, and hence the assumption of independence of observations is no longer met (Hong et al., 2014). In this paper, individuals’ residential environments are likely to be similar if they live in the same neighbourhood or city, so the inference of significance from the ordinary least squares model might be incorrect. Many researchers have applied the multilevel modelling framework to solve this problem because it can consider the spatial autocorrelation by estimating coefficients varying by group and allows researchers to differentiate heterogeneity existing among both individuals and groups (Hong et al., 2014; Schwanen et al., 2004; Zhu et al., 2017).

Using a multilevel linear regression model, we regressed commute duration on built environment at both city and neighbourhood levels and individual socioeconomic attributes. According to Leckie (2013), the empty model is as

C D_{ijk} = β_{0} + v_{k} + u_{jk} + e_{ijk}

(1)

where the outcome variable is commute duration (CD). The CD_ijk is the observed two-way commute duration for respondent i in neighbourhood j in city k, β₀ is the mean response across all cities, v_k is the effect of city k, u_jk is the effect of neighbourhood j within city k, and e_ijk is the residual error term.

We set our baseline model as follows,

\begin{array}{l} C D_{i j k} = β_{0} + β_{c i t y} B E_{c i t y, k} \\ + β_{n b h d} B E_{n b h d, j k} \\ + β_{S E} S E_{i j k} + v_{k} + u_{j k} + e_{i j k} \end{array}

(2)

where β₀+β_cityBE_city,k+β_nbhdBE_nbhd,jk+β_SESE_ijk is the fixed part of the model and v_k+u_jk+e_ijk is the random part of the model. BE_nbhd and BE_city are built environment variables at neighbourhood and city levels, respectively. In the fixed part of the model, BE_city,k, BE_nbhd,jk and SE_ijk are explanatory variables at levels 3, 2 and 1, respectively. SE is the variable of individual socioeconomic attributes. In addition, the workplace built environment might also affect commute duration. However, we do not include it in our model owing to data unavailability. To some extent, the built environment at the city level could partly reflect the features of the workplace built environment.

We do not consider some variables in the baseline model, such as commute price, mode choice, etc., because they are ‘bad control’ variables. The ‘bad control’ variable within the causal model is said to be the control variable that is determined by other independent variables, leading to dramatic misinterpretations of coefficient estimates (Angrist and Pischke, 2008). In our study, both travel price and mode choice are outcomes of built environment elements, so we should not control them, at least in the baseline model. However, we still controlled them in the robust model in order to examine whether the results of the baseline model are robust.

We tested for the moderating effects of the city-level built environment on the relationship between the neighbourhood-level built environment and commute duration using the random intercepts model. Based on the baseline model, the moderating effects model is written as

C D_{ijk} = β_{0} + β_{city} B E_{city, k} + β_{nbhd} B E_{nbhd, jk} + β_{m} (B {E_{city,}}_{k} \times B E_{nbhd, jk}) + β_{SE} S E_{ijk} + v_{k} + u_{jk} + e_{ijk}

(3)

where β_m is the coefficient of moderating effects. The multilevel linear regression was estimated with the software package Stata (version 13).

Variables

Table 1 shows the descriptive statistics of built environment variables. ‘City’ in this paper is defined as the city proper. Built environment elements at the city level include population size, population density, roads area and availability of metro. ‘Neighbourhood’ in this paper is defined as the area administered by a neighbourhood committee (juweihui), which is the smallest administrative unit in urban China (Luo et al., 2017). The area of a neighbourhood varies, and the median of neighbourhood areas is 1.28 km² in this study. The variables of the neighbourhood built environment include population density, diversity, jobs–housing balance and distance to the nearest business centre and transit station at the neighbourhood level. Diversity reflects the spatial balance among diversified destinations, measured by the sum of nine categories of facilities in the neighbourhood, including primary schools, middle schools, libraries or reading rooms, senior centres, playgrounds, squares or parks, hospitals, fitness centres and banks. The jobs–housing balance and distance to the nearest business centre and to the transit station are measured by individuals’ residential location.

Table 1.

Descriptive statistics of built environment variables (sample size = 3453).

