You are on page 1of 15

Transportation Research Part D 52 (2017) 480–494

Contents lists available at ScienceDirect

Transportation Research Part D


journal homepage: www.elsevier.com/locate/trd

Examining the effects of the built environment and residential


self-selection on commuting trips and the related CO2
emissions: An empirical study in Guangzhou, China
Xiaoshu Cao a,b,⇑, Wenyue Yang c
a
School of Geography Science and Planning, Sun Yat-sen University, Guangzhou 510275, China
b
Institute of Transport Geography and Spatial Planning, Shaanxi Normal University, Xi’an 710119, China
c
College of Forestry and Landscape Architecture, South China Agricultural University, Guangzhou 510642, China

a r t i c l e i n f o a b s t r a c t

Article history: Numerous studies have established the link between the built environment and travel
Received 7 December 2015 behavior. However, fewer studies have focused on environmental costs of travel (such as
Revised 16 January 2017 CO2 emissions) with respect to residential self-selection. Combined with the application
Accepted 7 February 2017
of TIQS (Travel Intelligent Query System), this study develops a structural equations model
Available online 1 March 2017
(SEM) to examine the effects of the built environment and residential self-selection on
commuting trips and their related CO2 emissions using data from 2015 in Guangzhou,
Keywords:
China. The results demonstrate that the effect of residential self-selection also exists in
Commuting trip
CO2 emissions
Chinese cities, influencing residents’ choice of living environments and ultimately affecting
Built environment their commute trip CO2 emissions. After controlling for the effect of residential self-
Residential self-selection selection, built environment variables still have significant effects on CO2 emissions from
structural equations model (SEM) commuting although some are indirect effects that work through mediating variables
China (car ownership and commuting trip distance). Specifically, CO2 emissions are negatively
affected by land-use mix, residential density, metro station density and road network den-
sity. Conversely, bus stop density, distance to city centers and parking availability near the
workplace have positive effects on CO2 emissions. To promote low carbon travel, interven-
tion on the built environment would be effective and necessary.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction

Over the last decade, energy consumption and related CO2 emissions from human activities contributing to global climate
change have caused widespread concern throughout the international community (Solomon, 2007; Ou et al., 2013). More-
over, because transportation is the fastest growing sector worldwide in energy consumption and CO2 emissions, reducing
CO2 emissions from transportation is considered a major challenge in achieving climate change mitigation goals (Ma
et al., 2015) and is perceived as the most difficult sector for which to reduce emissions (Marsden and Rye, 2010; Brand
et al., 2012). The New Urbanism movement emerging in the U.S. in the late 1980s proposed reconnecting transport with land
use and attempted to reduce the environmental impact of urban and regional development through land use planning and
related policies (Newman and Kenworthy, 1996; Crane and Crepeau, 1998). Meanwhile, the Compact City Policy in Europe

⇑ Corresponding author at: School of Geography Science and Planning, Sun Yat-sen University, Guangzhou 510275, China.
E-mail addresses: caoxsh@mail.sysu.edu.cn (X. Cao), yangwenyue900780@163.com (W. Yang).

http://dx.doi.org/10.1016/j.trd.2017.02.003
1361-9209/Ó 2017 Elsevier Ltd. All rights reserved.
X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494 481

advocated high-density and mixed-use neighborhoods to encourage shorter trips and more non-motorized trips, and to
reduce car use and travel distances through urban planning (Van Acker and Witlox, 2010).
In this context, a considerable amount of research in the planning field considers the relationship between travel behavior
and the built environment (Handy et al., 2002). Most of these, however, focus on travel variables, including trip frequencies,
trip lengths, mode choices or modal splits and person miles traveled (PMT), vehicle miles traveled (VMT) or vehicle hours
traveled (VHT), and no consistent conclusions have been reached (Ewing and Cervero, 2001, 2010). Moreover, less focus
has been placed on the environmental costs of travel, such as CO2 emissions and considerably less research considers the
impact of built environments on travel CO2 emissions by taking residential self-selection into account (Cao et al., 2009;
Cao, 2014). Whether the observed effect of the built environment on travel CO2 emissions is from the built environment
itself, or because individuals choose their residential built environment based on their socio-demographic characteristics
and travel related attitudes and preferences, thereby forming specific travel patterns and their CO2 emissions, these two
cases are quite different. In the latter case, residential self-selection is at work and at least partially contributes to differences
in CO2 emissions from travel. If the effect of residential self-selection exists but is not accounted for, we are likely to over-
estimate the effects of the built environment and mislead the development of low-carbon land use and transport policies.
Commuting is regarded as the most important urban traffic. Over the last few decades, numerous studies have examined
the relationship between commuting patterns and land use patterns (e.g., Cervero, 1996; Banister et al., 1997; Layman and
Horner, 2010; Zhao et al., 2011; Ding et al., 2014a; Mendiola et al., 2014; Liu et al., 2016). With rapid urbanization and
motorization, CO2 emissions from commuting trips will increase significantly and exacerbate the global climate change
problem (Wang et al., 2016). Therefore, this study focuses on commuting trips and their related CO2 emissions, providing
a guide for future low-carbon city and low-carbon transport planning.
Chinese cities, especially large cities, are now experiencing suburbanization and dispersion of development; the phe-
nomenon is similar to that observed in the U.S. (Cao, 2015b). However, Chinese built environment characteristics and res-
idents’ travel preferences are different from those in Western countries (Wang and Lin, 2014). Is the effect of residential
self-selection in Chinese cities the same as in Western cities (especially North America)? How and to what extent does
the built environment affect CO2 emissions from commuting after controlling for the effect of residential self-selection?
To address the above two questions, we use the city of Guangzhou in China as the study area and present an empirical study
examining the effects of the built environment and residential self-selection on commuting trips and related CO2 emissions.
The remainder of the paper is organized as follows: Section 2 briefly reviews the relevant literature. Section 3 focuses on
the methodology used in the analysis. Section 4 examines the estimation results of the models and explains the effects of
socio-demographics, built environments and attitudes and preferences on CO2 emissions from commuting trips. Section 5
provides primary conclusions and policy recommendations.

2. Literature review

2.1. Built environments, travel behavior and CO2 emissions

The built environment is mainly composed of land use, urban design and the transportation system (Handy et al., 2002).
Cervero and Kockelman (1997) proposed the ‘‘three Ds” of density, diversity and design as measures of the built environ-
ment. Subsequently, destination accessibility and distance to transit were added, as the other two Ds, while demand man-
agement, including parking availability and cost, was considered the sixth D in several studies (Ewing and Cervero, 2001,
2010). Numerous studies have focused on the effects of the built environment on individual travel behavior measured in sev-
eral ways including travel mode choice, travel distance, travel frequency, travel purpose or travel time (Crane, 2000; Ewing
and Cervero, 2001, 2010; Handy et al., 2005; Boarnet, 2011). Although the results of these studies were not entirely consis-
tent, they shared some commonalities: As summarized by Ewing and Cervero (2001), trip frequencies were found to be
mainly determined by socio-demographics followed by the built environment. Trip lengths were found to be primarily a
function of the built environment followed by a function of socio-demographics. Mode choice depended on the built envi-
ronment and socioeconomics, although the effect of the latter may be greater. For overall vehicle miles traveled (VMT) or
vehicle hours traveled (VHT), the effects of the built environment were found to be much more significant. A study by
the authors conducted a decade later included a meta-analysis of the built environment-travel literature and obtained
results similar to the previous study. In this new study, more built environment variables were estimated simultaneously
and were controlled for more confounding effects. The study concluded that the combined effect of several built environ-
ment variables on travel was substantial (Ewing and Cervero, 2010).
In the past decade, based on national time series data or cross-section data of cities, many aggregated studies at the macro
level of states and regions have been performed on the relationship between the urban form (such as urban density, city size,
built-up area and road density) and transportation energy consumption and related CO2 emissions (Newman and
Kenworthy, 1989; Kenworthy and Laube, 1996; Ewing, 1997; Hughes et al., 2004; Newman, 2006; Shim et al., 2006; Lu
et al., 2007; Alford and Whiteman, 2009; Hankey and Marshall, 2010; Zhao, 2010; Aguiléra and Voisin, 2014; Yang et al.,
2015). At the microscopic and disaggregate levels, existing studies were mainly based on questionnaire data. They found
a negative correlation between residential density and transportation energy consumption (Brownstone and Golob, 2009;
Modarres, 2013). While some argued that densities at the workplace played a greater role in reducing transport energy con-
482 X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494

sumption and related greenhouse gas emissions than at the home (Ding et al., 2014b), a case study in Montreal by Zahabi
et al. (2012) showed that the effects of built environment variables were statistically significant. 10% increase in density,
transit accessibility and land-use mix implied 3.5%, 5.8% and 2.5% reduction in travel related greenhouse gas emissions,
respectively. Similarly, Hong and Goodchild’s study in Puget Sound region showed that a 100% increase in residential density,
land-use mix and intersection density in urban areas could generate 31.2–34.4% reductions in transportation emissions
(Hong and Goodchild, 2014). Moreover, Hong (2015) found a non-linear relationship between density and transportation
CO2: the emissions reduction effect of population density became insignificant above a certain level. However, in some other
studies, the association between residential density and CO2 emissions from travel was found to be insignificant (Barla et al.,
2011; Jiang et al., 2011; Xiao et al., 2011). Subway accessibility was found to be inversely associated with CO2 emissions from
commuting in Beijing (Ma et al., 2015), while Xiao et al.’s study showed that bus accessibility had a positive effect on travel
CO2 emissions (Xiao et al., 2011). Meanwhile, many other studies focused more on the impact of socio-demographic vari-
ables on travel CO2 emissions and only considered a small number of built environment variables, such as location, popu-
lation density, distance, and perceived accessibility (Brand, 2009; Barla et al., 2011; Ko et al., 2011; Brand et al., 2013). In
general, the above-mentioned studies did not reach the same conclusions. Moreover, most of them only assessed the direct
effects of built environment variables on travel CO2 emissions and their significance, and ignored the indirect effects (which
may affect other variables, ultimately affecting travel CO2 emissions).

2.2. The effect of residential self-selection

The determination of people’s travel behavior and related CO2 emissions by the built environment is a different phe-
nomenon from self-selection effects characterized by residents moving into a particular built environment based on their
travel mode preferences and land use patterns (Mokhtarian and Cao, 2008). For example, Handy and Clifton (2001) found
that residents who preferred walking might consciously choose to live in walkable communities; thus, they walked more
and emitted less CO2. Hence, the built environment, as well as travel preferences of residents, and residential choice, affected
the travel choice and CO2 emissions. Thus, the effect of residential self-selection may be at work, and the effect of travel-
related attitudes and preferences in housing choice should be considered in travel behavior studies. In addition to attitudes
and preferences, residential self-selection also resulted from socio-demographics (Mokhtarian and Cao, 2008). For example,
low-income residents without a car are likely to choose to live in neighborhoods with adequate transit service and hence
travel by transit more and emit less CO2. This case shows that it’s individuals’ economic constraints rather than good transit
facilities that have a true and direct effect on their mode choice and travel-related CO2 emissions (Cao et al., 2009). Therefore,
it is important to consider socio-demographic variables in investigating the effect of residential self-selection. If the effect of
residential self-selection cannot be distinguished, then the impact of built environments on travel behavior (including the
environmental cost of travel, e.g., CO2 emissions) would be over-estimated. This overestimation would affect transport
and land use policy decisions (Bhat and Guo, 2007; Bohte et al., 2009; Cao et al., 2009; Cao, 2015a,b).
Existing studies have determined the effects of residential self-selection and reached important conclusions (Cao et al.,
2007; Pinjari et al., 2008; Zhou and Kockelman, 2008; Bhat et al., 2013; He and Zhang, 2014). Most studies suggested that
the effects of built environment variables remained statistically significant, even after controlling for statistically significant
effects of residential self-selection that complicated the relationship between built environments and travel behavior
(Mokhtarian and Cao, 2008; Cao et al., 2009; Hong and Shen, 2013; Cao, 2015a,b). With regard to studies of the impact of
self-selection on commuting, most focused on mode choice and trip distances. The built environment was usually found
to have significant impact on commute mode choice behavior after accounting for residential self-selection effect
(Cervero, 2007; Pinjari et al., 2007; Chen et al., 2008). But research by Scheiner (2010), neither residential self-selection
nor lifestyles were found to have a strong impact on commute distances. Although some studies have explored the influence
of land use characteristics and residential self-selection on multiple travel outcomes for commuting (Jahanshahi et al., 2015;
Jahanshahi and Jin, 2016b), little research has examined the effect of residential self-selection on CO2 emissions from com-
muting and most empirical studies on the theory of residential self-selection are from the West (Cao, 2014; Wang and Lin,
2014). In the case study of Hangzhou, China, Naess (2010) held a different opinion on the effect of residential self-selection.
Transport development in developing countries is quite different from that in developed countries (Wright and Fulton, 2005;
Sperling and Salon, 2011). Moreover, Chinese cities differ significantly from European and American cities in terms of the
characteristics, structure and per capita level of transport CO2 emissions (Dodman, 2009; Yang et al., 2015). In addition,
the Chinese have special travel-related attitudes and preferences, such as preferences for short commutes (Wang and Lin,
2014). Therefore, studying the effects of built environments and residential self-selection on CO2 emissions from commuting
in Chinese cities may lead to different conclusions from similar research focused on Western cities.

3. Methodology

3.1. Study area and communities surveyed

We used Guangzhou, which is the largest city in southern China and includes 10 districts, as the study area in this paper.
It includes 2055 communities and covers an area of 3647.43 km2. Over the past decade, the growth rate of car ownership in
X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494 483

Guangzhou has been rapid despite a well-developed public transport system. The city, therefore, has great research signif-
icance as a study area, and could also serve as an excellent reference for other cities, especially those with similar city scale,
urban form and traffic structures.
To account for the effect of residential self-selection and to ensure that residents are able to choose their own homes
according to their attitudes and preferences, the communities which were primarily commercial housing were first selected.
On this basis, in order to ensure statistical significance of model estimates and represent communities in different built envi-
ronments, fifteen communities that varied systematically in terms of location, residential density, land use diversity, public
transport supply level and road facilities supply level were selected for the survey (as shown in Fig. 1). Face-to-face survey
interviews were conducted from May to July 2015 to collect respondents’ socio-demographic attributes, attitudes, prefer-
ences, travel modes, etc. In addition, respondents’ residential and work addresses were used to measure the commuting trip
distance and the related CO2 emissions. We collected 1239 valid questionnaires in total.

3.2. Variables and data

Additional data besides that obtained from the survey focused on built environment elements including distance to city
centers (DTC), land-use mix (LUM), residential density (RD), bus stop density (BSD), metro station density (MSD) and road
network density (RND) at the community level. We principally measured these built environment elements using GIS.
Specifically, the distance to the city center for each community was measured through the average Euclidean distance from
the center of the community to 16 urban public centers of different types. The land-use mix of the community was calculated
by methods similar to those used in previous studies (Frank et al., 2004; Moniruzzaman et al., 2013), but using points of

Fig. 1. The communities surveyed in the study area of Guangzhou.