Variable	Description	Mean	SD	Min	Max
City built environment
Population size (ln)	The total of city district population (10,000 persons)	5.51	1.00	3.44	7.49
Population density_C (ln)	Population per city proper area (person/km²)	6.97	0.75	4.08	8.54
Buses per 10,000 persons	Number of buses per 10,000 population (unit)	15.03	16.84	0.96	98.53
Road area per capital	Road area per person (m²)	14.88	8.81	3.06	54.37
Metro stations per capital	1 = city with metro; 0 = otherwise	0.07	0.11	0	0.42
Neighbourhood built environment
Population density_N (ln)	Population per neighbourhood area (person/km²)	8.45	2.14	1.45	12.55
Diversity	The sum of 9 categories destinations	5.64	1.69	1	9
Jobs–housing balance	1 = respondents’ workplace in the neighbourhood; 0 = otherwise	0.48	0.50	0	1
Business centre	1 = distance to the nearest business centre below 1000 m; 0 = otherwise	0.54	0.50	0	1
Transit station	1 = distance to the nearest transit station below 1000 m; 0 = otherwise	0.92	0.26	0	1

Table 2 shows the descriptive statistics of individual socioeconomic attributes. Like most previous studies, the control variables in the model include individuals’ gender, age, hukou, education, marriage, family size, household income and renting status.

Table 2.

Descriptive statistics of individual socioeconomic attributes.

Variable	Description	N	Mean	SD	Min	Max
Male	1 = male; 0 = female	3453	0.53	0.50	0	1
Age	Respondents’ age	3453	39.78	10.58	16	65
Hukou	1 = hukou in the local district; 0 = otherwise	3453	0.79	0.41	0	1
Education	Respondents’ education years	3453	12.21	3.44	0	16
Married	1 = married; 0 = otherwise	3453	0.86	0.35	0	1
Family size	The number of family members living together	3453	3.17	1.31	1	11
Household income (ln)	Household annual income (unit: 10,000 yuan)	3453	11.09	1.03	0	15.61
Renting	1 = renting a house; 0 = otherwise	3453	0.19	0.40	0	1
Commute price	Commute price (yuan)	3441	6.98	39.26	0	1000
Ebike ownership	The number of ebikes	3450	0.48	0.76	0	3
Car ownership	The number of cars	3448	0.35	0.56	0	3

Results and discussions

The results of the intraclass correlation coefficients

Table 3 shows the results of the multilevel linear model for commute duration. The intraclass correlation coefficients (ICCs) at the city and neighbourhood levels are 0.075 and 0.197, respectively, in the null model, indicating that commute duration within the same neighbourhood has a somewhat higher correlation. After controlling for confounding variables at the individual level (model 1), the ICCs are still more than 6% at both the neighbourhood and city levels. This suggests that respondents’ commute duration in the same city should not be ignored. The findings are in line with previous studies that indicate that the built environment at different levels affects commuting behaviour (Ewing and Cervero, 2010; Hong et al., 2014; Yin and Sun, 2018).

Table 3.

The results of multilevel linear models for commute duration.