484 X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494

interest (POIs) rather than land use geospatial data, because they better reflected the diversity of the land use. Thirteen types
of POIs were considered, including residences, hotels, restaurants, supermarkets, parks, squares, shopping malls, commercial
buildings, schools, hospitals, banks, sports venues and government organizations. We spatially visualized the residential
density for each community based on the data of the 6th National Population Census. Bus stop density and metro station
density were calculated using kernel density estimation in ArcGIS, wherein the station location data were extracted from
the Baidu map application through API (Application Programming Interface). Road network density was also measured in
ArcGIS based on line density. The built environment characteristics of each community surveyed are summarized in Table 1.
Moreover, we collected the data on the convenience of parking near work places (WCP) in the questionnaires.
Commuting trip distance (TD) was also measured using the TIQS (Travel Intelligent Query System), which we developed
based on the Baidu map LBS (Location Based Service) open platform. CO2 emissions from commuting trips were calculated
using the following formula:
CEi ¼ MTDi  EF m ð1Þ
where CEi represented the CO2 emissions from the commuting trip (we focused on the one-way commuting trip from res-
idence to work place) for respondent i, MTDi was a portion of the commuting trip distance respondent i traveled via a motor-
ized mode, and EFm was the emission factor associated with the travel mode m in the related commuting trip for respondent
i. The emission factor for motorized travel modes referencing Entwicklungsbank (2008) is given in Table 2. Noted that TD
was not equal to motorized travel distance (MTD) because in addition to MTD, it also included non-motorized travel dis-
tance, such as the walking distance from an individual’s home to a bus stop.
The socio-demographic characteristics collected from questionnaires included gender, age, household size (HS), the pres-
ence of children under 16 years old (CHILD), education (EDU), hukou, monthly household income per capita (INCOME) and
car ownership (CAR). Table 3 presents the socio-demographic characteristics of the sample.
Respondents were asked to identify their travel-related attitudes and preferences, which included attitude toward low
carbon travel (low carbon travel is important, AP1), attitude toward car travel as a symbol of identity (car travel is a symbol
of identity, AP2), preference for sports and exercise (I like sports and exercise, AP3), preference for walking and bicycling (I
like walking and riding a bicycle, AP4), preference for traveling using public transport (I like traveling by public transport,
AP5), preference for driving (I like traveling by car, AP6), and preference for working close to the place of residence (I like
to work nearby, AP7), on a five-point scale where 1 indicates ‘‘strongly disagree” or ‘‘not at all true”, and 5 indicates ‘‘strongly
agree” or ‘‘entirely true”.

3.3. Structural equations model (SEM)

The literature review briefly highlighted the relationship between built environments, travel behavior and related CO2
emissions, as well as the effects of residential self-selection. Hence, structural equations modeling, which has been fre-
quently applied to investigate the complex relationship between the built environment and travel behavior (Bagley and
Mokhtarian, 2002; Cao et al., 2007; Van Acker and Witlox, 2010; Aditjandra et al., 2012; Wang and Lin, 2013), was employed
in the analysis for this paper. Compared to traditional multivariate regression models, SEM is more capable of resolving the
endogeneity problem between variables and of exploring direct, indirect and total effects among exogenous and endogenous
variables (Cao et al., 2007; Cervero and Murakami, 2010; Aditjandra et al., 2012; Jahanshahi and Jin, 2016a,b). An SEM with-
out latent variables can be expressed (Lu and Pas, 1999; Cao et al., 2007) as follows:

Table 1
Built environment characteristics of the fifteen surveyed communities.

Community District Population Sample DTC LUM RD BSD MSD RND


m – Person/km2 unit/km2 unit/km2 km/km2
Fuli (FL) Liwan 8831 63 7366 0.54 114,489 8.91 0.68 8.93
Wuyang (WY) Yuexiu 14,110 88 4963 0.57 39,885 6.28 1.05 7.63
Yijingcuiyuan (YJCY) Haizhu 13,979 75 7230 0.48 24,695 6.89 0.23 6.99
Guangdahuayuan (GDHY) Haizhu 17,466 102 8042 0.18 32,147 6.09 0.36 7.97
Fangcaoyuan (FCY) Tianhe 4230 39 5929 0.35 63,200 7.72 0.67 7.28
Junjinghuayuan (JJHY) Tianhe 12,320 109 9341 0.36 13,827 4.85 0.36 6.43
Zhonghaikangcheng (ZHKC) Tianhe 8016 69 10,714 0.27 17,580 4.56 0.21 5.86
Huiqiaoxincheng (HQXC) Baiyun 15,044 121 9491 0.47 56,825 8.07 0.02 8.68
Fulicheng (FLC) Baiyun 4372 41 14,051 0.27 10,343 5.70 0.00 4.78
Jinbi (JB) Huangpu 10,685 89 13,364 0.40 63,149 4.75 0.10 5.38
Wankehuayuan (WKHY) Huangpu 4747 34 17,121 0.25 29,717 4.45 0.29 4.48
Luoxixincheng (LXXC) Panyu 12,599 109 10,999 0.25 13,938 5.15 0.25 4.81
Lijianghuayuan (LJHY) Panyu 11,056 95 12,130 0.41 9989 5.32 0.21 4.42
Qifuxincun (QFXC) Panyu 27,136 159 19,640 0.25 6980 1.38 0.00 2.83
Dongyi (DY) Panyu 7874 46 24,462 0.57 20,503 3.52 0.12 4.31
Total 172,465 1239 11,656 0.37 34,484 5.58 0.30 6.05
X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494 485

Table 2
Specific energy consumption and CO2 emission factor for motorized travel modes. Source: Entwicklungsbank (2008).

Motorized travel Final energy consumption Capacity Primary energy CO2 (g/Pkm)
modes (l/100 km, KWh/km) (Persons) consumption (MJ/Pkm)
Passenger car 11.0 1.3 0.84 233.1
Urban bus 35.0 40 0.35 26.0
Coach 30.0 44.0 0.27 20.3
Metro 5.0 216 0.26 20.9

Table 3
Socio-demographic characteristics of the sample.

Variable Level N (%) Median CE (g)


Gender 0 for male 694 (56.05%) 366.27
1 for female 545 (43.95%) 249.53
Age 1 represents age 16–24 137 (11.08%) 130.12
2 represents age 25–34 605 (48.79%) 294.17
3 represents age 35–44 426 (34.39%) 439.70
4 represents age 45 71 (5.73) 266.39
Household size 1 represents 1 people 38 (3.06%) 253.92
2 represents 2 people 140 (11.34%) 274.81
3 represents 3 people 429 (34.65%) 284.73
4 represents 4 people 355 (28.66%) 369.69
5 represents 5 people 276 (22.29%) 316.73
Any child under 16 0 for no 414 (33.38%) 266.00
1 for yes 825 (66.62%) 350.85
Education 1 represents senior high school and below 151 (12.19%) 100.64
2 represents junior college 357 (28.79%) 277.21
3 represents bachelor degree 551 (44.46%) 348.72
4 represents master degree or above 180 (14.52%) 420.74
Hukou 0 for other cities 584 (47.13%) 218.71
1 for Guangzhou 655 (52.87%) 373.10
Household monthly incomes per capita 1 represents income 3999 RMB 128 (10.33%) 120.31
2 represents income 4000–5999 RMB 221 (17.83%) 208.59
3 represents income 6000–7999 RMB 208 (16.82%) 319.69
4 represents income 8000–9999 RMB 202 (16.31%) 389.13
5 represents income 10,000–14,999 RMB 208 (16.82%) 533.57
6 represents income 15,000 RMB 271 (21.91%) 357.44
Car ownership 0 for no 488 (39.36%) 195.57
1 for yes 751 (60.64%) 470.91

y ¼ By þ Cx þ f ð2Þ

where y refers to the vector of endogenous variables, x refers to the vector of exogenous variables, B is the matrix of coef-
ficients representing the effects of endogenous variables on other endogenous variables, C is the matrix of coefficients rep-
resenting the effects of exogenous variables on endogenous variables, and f is the vector of errors in the equation.
To examine the effects of the built environment on commuting trips and the related carbon emissions after controlling for
the effect of residential self-selection and to explore the effect of the residents’ attitudes and preferences, two SEMs were
constructed. One model (SEM 1) did not consider the attitude and preference variables; car ownership (CAR), commuting
trip distance (TD) and CO2 emissions from commuting trip (CE) were set as endogenous variables. Specifically, CAR was con-
sidered to mediate the relationship between the exogenous variables and other endogenous variables, as well as TD (Cao
et al., 2007; Van Acker and Witlox, 2010; Shen et al., 2016), while the other socio-demographic variables and built environ-
ment variables were set as exogenous variables. The other model (SEM 2) considered the attitude and preference variables
used as exogenous variables, and the built environment variables as endogenous variable. The links between the variables in
SEM 1 were preserved in SEM 2. Moreover, SEM 2 investigated the links between attitude and preference variables and
endogenous variables, as well as the links between socio-demographic variables and the built environment variables. If
the links are statistically significant, it indicates that residents chose their place of residence with different built environment
characteristics according to their different attitudes and preferences or socio-demographic attributes.
The conceptual framework for constructing these two SEMs is presented in Fig. 2. If the effects of attitude and preference
variables on the built environment and on CE or TD are significant in SEM 2 simultaneously, it means that the effect of res-
idential self-selection exists. Because residential self-selection results from two sources of attitudes and preferences, and
socio-demographics (Mokhtarian and Cao, 2008), we can also infer the existence of residential self-selection if the effects
486 X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494