	Null model	Model 1	Model 2	Model 3	Model 4	Model 5
City built environment
Population size			0.198^*** (0.054)		0.168^*** (0.043)	0.173^*** (0.043)
Population density_C			0.120^**(0.059)		0.121^*** (0.045)	0.116^*** (0.044)
Population density_C square			0.103^*** (0.035)		0.074^*** (0.027)	0.075^*** (0.027)
Buses per 10,000 persons			0.000 (0.004)		−0.001 (0.003)	−0.000 (0.003)
Road area per capital			0.001 (0.005)		0.003 (0.004)	0.003 (0.004)
Metro stations per capital			−0.264 (0.656)		−0.361 (0.489)	−0.401 (0.487)
Neighbourhood built environment
Population density_N				0.039^** (0.019)	0.030^** (0.014)	0.028^** (0.014)
Population density_N square				−0.000 (0.005)
Diversity				−0.020 (0.019)	−0.016 (0.017)	−0.013 (0.017)
Jobs–housing balance				−0.554^*** (0.033)	−0.557^*** (0.033)	−0.553^*** (0.033)
Business centre				−0.074^** (0.036)	−0.072^** (0.036)	−0.067 (0.036)
Transit station				−0.160^** (0.063)	−0.153^** (0.062)	−0.162^*** (0.062)
Socioeconomic attributes
Male		0.090^*** (0.031)	0.090^*** (0.031)	0.057 (0.030)	0.058^** (0.030)	0.049 (0.030)
Age		−0.003 (0.002)	−0.003 (0.002)	−0.001 (0.002)	−0.001 (0.002)	−0.001 (0.002)
Hukou		0.153^*** (0.048)	0.162^*** (0.048)	0.156^*** (0.046)	0.165^*** (0.046)	0.168^*** (0.046)
Education		0.027^*** (0.006)	0.026^*** (0.006)	0.019^*** (0.005)	0.019^*** (0.005)	0.018^*** (0.005)
Married		−0.060 (0.053)	−0.055 (0.053)	−0.036 (0.051)	−0.031 (0.051)	−0.032 (0.051)
Family size		0.006 (0.013)	0.007 (0.013)	−0.001 (0.012)	−0.000 (0.012)	0.002 (0.013)
Household income		0.019 (0.016)	0.016 (0.016)	0.017 (0.016)	0.013 (0.016)	0.015 (0.016)
Renting		−0.097^** (0.048)	−0.096^** (0.048)	−0.096^** (0.046)	−0.093^** (0.046)	−0.094^** (0.046)
Commute price						0.002^*** (0.000)
Ebike ownership						−0.012 (0.024)
Car ownership						−0.053 (0.030)
Constant	3.096^*** (0.051)	2.571^*** (0.208)	0.651 (0.514)	2.873^*** (0.290)	1.170^*** (0.432)	1.167^*** (0.433)
The random part
ln(sd(City_constant))	−1.317^*** (0.191)	−1.418^*** (0.203)	−2.105^*** (0.486)	−1.795^*** (0.272)	−10.848^*** (3.641)	−11.385^*** (3.554)
ln(sd(Neighbourhood_constant)	−1.067^*** (0.096)	−1.106^*** (0.097)	−1.117^*** (0.095)	−1.255^*** (0.103)	−1.268^*** (0.085)	−1.279^*** (0.092)
ln(sd(Residual)	−0.128^*** (0.012)	−0.140^*** (0.012)	−0.139^*** (0.012)	−0.176^*** (0.012)	−0.176^*** (0.015)	−0.179^*** (0.012)
Observations	3453	3453	3453	3453	3453	3436
LL	−4584.650	−4541.190	−4528.515	−4397.469	−4383.941	−4350.143
Chi square		88.443	127.494	406.275	478.906	520.530
AIC	9177.300	9106.379	9093.031	8830.938	8813.883	8752.286
ICC_City	0.075	0.063	0.017	0.034	0.000	0.000
ICC_ Neighbourhood\|City	0.197	0.182	0.139	0.134	0.101	0.100

Notes: Standard errors in parentheses; **p < 0.05; ***p < 0.01.

Built environment and commute duration at different levels

Models 2 and 3 show the results adjusted for built environment variables at the city and neighbourhood levels, respectively, while model 4 controlled for the two levels of the built environment simultaneously. We used the quadratic terms of population density at both city and neighbourhood levels to avoid potential non-linearity. However, the results show the quadratic term of population density at the neighbourhood level is insignificant (model 3); thus, we eliminated it from model 4. Most built environment variables were statistically significant, suggesting that the built environment elements at both the neighbourhood and city levels could impact commute duration. However, after controlling for both city and neighbourhood built environments (model 4), the coefficients of population size and density at both the city and neighbourhood levels observably decreased, suggesting that the estimates of built environment elements would be biased if the built environment at the city or neighbourhood level was omitted. Model 5 is the robust model, which includes the commute price and the number of vehicles. The results of model 5 are similar to model 4, except the business centre, which shows a negative association with commute duration marginally (p = 0.60). As expected, commute price is positive with commute duration.

The effects of each variable are discussed below based on model 4 in relation to the results of the baseline model.