Fig. 2. Conceptual framework for the SEM construction.

of socio-demographic variables on the built environment and on CE or TD are significant simultaneously. On this basis, if the
effects of the built environment on CE are significant in SEM 1, but not significant in SEM 2, it shows that the significant asso-
ciation between the built environment and CE in SEM 1 can be attributed to the effect of residential self-selection rather than
the built environment itself. If the effects of the built environment on CE are statistically significant in SEM 2, it indicates that
after controlling for the effect of residential self-selection, the built environment still has significant direct effects on CE. The
SEMs were estimated using Amos 21.0. Because the data did not have a multivariate normal distribution, the Bollen-Stine
bootstrap method was used to develop the SEMs (Chowdhury and Ceder, 2013; Ma et al., 2014). The number of bootstrap
samples we set was 2000. The ordinal categorical variables, such as AGE, HS, EDU, INCOME and attitude and preference vari-
ables, were introduced into the models as continuous variables.

4. Results

4.1. Goodness-of fit for SEMs

According to the conceptual framework for the SEM construction mentioned in the preceding paragraph, the two models
were estimated successively. The links with no statistical significance (p > 0.1) were removed, and the model was re-
estimated. The model was modified and improved according to the modification indices (M.I.) provided by AMOS 21.0.
The model results estimated with the Bootstrap method were highly similar to the results from the maximum likelihood
estimation (Appendixes A and B). Table 4 showed the goodness-of-fit indices of the two models and their corresponding ref-
erence values. All of the indices suggested that the models fitted the data well (Wu, 2010).

4.2. Path analysis without considering the residential self-selection

In terms of endogenous variables, both CAR and TD had highly significant positive direct effects on CE (Fig. 3, Table 5 and
Appendix A). Moreover, CAR had a positive direct effect on TD at the 0.05 significance level, indicating that the residents who
owned cars had a longer commuting trip distance, causing this factor to have a positive indirect effect on CE. However, TD
had a negative direct effect on CAR at the 0.05 significance level, thereby having a negative indirect effect on CE. This also
meant that residents with a longer commute were not more likely to own a car. Overall, the direct effect of TD on CE was
much larger than the indirect effect, and the total effect of TD on CE is positive.
For socio-demographic variables, only GENDER had a negative total effect on CE, which indicated that women (GEN-
DER = 1) had lower CE than men. In contrast, all of the other variables had a positive total and indirect effect on CE, indicating

Table 4
Model fit indices for the two models.

Model fit indices Reference value Model-based value


SEM without considering AP SEM considering AP
Chi-square (v2) 63.771 146.327
Degrees of freedom (df) 68 166
Bollen-Stine bootstrap P-value >0.05 0.649 0.852
Goodness of Fit Index (GFI) >0.9 0.991 0.985
Adjusted Goodness of Fit Index (AGFI) >0.9 0.979 0.973
Comparative Fit Index (CFI) >0.9 1.000 1.000
Normed Fit Index (NFI) >0.9 0.989 0.977
Non-Normed Fit Index (NNFI) >0.9 1.002 1.005
Root Mean Square Error of Approximation (RMSEA) <0.05 0.000 0.000
X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494 487

Fig. 3. SEM path diagram explaining the effects of socio-demographic variables and built environment variables on CE.

Table 5
Standardized total, direct and indirect effects of variables on endogenous variables.

Endogenous variables CAR TD CE


Effect Total Direct Indirect Total Direct Indirect Total Direct Indirect
Exogenous variables
Socio-demographics
HS 0.208a 0.215a 0.007b 0.031b – 0.031b 0.077a – 0.077a
GENDER – – – – – – 0.092a 0.092a –
AGE 0.082a 0.084a 0.003b 0.012b – 0.012b 0.126a 0.096a 0.030a
EDU 0.078b 0.111a 0.033a 0.158a 0.146a 0.012b 0.032 0.065b 0.097a
HUKOU 0.069b 0.071c 0.002c 0.010c – 0.010c 0.079a 0.054c 0.025b
INCOME 0.297a 0.307a 0.009b 0.045b – 0.045b 0.110a – 0.110a
Built environment
DTC 0.079a – 0.079a 0.381a 0.393a 0.012b 0.153a – 0.153a
LUM 0.078b 0.080b 0.002b 0.012c – 0.012c 0.109a 0.080b 0.029b
RD 0.030b – 0.030b 0.144b 0.149b 0.005b 0.058b – 0.058b
BSD 0.093a – 0.093a 0.447a 0.461a 0.014b 0.292a 0.113a 0.179a
MSD 0.079b 0.082b 0.002b 0.012b – 0.012b 0.029b – 0.029b
RND 0.064b – 0.064b 0.310a 0.319a 0.010b 0.124a – 0.124a
WCP – – – – – – 0.107a 0.107a –
Endogenous variables
CAR 0.030b – 0.030b 0.146b 0.151b 0.005b 0.359a 0.301a 0.059b
TD 0.202b 0.208b 0.006b 0.030b – 0.030b 0.388a 0.463a 0.075b

Note: Links that are not included in the model are indicated by ‘–’.
a
Significant at the 0.01 level.
b
Significant at the 0.05 level.
c
Significant at the 0.1 level.
488 X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494

that older residents and those with a larger household size, higher education, higher household monthly incomes per capita
and hukou of Guangzhou (HUKOU = 1) were more likely to own a car and to emit more CO2 from commuting trips. Because
CHILD did not display significant direct or indirect effects on CE, the links between it and other endogenous variables in the
process of model modification were removed.
With respect to built environment variables, only LUM, BSD and WCP had a significant direct effect on CE, while all of the
variables had significant indirect effects on CE. This indicated that although some built environment variables (DTC, RD, MSD
and RND) had no direct effects on CE, they had indirect effects on CE through the effects of mediating variables (CAR, TD or
both).
Based on their total effects, LUM, RD, MSD and RND had negative effects on CE, indicating that residents living in com-
munities with higher land-use mix, higher residential density, higher metro station density and higher road network density
tended to emit less CO2 from commuting trips. Specifically, LUM and MSD had significant negative direct effects on CAR,
while RD and RND had significant negative direct effects on TD, hence having significant indirect effects on CE. The findings
on the negative effect of resident community density on commuting trip distance and the related CO2 emissions were con-
trary to the conclusion of the empirical research by Brand et al. (2013) in the UK and Xiao et al. (2011) in Beijing, China.
In contrast, DTC, BSD and WCP had positive total effects on CE, suggesting that individuals living far from the city centers,
living in higher bus stop density communities or working near locations with ample parking spaces would probably have
more CO2 emissions from commuting trips. The findings of the positive effect of BSD on CE were contrary to many studies
in Western countries in which higher transit accessibility was inversely associated with commuting distance and related CO2
emissions, but it was consistent with Xiao et al.’s study in Beijing (Xiao et al., 2011). This is likely to be caused by the fol-
lowing two reasons. First, BSD was strongly associated with TD and the standardized direct effect coefficient was 0.461,
which meant the more bus stops, the longer commuting trip distance. For a specific trip, traveling by bus usually produces
a longer distance than that by car, by walking or by cycling because of the detours in bus routes. This explained why BSD was
strongly associated with TD to a certain extent. Secondly, compared to metro and non-motorized modes, bus was more
carbon-intensive. High bus stop density would encourage more bus travel rather than subway travel or non-motorized tra-
vel, which would generate more CO2 emissions. For example, an individual may prefer to use the bus to commute instead of
walking or riding a bicycle, even for short-distance commute trips, which would lead them to emit more carbon-dioxide.

4.3. Path analysis considering residential self-selection

When the attitude and preference variables were incorporated into SEM 2 with the built environment variables as
endogenous variables, most of the links constructed in SEM 1 did not show substantial changes, except for the link between
HUKOU and CE, which became insignificant (Fig. 4, Tables 6 and 7 and Appendix B). Moreover, some of the socio-
demographic variables and attitude and preference variables had significant direct effects on the built environment, thereby
indirectly influencing CE.
As shown in Table 6, in addition to AP3 and AP7, all of the other attitude and preference variables had direct or indirect
effects on CE. The total effects of AP1, AP2, AP4, AP5 and AP6 on CE were 0.014, 0.013, 0.012, 0.055 and 0.212, respec-
tively, indicating that residents who preferred to walk and ride a bicycle or those who preferred to travel by transit tended to
emit less CO2 from commuting trips, while individuals who preferred traveling by car exhibited the reverse. However, the
results also noted that individuals who agreed with the importance of low carbon travel did not have reduced CO2 emissions
from commuting trips and that individuals who regard car travel as a symbol of identity also did not necessarily emit more
CO2 from commuting. Meanwhile, the significant direct effects of some socio-demographic variables and attitudes and pref-
erence variables on the built environment variables (one or several of them) implied that residents with different socio-
demographic characteristics would choose to live in communities with different built environment characteristics according
to their attitudes and preferences. This would affect their decisions regarding owning a private car and commuting distance,
ultimately influencing their CO2 emissions from commuting trips. These results demonstrated the existence of residential
self-selection effect, which should not be ignored in the model. After controlling for the effect of residential self-selection,
the built environment still had significant (direct or indirect) effects on CE (Table 7). However, the total and direct effects
of almost all built environment variables on CE were diminished in SEM 2 except MSD. The differences of their coefficients
in Tables 5 and 7 reflects the extent of effects that were erroneously attributed to the built environment if ignoring the res-
idential self-selection effect. Specifically, the total effects on CE that were erroneously attributed to DTC, LUM, RD, BSD, RND
and WCP were 0.011, 0.003, 0.002, 0.015, 0.024 and 0.015, respectively (Table 8).

5. Conclusions and policy implications

China is experiencing rapid suburbanization, where new towns and suburbs characterized by large blocks and wide roads
have abandoned the traditional community model based on walking and public transport (Cao, 2015a,b). This unsustainable
pattern of suburbanization and spreading development makes their residents increasingly dependent on cars and is likely to
occur in other developing countries. Hence, using Chinese cities as study areas is of great importance. This study investigated
the complex influence mechanisms between socio-demographics, built environments, attitudes and preferences, and com-
muting trips and their CO2 emissions by employing structural equations modeling based on survey data. Two SEMs were
X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494 489

Fig. 4. SEM path diagram explaining the complex influence mechanisms between socio-demographic variables, built environment variables, attitude and
preference variables and CE.

constructed to address the effects of residential self-selection and to improve the robustness of the estimation results. Self-
selection is seldom studied in China (Wang and Lin, 2014; Cao, 2015a,b). This study fills the gaps by studying the issue from a
developing country and enriches the literature of self-selection, which previously was mainly conducted in developed
countries.
The conclusions of our research and policy implications for low-carbon land use and transport planning in China’s large
cities include the following: (1) Similar to North American cities, the effect of residential self-selection was demonstrated to
exist in Chinese cities, indicating that it should not be ignored in future studies of travel in the Chinese context. Residents’
attitudes and preferences will affect their residential choices, and ultimately affect their commuting trip CO2 emissions. Indi-
viduals with preferences for walking and cycling or those who prefer to travel by transit tend to emit less CO2 from commut-
ing trips, while individuals who prefer traveling by car tend to emit more. Therefore, reshaping residents’ attitudes and
preferences and cultivating the habit of walking and cycling have great significance for promoting low-carbon travel. (2)
After controlling for the effects of residential self-selection, built environment variables still had significant effects on CO2
emissions from commuting trips although some only indirectly through the mediating effects of car ownership, commuting
trip distance, or both. This pattern indicates that built environment elements did not necessarily have a direct impact on CO2
emissions from commuting, but improving the built environment would eventually lead to the mitigation of CO2 emissions
from commuting. Therefore, it is necessary to intervene in the built environment through land use and transport policies to
create a low-carbon urban spatial structure. (3) Specifically, land-use mix, residential density, metro station density and road
network density had negative effects on CO2 emissions from commuting trips, while distance to city centers, bus stop density
and convenience of parking near the workplace had positive effects on CO2 emissions. This indicated that providing a variety
of service facilities within a community reduced the need to have a car and promoted non-motorized travel for residents.
Therefore, mixed use of land in communities is an important strategy to promote low-carbon travel. Compared to suburban
communities characterized by large blocks and wide roads, a more sustainable community mode conducive to promoting
low carbon travel involved proximity to urban public centers, dense road networks and high residential densities. Providing
adequate subway services was more effective in reducing residents’ CO2 emissions from commuting than increasing bus ser-
vice. Meanwhile, reducing the efficacy of the automotive system, such as by controlling the number of parking spaces in the
employment centers, constituted efficient land use policies to change the travel behavior of residents and reduce related CO2
emissions.
490 X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494

Table 6
Standardized total, direct and indirect effects of exogenous variables on endogenous variables.

Exogenous variables Effect DTC LUM RD BSD MSD RND WCP CAR TD CE
HS Total – – – – – – – 0.197a 0.048a 0.074a
Direct – – – – – – – 0.211a – –
Indirect – – – – – – – 0.014a 0.048a 0.074a
CHILD Total – – – – 0.037c – – 0.003b 0.001b 0.001c
Direct – – – – 0.037c – – – – –
Indirect – – – – – – – 0.003b 0.001b 0.001c
GENDER Total – – 0.039b – – – – 0.002b 0.006b 0.086a
Direct – – 0.039b – – – – – – 0.084a
Indirect – – – – – – – 0.002b 0.006b 0.002b
AGE Total – – – – – – – 0.079a 0.019a 0.127a
Direct – – – – – – – 0.085a – 0.097a
Indirect – – – – – – – 0.006a 0.019a 0.030a
EDU Total – 0.040c – – – 0.030b – 0.075b 0.156a 0.033
Direct – 0.040c – – – 0.030b – 0.118a 0.129a 0.062b
Indirect – – – – – – – 0.042a 0.027a 0.095a
HUKOU Total 0.052c – – 0.078a – 0.080a – 0.057c 0.004 0.070c
Direct 0.052c – – 0.078a – 0.080a – 0.058c – 0.043
Indirect – – – – – – – 0.001 0.004 0.027c
INCOME Total – – – – – 0.039a – 0.283a 0.058a 0.102a
Direct – – – – – 0.039a – 0.299a – –
Indirect – – – – – – – 0.017a 0.058a 0.102a
AP1 Total – – – 0.043a 0.053b – – 0.000 0.018a 0.014a
Direct – – – 0.043a 0.053b – – – – –
Indirect – – – – – – – 0.000 0.018a 0.014a
AP2 Total – – – 0.047a – – – 0.005a 0.019a 0.013a
Direct – – – 0.047a – – – – – –
Indirect – – – – – – – 0.005a 0.019a 0.013a
AP4 Total – 0.080a – – – 0.034b – 0.003 0.010a 0.012a
Direct – 0.080a – – – 0.034b – – – –
Indirect – – – – – – – 0.003 0.010a 0.012a
AP5 Total 0.040c 0.045c 0.064a 0.090a 0.075b – 0.067b 0.036a 0.084b 0.055c
Direct 0.040c 0.045c 0.064a 0.090a 0.075b – 0.067b – 0.080b 0.086a
Indirect – – – – – – – 0.036a 0.003 0.031b
AP6 Total – – – – – – 0.072b 0.148a 0.036a 0.212a
Direct – – – – – – 0.072b 0.158a – 0.150a
Indirect – – – – – – – 0.010a 0.036a 0.062a

Note: Links that are not included in the model are indicated by ‘–’.
a
Significant at the 0.01 level.
b
Significant at the 0.05 level.
c
Significant at the 0.1 level.