Population size has a positive effect on commute duration, suggesting that living in a city with a larger population size is associated with a longer commute duration. Specifically, all other variables being equal, a 1% increase in population size is associated with 0.17% longer commute duration. Controlling for population density, the increase in population size is equivalent to the expansion of the city’s built areas, with increased commuting distance and duration. A study conducted by Schwanen et al. (2004) identified the positive relationship between urban area and disaggregate commute duration.

Population density at both the city and neighbourhood levels has a positive effect on commute duration. This finding is consistent with Schwanen et al. (2003), who suggested that the distance-shortening effect of population density is eclipsed by the congestion effect. This finding also supports Zhu et al. (2017), who found population density at the city level is a more important factor than that at the neighbourhood level.

Moreover, the quadratic term of city population density also shows a positive association with commute duration, indicating that the effects of population density at the city level on commute duration are consistently strengthened by the increase of city population density. More specifically, all else being equal, a 1% increase in population density at the city level is related to an increase in the commute duration of 0.12%, while a 1% increase in population density at the neighbourhood level is only related to an increase in the commute duration of about 0.03%. This is plausible because the higher the population density, the greater the travel demands of commuters and non-commuters, leading to traffic congestion during rush hours and a longer commute duration (Melia et al., 2011). In the Chinese context, many cities with extremely high population density and heavy traffic would experience increased transportation congestion and prolonged commute duration if city population density were to constantly increase.

As expected, jobs–housing balance negatively correlates with commute duration, suggesting that residents would have a shorter commute duration if they have jobs nearby. This is consistent with Cervero and Duncan (2006), who found that jobs–housing balance could shorten the commute distance and duration.

Business centres located within 1 km of a respondent’s residence are negatively related to commute duration, indicating that if residents’ homes are located near a business centre, commute duration could be reduced. This may be due to the fact that residents would have a higher likelihood of attaining a job if their residential location was close to the business centre, reducing commute distance and time (Zhao, 2013).

The presence of a transit station within 1 km of a person’s residence also has negative effects on commute duration, implying that a resident’s home being close to a transit station would reduce commute time, in accordance with Zhao and Li (2016). The reason might be that a short distance to a transit station shortens walking time, thus shortening the overall commute duration.

Moderating effects of the built environment

We put the terms indicating the interaction between the significant built environment variables from the baseline model (model 4) into the moderating effects models to determine the moderating effects of the built environment (Table 4). Overall, city-level built environment elements could moderate the impacts of neighbourhood-level built environment elements on commute duration.

Table 4.

The results of the moderating effects models for commute duration.

	Model 6	Model 7	Model 8
City built environment
Population size (ln)	0.087^* (0.049)	0.158^*** (0.047)	0.165^*** (0.046)
Population density_C (ln)	0.116^** (0.049)	0.124^** (0.050)	0.174^*** (0.052)
Population density square_C (ln)	0.089^*** (0.029)	0.080^*** (0.030)	0.075^*** (0.029)
Buses per 10,000 persons	0.001 (0.003)	−0.000 (0.003)	−0.000 (0.003)
Road area per capital	0.003 (0.004)	0.003 (0.005)	0.002 (0.004)
Metro stations per capital	−0.515 (0.540)	−0.392 (0.552)	−0.393 (0.536)
Neighbourhood built environment
Population density_N	0.030^** (0.014)	0.030^** (0.014)	0.028^** (0.014)
Diversity	−0.010 (0.018)	−0.015 (0.018)	−0.014 (0.018)
Jobs–housing balance	−0.550^*** (0.033)	−0.555^*** (0.033)	−0.553^*** (0.033)
Business centre	−0.074^** (0.036)	−0.073^** (0.036)	−0.062 (0.036)
Transit station	−0.158^** (0.062)	−0.193^*** (0.065)	−0.148^** (0.063)
Moderating effect
Population size × jobs–housing balance	−0.167^*** (0.034)
Population size × transit station		−0.128^** (0.060)
population density_C× business centre			0.120^*** (0.044)
Constant	1.615^*** (0.468)	1.244^*** (0.469)	0.828 (0.476)
Individual socioeconomic attributes	Controlled	Controlled	Controlled
The random part
ln(sd(City_constant))	−2.590^** (1.023)	−2.535^*** (0.909)	−2.654^** (1.095)
ln(sd(neighbourhood_constant)	−1.261^*** (0.104)	−1.240^*** (0.101)	−1.257^*** (0.102)
ln(sd(Residual)	−0.178^*** (0.012)	−0.175^*** (0.012)	−0.175^*** (0.012)
Observations	3453	3453	3453
LL	−4432.933	−4442.382	−4441.376
Chi²	486.110	459.376	467.843
AIC	8913.867	8932.765	8930.753

Notes: Standard errors in parentheses; **p < 0.05; ***p < 0.01.