Table 7
Standardized total, direct and indirect effects of endogenous variables on each other.

Endogenous variables CAR TD CE


Effect Total Direct Indirect Total Direct Indirect Total Direct Indirect
DTC 0.108a – 0.108a 0.371a 0.398a 0.026a 0.142a – 0.142a
LUM 0.074b 0.080b 0.005b 0.018b – 0.018b 0.106a 0.078b 0.028b
RD 0.043a – 0.043a 0.146a 0.156a 0.010a 0.056a – 0.056a
BSD 0.116a – 0.116a 0.396a 0.424a 0.028a 0.277a 0.125a 0.152a
MSD 0.093a 0.100a 0.007a 0.023a – 0.023a 0.035a – 0.035a
RND 0.076a – 0.076a 0.260a 0.279a 0.018a 0.100a – 0.100a
WCP – – – – – – 0.092a 0.092a –
CAR 0.066a – 0.066a 0.226a 0.242a 0.016a 0.352a 0.265a 0.087a
TD 0.273a 0.292a 0.019a 0.066a – 0.066a 0.358a 0.461a 0.103a

Note: Links that are not included in the model are indicated by ‘–’.
a
Significant at the 0.01 level.
b
Significant at the 0.05 level.
c
Significant at the 0.1 level.

Although the ratio (1239/24 > 51) between sample size and the number of observed variables is more than triple the rec-
ommended threshold (=15) (Stevens, 2012), the sample size is still modest compared to some existing studies. Based on
these survey data, our empirical study in Guangzhou may draw a conclusion inconsistent with some existing studies. This
likely results from a failure to consider the effects of residential self-selection and to overestimate (or underestimate) the
X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494 491

Table 8
Differences of total and direct effect of built environment variables between considering and without considering the residential self-selection.

SEM 1 SEM 2 Differences


Effect Total Direct Total Direct Total Direct
DTC 0.153 – 0.142 – 0.011 –
LUM 0.109 0.08 0.106 0.078 0.003 0.002
RD 0.058 – 0.056 – 0.002 –
BSD 0.292 0.113 0.277 0.125 0.015 0.012
MSD 0.029 – 0.035 – 0.006 –
RND 0.124 – 0.1 – 0.024 –
WCP 0.107 0.107 0.092 0.092 0.015 0.015

effects of the built environment; some only estimated the direct effects of the built environment on CO2 emissions from com-
muting, ignoring the indirect effects that work through mediating variables. Moreover, the difference in findings also can be
attributed to the differences between countries, and between cities within countries. Meanwhile, we must clarify that,
because of the lack of vehicular speed-emission data in China, our CO2 estimates are distance based and have not accounted
for the emission factors of different vehicle speeds. We have no choice but to use the average CO2 emission factor for each
motorized travel modes to estimate our models, similar to methods proposed by existing research (e.g., Ma et al., 2015;
Wang et al., 2016; Aguiléra and Voisin, 2014; Barla et al., 2011; Huang et al., 2013; Xiao et al., 2011; Chai et al., 2012).

Acknowledgments

This work was supported by the National Natural Science Foundation of China (41171139, 41130747, 41671160). We
would like to thank Dr. Xinyu (Jason) Cao for his constructive comments and the International Association for China Planning
(IACP) for organizing this special issue. We are also very grateful to the anonymous reviewers for their valuable comments,
which helped improve the quality of the paper. Any remaining errors are, of course, our own responsibility.

Appendix A. Estimated regression weights (unstandardized) between the independent variables and the dependent
variables in SEM without considering attitudes and preferences

Parameter SE SE-SE Mean Bias SE-Bias P (BC)


CAR HS 0.015 0 0.101 0.001 0 0.001
CAR AGE 0.022 0 0.055 0 0 0.01
CAR EDU 0.02 0 0.061 0.001 0 0.002
TD EDU 301.608 4.769 1412.056 12.619 6.744 0.001
CAR HR 0.036 0.001 0.071 0.001 0.001 0.052
CAR INCOME 0.011 0 0.09 0 0 0.001
TD RD 0.017 0 0.045 0 0 0.016
TD BSD 384.99 6.087 1926.687 22.341 8.609 0.001
TD RND 411.829 6.512 1448.393 30.529 9.209 0.002
TD DTC 0.136 0.002 0.752 0 0.003 0.001
CAR LUM 0.152 0.002 0.351 0.006 0.003 0.029
CAR MSD 0.068 0.001 0.143 0.001 0.002 0.03
CE CAR 87.053 1.376 944.187 0.964 1.947 0.001
CE TD 0.011 0 0.083 0 0 0.001
CE GENDER 82.892 1.311 284.506 1.338 1.854 0.001
CE AGE 61.622 0.974 192.689 2.998 1.378 0.001
CE EDU 53.378 0.844 114.629 0.887 1.194 0.034
CE HR 93.171 1.473 167.376 1.28 2.083 0.087
CE BSD 30.071 0.475 84.058 0.295 0.672 0.009
CE WCP 88.091 1.393 336.249 0.924 1.97 0.001
CE LUM 470.175 7.434 1085.734 6.694 10.513 0.027
TD CAR 1402.376 22.174 2693.305 54.096 31.358 0.047
CAR TD 0 0 0 0 0 0.012
Note: SE represents the Bootstrap estimates of standard error, SE-SE represents an approximate standard error for the standard error, Mean represents the
mean of the Bootstrap samples of the quantity estimated, Bias represents the difference between the average of b estimates obtained from b Bootstrap
samples and the single estimate obtained from the original sample (ML); SE-Bias represents an approximate standard error for the bias estimate, and P(BC)
represents the P-value of the parameter in the bias-corrected bootstrap confidence intervals.
492 X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494

Appendix B. Estimated regression weights (unstandardized) between the independent variables and the dependent
variables in SEM considering attitudes and preferences

Parameter SE SE-SE Mean Bias SE-Bias P(BC)


RD GENDER 1121.454 35.463 2208.213 34.732 50.153 0.045
MSD CHILD 0.012 0 0.022 0 0.001 0.08
RND AP4 0.031 0.001 0.081 0 0.001 0.03
DTC AP5 116.986 3.699 238.893 7.699 5.232 0.053
MSD AP5 0.01 0 0.027 0.001 0 0.017
BSD AP5 0.039 0.001 0.238 0.001 0.002 0.004
RD AP5 894.934 28.3 2295.363 21.778 40.023 0.005
LUM AP5 0.004 0 0.007 0 0 0.082
LUM AP4 0.003 0 0.011 0 0 0.004
RND INCOME 0.016 0 0.044 0 0.001 0.003
RND HUKOU 0.092 0.003 0.31 0 0.004 0.003
RND EDU 0.028 0.001 0.064 0.002 0.001 0.025
BSD AP1 0.041 0.001 0.12 0.001 0.002 0.007
BSD AP2 0.03 0.001 0.116 0.001 0.001 0.005
BSD HUKOU 0.099 0.003 0.326 0.003 0.004 0.004
MSD AP1 0.009 0 0.02 0 0 0.03
DTC HUKOU 225.876 7.143 466.481 1.375 10.101 0.063
LUM EDU 0.003 0 0.005 0 0 0.088
CAR HS 0.015 0 0.098 0 0.001 0.003
CAR AGE 0.021 0.001 0.057 0.001 0.001 0.006
CAR EDU 0.019 0.001 0.064 0.001 0.001 0.002
CAR HUKOU 0.035 0.001 0.059 0.002 0.002 0.08
CAR INCOME 0.011 0 0.088 0.001 0 0.005
TD RD 0.016 0.001 0.049 0.001 0.001 0.006
TD BSD 387.211 12.245 1789.918 30.553 17.317 0.006
TD RND 401.586 12.699 1256.055 17.979 17.959 0.004
TD DTC 0.139 0.004 0.759 0.001 0.006 0.004
CAR LUM 0.154 0.005 0.348 0.005 0.007 0.027
CAR MSD 0.063 0.002 0.174 0.001 0.003 0.005
CAR AP6 0.022 0.001 0.108 0.001 0.001 0.006
TD EDU 302.989 9.581 1259.698 0.69 13.55 0.005
TD AP5 362.526 11.464 884.565 2.945 16.213 0.043
WCP AP5 0.022 0.001 0.042 0 0.001 0.049
WCP AP6 0.024 0.001 0.049 0.001 0.001 0.047
CE CAR 87.251 2.759 827.06 1.426 3.902 0.004
CE GENDER 86.596 2.738 262.375 3.707 3.873 0.004
CE AGE 59.05 1.867 195.324 3.222 2.641 0.003
CE EDU 53.259 1.684 107.062 1.164 2.382 0.044
CE BSD 29.323 0.927 92.553 0.349 1.311 0.004
CE WCP 84.797 2.682 290.617 2.037 3.792 0.006
CE LUM 472.98 14.957 1067.703 21.638 21.152 0.039
CE AP5 54.33 1.718 167.131 1.358 2.43 0.003
CE HUKOU 92.027 2.91 132.116 0.553 4.116 0.151
CE TD 0.012 0 0.082 0 0.001 0.003
CE AP6 60.815 1.923 312.735 2.195 2.72 0.004
TD CAR 1162.758 36.77 4264.374 31.045 52 0.004
CAR TD 0 0 0 0 0 0.004
Note: SE represents the Bootstrap estimates of standard error, SE-SE represents an approximate standard error for the standard error, Mean represents the
mean of the Bootstrap samples of the quantity estimated, Bias represents the difference between the average of b estimates obtained from b Bootstrap
samples and the single estimate obtained from the original sample (ML); SE-Bias represents an approximate standard error for the bias estimate, and P(BC)
represents the P-value of the parameter in the bias-corrected bootstrap confidence intervals.
X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494 493

References

Aditjandra, P.T., Cao, X.J., Mulley, C., 2012. Understanding neighbourhood design impact on travel behaviour: an application of structural equations model to
a British metropolitan data. Transport. Res. Part A: Policy Pract. 46, 22–32.
Aguiléra, A., Voisin, M., 2014. Urban form, commuting patterns and CO2 emissions: What differences between the municipality’s residents and its jobs?
Transport. Res. Part A: Policy Pract. 69, 243–251.
Alford, G., Whiteman, J., 2009. Macro-urban form and transport energy outcomes: Investigations for Melbourne. Road Transport Res. 18, 53–67.
Bagley, M.N., Mokhtarian, P.L., 2002. The impact of residential neighborhood type on travel behavior: a structural equations modeling approach. Ann. Reg.
Sci. 36, 279–297.
Banister, D., Watson, S., Wood, C., 1997. Sustainable cities: transport, energy, and urban form. Environ. Plan. B: Plan. Des. 24, 125–143.
Barla, P., Miranda-Moreno, L.F., Lee-Gosselin, M., 2011. Urban travel CO2 emissions and land use: a case study for Quebec City. Transport. Res. Part D-
Transport Environ. 16, 423–428.
Bhat, C.R., Guo, J.Y., 2007. A comprehensive analysis of built environment characteristics on household residential choice and auto ownership levels.
Transport. Res. Part B: Methodol. 41, 506–526.
Bhat, C., Paleti, R., Pendyala, R., et al, 2013. Accommodating immigration status and self-selection effects in a joint model of household auto ownership and
residential location choice. Transport. Res. Rec.: J. Transport. Res. Board, 142–150.
Boarnet, M.G., 2011. A broader context for land use and travel behavior, and a research agenda. J. Am. Plan. Assoc. 77, 197–213.
Bohte, W., Maat, K., van Wee, B., 2009. Measuring attitudes in research on residential self-selection and travel behaviour: a review of theories and empirical
research. Transport Rev. 29, 325–357.
Brand, C., 2009. ‘‘Hockey Sticks” made of carbon unequal distribution of greenhouse gas emissions from personal travel in the United Kingdom. Transp. Res.
Rec., 88–96
Brand, C., Goodman, A., Rutter, H., et al, 2013. Associations of individual, household and environmental characteristics with carbon dioxide emissions from
motorised passenger travel. Appl. Energy 104, 158–169.
Brand, C., Tran, M., Anable, J., 2012. The UK transport carbon model: an integrated life cycle approach to explore low carbon futures. Energy Policy 41, 107–
124.
Brownstone, D., Golob, T.F., 2009. The impact of residential density on vehicle usage and energy consumption. J. Urban Econ. 65 (1), 91–98.
Cao, J., 2014. Residential self-selection in the relationships between the built environment and travel behavior: introduction to the special issue. J. Transport
Land Use 7, 1–3.
Cao, X.J., 2015a. Examining the impacts of neighborhood design and residential self-selection on active travel: a methodological assessment. Urban Geogr.
36, 236–255.
Cao, X.J., Mokhtarian, P.L., Handy, S.L., 2009. Examining the impacts of residential self-selection on travel behaviour: a focus on empirical findings. Transport
Rev. 29, 359–395.
Cao, X., Mokhtarian, P.L., Handy, S.L., 2007. Do changes in neighborhood characteristics lead to changes in travel behavior? A structural equations modeling
approach. Transportation 34, 535–556.
Cao, X., 2015b. Examining the relationship between neighborhood built environment and travel behavior: a review from the US perspective. Urban Plan. Int.
30, 46–52 (in Chinese).
Cervero, R., 1996. Traditional neighborhoods and commuting in the San Francisco Bay Area. Transportation 23, 373–394.
Cervero, R., 2007. Transit-oriented development’s ridership bonus: a product of self-selection and public policies. Environ. Plan. A 39, 2068–2085.
Cervero, R., Kockelman, K., 1997. Travel demand and the 3Ds: density, diversity, and design. Transport. Res. Part D: Transport Environ. 2, 199–219.
Cervero, R., Murakami, J., 2010. Effects of built environments on vehicle miles traveled: evidence from 370 US urbanized areas. Environ. Plan. A 42, 400–418.
Chai, Y., Xiao, Z., Liu, Z., 2012. Low-carbon optimization strategies based on CO2 emission mechanism of household daily travels: a case study of Beijing.
Geograph. Res. 31, 334–344 (in Chinese).
Chen, C., Gong, H., Paaswell, R., 2008. Role of the built environment on mode choice decisions: additional evidence on the impact of density. Transportation
35, 285–299.
Chowdhury, S., Ceder, A., 2013. A psychological investigation on public-transport users’ intention to use routes with transfers. Int. J. Transport. 1, 1–20.
Crane, R., 2000. The influence of urban form on travel: an interpretive review. J. Plan. Literat. 15, 3–23.
Crane, R., Crepeau, R., 1998. Does neighborhood design influence travel?: A behavioral analysis of travel diary and GIS data. Transport. Res. Part D: Transport
Environ. 3, 225–238.
Ding, C., Lin, Y., Liu, C., 2014a. Exploring the influence of built environment on tour-based commuter mode choice: a cross-classified multilevel modeling
approach. Transport. Res. Part D: Transport Environ. 32, 230–238.
Ding, C., Wang, Y., Xie, B., et al, 2014b. Understanding the role of built environment in reducing vehicle miles traveled accounting for spatial heterogeneity.
Sustainability 6, 589–601.
Dodman, D., 2009. Blaming cities for climate change? An analysis of urban greenhouse gas emissions inventories. Environ. Urban. 21, 185–201.
Entwicklungsbank, K., 2008. Transport in China: Energy Consumption and Emissions of Different Transport Modes.
Ewing, R., 1997. Is Los Angeles-style sprawl desirable? American Planning Association. J. Am. Plan. Assoc. 63, 107.
Ewing, R., Cervero, R., 2001. Travel and the built environment: a synthesis. Transport. Res. Rec.: J. Transport. Res. Board 1780, 87–114.
Ewing, R., Cervero, R., 2010. Travel and the built environment. J. Am. Plan. Assoc. 76, 265–294.
Frank, L.D., Andresen, M.A., Schmid, T.L., 2004. Obesity relationships with community design, physical activity, and time spent in cars. Am. J. Prev. Med. 27,
87–96.
Handy, S.L., Boarnet, M.G., Ewing, R., et al, 2002. How the built environment affects physical activity – views from urban planning. Am. J. Prev. Med. 23S, 64–
73.
Handy, S.L., Clifton, K.J., 2001. Local shopping as a strategy for reducing automobile travel. Transportation 28, 317–346.
Handy, S., Cao, X., Mokhtarian, P., 2005. Correlation or causality between the built environment and travel behavior? Evidence from Northern California.
Transport. Res. Part D: Transport Environ. 10, 427–444.
Hankey, S., Marshall, J.D., 2010. Impacts of urban form on future US passenger-vehicle greenhouse gas emissions. Energy Policy 38, 4880–4887.
He, X., Zhang, L., 2014. Quantifying the self-selection effects in residential location choice with a structural equation model. Transport. Res. Rec.: J. Transport.
Res. Board, 153–161.
Hong, J., 2015. Non-linear influences of the built environment on transportation emissions: focusing on densities. J. Transport Land Use 10 (1).
Hong, J., Goodchild, A., 2014. Land use policies and transport emissions: Modeling the impact of trip speed, vehicle characteristics and residential location.
Transport. Res. Part D: Transport Environ. 26, 47–51.
Hong, J., Shen, Q., 2013. Residential density and transportation emissions: examining the connection by addressing spatial autocorrelation and self-
selection. Transport. Res. Part D: Transport Environ. 22, 75–79.
Huang, J., Du, N., Liu, P., et al, 2013. An exploration of land use mix around residence and family commuting caused carbon emission: a case study of Wuhan
city in china. Urban Plan. Int. 28, 25–30 (in Chinese).
Hughes, B., Chambers, L., Lansdell, H., et al, 2004. Cities, area and transport energy. Road Transport Res. 13, 72.
Jahanshahi, K., Jin, Y., 2016a. The built environment typologies in the UK and their influences on travel behaviour: new evidence through latent
categorisation in structural equation modelling. Transport. Plan. Technol. 39, 59–77.
Jahanshahi, K., Jin, Y., 2016b. Trendbreaking influences of built form upon travel in UK cities: evidence from new quantifications of within-and between-
built-form variations. Transport. Res. Board 95th Annual Meeting. http://dx.doi.org/10.3141/2564-04.
494 X. Cao, W. Yang / Transportation Research Part D 52 (2017) 480–494