Population size could strengthen the negative effects of the jobs–housing balance and transit station proximity on commute duration. This suggests that in a larger city, that is, a more congested city as the coefficient of city population size shows, jobs–housing balance and transit station proximity could reduce commute duration.

City population density could weaken the diminishing effects of a nearby business centre on commute duration. It could be attributed to the Chinese context wherein cities with higher population density stimulate more commute demands, especially in the neighbourhoods close to the city’s business centre, and lead to traffic congestion and a longer commute duration (Zhao and Li, 2016).

Conclusion

Previous studies on the impacts of the built environment on commute duration may be biased because they disregard the moderating effects among built environment attributes and the simultaneous impacts of the built environment at both neighbourhood and city levels. To fill these gaps, we examined the relationship between both levels of the built environment and commuting duration, as well as the moderating effects among different levels of the built environment based on a disaggregated sample from China. After controlling for individual socioeconomic attributes, the empirical results suggest that the built environment at both neighbourhood and city levels affects commute duration significantly. Specifically, residents will have a longer commute duration if they live in cities with higher population size and density or in neighbourhoods with higher population density. Conversely, residents will have a shorter commute duration if their workplaces are in the same neighbourhood as their residences or if they live closer to a business centre or transit station.

Our findings also indicate that the effects of the built environment at the neighbourhood level could be moderated by city-level built environment factors. Specifically, the time-shortening effects of jobs–housing balance and proximity to a transit station on commute duration are strengthened by city population size. However, the time-shortening effect of proximity to a business centre is inhibited by city population density. Moreover, the impact of higher population density at the city level on commute duration is non-linear. That is, as the city population density increases, residents’ commute duration increases quadratically.

Constrained by data availability, there are some limitations in this study. First, we only analyse the built environment at both neighbourhood and city scales, ignoring the workplace built environment. Although the workplace built environment might be less influential on commute than the residence built environment in China (Sun et al., 2017), it should be considered in the framework of social ecological models in the future. Second, although most related built environment elements are covered in this study, some variables are crude or omitted, such as ‘design’ at the neighbourhood level and jobs–housing balance at the city level. In the future, if data are available, controlling more built environment elements could allow more accurate estimation of the net effects of each built environment variable. These limitations leave room for improvement by future studies. However, as one of the earliest studies focusing on the impacts of the built environment at different levels on commute duration and the moderating effects among built environmental factors, this research contributes to the literature by providing empirical evidence from China.

It also has important policy implications for reducing travel duration. Urban planners should pay greater attention to the impacts of the built environment at both the city and neighbourhood levels on commute duration and try to mitigate the negative effects caused by overly compact development patterns. Specifically, maintaining a reasonable city size and population density and promoting jobs–housing balance and destination accessibility help shorten commute duration. At the same time, urban planners should also take notice of the interaction effects among built environment factors on commute duration. Thus, we should not focus on the single built environment element because of the interaction among the elements of the built environment. For example, promoting a reasonably larger city population size would augment the time-shortening effects of proximity to the workplace or transit station on commute duration.

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work is financially supported by the Major Program of the National Social Science Foundation of China, No. 17ZDA068.

ORCID iD

Chun Yin

References

Angrist

Pischke

J-S

(2008) Mostly Harmless Econometrics: An Empiricist’s Companion. Massachusetts Institute of Technology and The London School of Economics. Princeton, NJ: Princeton University Press.

Antipova

Wang

Wilmot

(2011) Urban land uses, socio-demographic attributes and commuting: A multilevel modeling approach. Applied Geography 31(3): 1010–1018.

Barnes

(2001) Population and employment density and travel behavior in large U.S. cities. Final report MN/RC – 2001-24. University of Minnesota, Minneapolis, MN, September.

Cervero

(1989) Jobs–housing balancing and regional mobility. Journal of the American Planning Association 55(2): 136–150.

Cervero

(1996) Jobs–housing balance revisited: Trends and impacts in the San Francisco Bay Area. Journal of the American Planning Association 62(4): 492–511.

Cervero

Day

(2008) Suburbanization and transit-oriented development in China. Transport Policy 15(5): 315–323.

Cervero

Duncan

(2006) Which reduces vehicle travel more: Jobs–housing balance or retail–housing mixing? Journal of the American Planning Association 72(4): 475–490.

Cervero

Kockelman

(1997) Travel demand and the 3Ds: Density, diversity, and design. Transportation Research Part D: Transport and Environment 2(3): 199–219.

Cervero

Murakami

Miller

(2010) Direct ridership model of bus rapid transit in Los Angeles County, California. Transportation Research Record: Journal of the Transportation Research Board 2145(1): 1–7.

10.

Chatman

(2008) Deconstructing development density: Quality, quantity and price effects on household non-work travel. Transportation Research Part A: Policy and Practice 42(7): 1008–1030.

11.

Chen

Gong

Paaswell

(2008) Role of the built environment on mode choice decisions: Additional evidence on the impact of density. Transportation 35(3): 285–299.

12.

Dai

Zhou

(2016) Spatial-temporal characteristics and factors influencing commuting activities of middle-class residents in Guangzhou City, China. Chinese Geographical Science 26(3): 410–428.

13.

Ding

Lin

Liu

(2014) Exploring the influence of built environment on tour-based commuter mode choice: A cross-classified multilevel modeling approach. Transportation Research Part D: Transport and Environment 32: 230–238.

14.

Ewing

Cervero

(2001) Travel and the built environment: A synthesis. Transportation Research Record: Journal of the Transportation Research Board 1780(1): 87–114.

15.

Ewing

Cervero

(2010) Travel and the built environment. Journal of the American Planning Association 76(3): 265–294.

16.

Ewing

Pendall

Chen

(2003) Measuring sprawl and its transportation impacts. Transportation Research Record: Journal of the Transportation Research Board 1831(1): 175–183.

17.

Giuliano

(1991) Is jobs–housing balance a transportation issue? Transportation Research Record: Journal of the Transportation Research Board 1305: 305–312.

18.

Giuliano

Small

(1993) Is the journey to work explained by urban structure? Urban Studies 30(9): 1485–1500.

19.

Hamilton

(1982) Wasteful commuting. Journal of Political Economy 90(5): 1035–1053.

20.

Handy

Boarnet

Ewing

, et al. (2002) How the built environment affects physical activity: Views from urban planning. American Journal of Preventive Medicine 23(2): 64–73.

21.

Heres

Niemeier

(2017) The past and future of research on the link between compact development and driving: Comment on ‘Does compact development make people drive less?’ Journal of the American Planning Association 83(2): 145–148.

22.

Hong

Shen

Zhang

(2014) How do built-environment factors affect travel behavior? A spatial analysis at different geographic scales. Transportation 41(3): 419–440.

23.

Izraeli

McCarthy

(1985) Variations in travel distance, time and modal choice among SMSAs. Journal of Transport Economics and Policy 19(2): 139–160.

24.

Kanaroglou

Higgins

Chowdhury

(2015) Excess commuting: A critical review and comparative analysis of concepts, indices, and policy implications. Journal of Transport Geography 44: 13–23.

25.

Kwan

(2012a) How GIS can help address the uncertain geographic context problem in social science research. Annals of GIS 18(4): 245–255.

26.

Kwan

(2012b) The uncertain geographic context problem. Annals of the Association of American Geographers 102(5): 958–968.

27.

Leckie

(2013) Three-Level Multilevel Models – Concepts. LEMMA VLE Module 11, 1–47. Available at: http://www.bristol.ac.uk/cmm/learning/course.html (accessed 22 July 2018).

28.

Lee

Gordon

Richardson

, et al. (2009) Commuting trends in U.S. cities in the 1990s. Journal of Planning Education and Research 29(1): 78–89.

29.

Levinson

Kumar

(1997) Density and the journey to work. Growth and Change 28(2): 147–172.

30.

Luo

Tian

Luo

, et al. (2017) Two-step optimization for spatial accessibility improvement: A case study of health care planning in rural China. BioMed Research International 2017: 1–12.

31.

K-R

Banister

(2006) Excess commuting: A critical review. Transport Reviews 26(6): 749–767.

32.

Melia

Parkhurst

Barton

(2011) The paradox of intensification. Transport Policy 18(1): 46–52.

33.

Meng

Zheng

(2011) Commuting time change and its influencing factors in Beijing. Progress in Geography 30(10): 1218–1224.

34.

Sallis

Owen

Fisher

(2008) Ecological models of health behaviour. In: Glanz

Rimer

Viswanath

(eds) Health Behaviour and Health Education: Theory, Research, and Practice. San Francisco, CA: Jossey-Bass, pp. 465–485.

35.

Sarzynski

Wolman

Galster

, et al. (2006) Testing the conventional wisdom about land use and traffic congestion: The more we sprawl, the less we move? Urban Studies 43(3): 601–626.

36.

Schwanen

Dieleman

Dijst

(2003) Car use in Netherlands daily urban systems: Does polycentrism result in lower commute times? Urban Geography 24(5): 410–430.

37.

Schwanen

Dieleman

Dijst

(2004) The impact of metropolitan structure on commute behavior in the Netherlands: A multilevel approach. Growth and Change 35(3): 304–333.

38.

Small

Song

(1992) ‘Wasteful’ commuting: A resolution. Journal of Political Economy 100(4): 888–898.

39.

Stutzer

Frey

(2008) Stress that doesn’t pay: The commuting paradox. Scandinavian Journal of Economics 110(151): 336–366.

40.

Sun

Yin

(2018) Relationship between multi-scale urban built environments and body mass index: A study of China. Applied Geography 94: 230–240.

41.

Sun

Ermagun

Dan

(2017) Built environmental impacts on commuting mode choice and distance: Evidence from Shanghai. Transportation Research Part D: Transport and Environment 52: 441–453.

42.

van Acker

Witlox

(2011) Commuting trips within tours: How is commuting related to land use? Transportation 38(3): 465–486.

43.

Wang

Chai

(2009) The jobs–housing relationship and commuting in Beijing, China: The legacy of Danwei. Journal of Transport Geography 17(1): 30–38.

44.

Wang

Chai

(2011) Built environment diversities and activity–travel behaviour variations in Beijing, China. Journal of Transport Geography 19(6): 1173–1186.

45.

White

(1988) Urban commuting journeys are not ‘wasteful’. Journal of Political Economy 96(5): 1097–1110.

46.

Yang

(2006) Transportation implications of land development in a transitional economy: Evidence from housing relocation in Beijing. Transportation Research Record: Journal of the Transportation Research Board 1954: 7–14.

47.

Yang

French

Holt

, et al. (2012) Measuring the structure of U.S. metropolitan areas, 1970–2000. Journal of the American Planning Association 78(2): 197–209.

48.

Yin

Sun

(2018) Disentangling the effects of the built environment on car ownership: A multi-level analysis of Chinese cities. Cities 74: 188–195.

49.

Zhang

(2004) The role of land use in travel mode choice: Evidence from Boston and Hong Kong. Journal of the American Planning Association 70(3): 344–360.

50.

Zhao

(2013) The impact of the built environment on individual workers’ commuting behavior in Beijing. International Journal of Sustainable Transportation 7(5): 389–415.

51.

Zhao

(2016) Restraining transport inequality in growing cities: Can spatial planning play a role? International Journal of Sustainable Transportation 10(10): 947–959.

52.

Zhao

Lü

de Roo

(2011) Impact of the jobs–housing balance on urban commuting in Beijing in the transformation era. Journal of Transport Geography 19(1): 59–69.

53.

Zhu

Liu

, et al. (2017) The impact of urban characteristics and residents’ income on commuting in China. Transportation Research Part D: Transport and Environment 57: 474–483.