Jahanshahi, K., Jin, Y., Williams, I., 2015. Direct and indirect influences on employed adults’ travel in the UK: new insights from the National Travel Survey
data 2002–2010. Transport. Res. Part A: Policy Pract. 80, 288–306.
Jiang, Y., He, D., Zegras, C., 2011. Impact of neighborhood land use on residents travel energy consumption. Urban Transport China 9, 21–29 (in Chinese).
Kenworthy, J.R., Laube, F.B., 1996. Automobile dependence in cities: an international comparison of urban transport and land use patterns with implications
for sustainability. Environ. Impact Assess. Rev. 16, 279–308.
Ko, J., Park, D., Lim, H., et al, 2011. Who produces the most CO2 emissions for trips in the Seoul metropolis area? Transport. Res. Part D: Transport Environ.
16, 358–364.
Layman, C., Horner, M., 2010. Comparing methods for measuring excess commuting and jobs-housing balance: empirical analysis of land use changes.
Transport. Res. Rec.: J. Transport. Res. Board, 110–117.
Liu, Q., Wang, J., Chen, P., et al, 2016. How does parking interplay with the built environment and affect automobile commuting in high-density cities? A
case study in China. Urban Studies. 0042098016667040.
Lu, I.J., Lin, S.J., Lewis, C., 2007. Decomposition and decoupling effects of carbon dioxide emission from highway transportation in Taiwan, Germany, Japan
and South Korea. Energy Policy 35, 3226–3235.
Lu, X., Pas, E.I., 1999. Socio-demographics, activity participation and travel behavior. Transport. Res. Part A: Policy Pract. 33, 1–18.
Ma, J., Liu, Z., Chai, Y., 2015. The impact of urban form on CO2 emission from work and non-work trips: the case of Beijing, China. Habitat Int. 47, 1–10.
Ma, L., Dill, J., Mohr, C., 2014. The objective versus the perceived environment: what matters for bicycling? Transportation 41, 1135–1152.
Marsden, G., Rye, T., 2010. The governance of transport and climate change. J. Transp. Geogr. 18, 669–678.
Mendiola, L., González, P., Cebollada, À., 2014. The link between urban development and the modal split in commuting: the case of Biscay. J. Transp. Geogr.
37, 1–9.
Modarres, A., 2013. Commuting and energy consumption: toward an equitable transportation policy. J. Transp. Geogr. 33, 240–249.
Mokhtarian, P.L., Cao, X., 2008. Examining the impacts of residential self-selection on travel behavior: a focus on methodologies. Transport. Res. Part B:
Methodol. 42, 204–228.
Moniruzzaman, M., Páez, A., Habib, K.M.N., et al, 2013. Mode use and trip length of seniors in Montreal. J. Transp. Geogr. 30, 89–99.
Naess, P., 2010. Residential location, travel, and energy use in the Hangzhou Metropolitan Area. J. Transport Land Use 3, 27–59.
Newman, P.W., Kenworthy, J.R., 1989. Gasoline consumption and cities: a comparison of US cities with a global survey. J. Am. Plan. Assoc. 55, 24–37.
Newman, P.W., Kenworthy, J.R., 1996. The land use—transport connection: an overview. Land Use Policy 13, 1–22.
Newman, P., 2006. The environmental impact of cities. Environ. Urban. 18, 275–295.
Ou, J., Liu, X., Li, X., et al, 2013. Quantifying the relationship between urban forms and carbon emissions using panel data analysis. Landscape Ecol. 28, 1889–
1907.
Pinjari, A.R., Pendyala, R.M., Bhat, C.R., et al, 2007. Modeling residential sorting effects to understand the impact of the built environment on commute mode
choice. Transportation 34, 557–573.
Pinjari, A., Eluru, N., Bhat, C., et al, 2008. Joint model of choice of residential neighborhood and bicycle ownership: accounting for self-selection and
unobserved heterogeneity. Transport. Res. Rec.: J. Transport. Res. Board, 17–26.
Scheiner, J., 2010. Social inequalities in travel behaviour: trip distances in the context of residential self-selection and lifestyles. J. Transp. Geogr. 18, 679–
690.
Shen, Q., Chen, P., Pan, H., 2016. Factors affecting car ownership and mode choice in rail transit-supported suburbs of a large Chinese city. Transport. Res.
Part A: Policy Pract. 94, 31–44.
Shim, G.E., Rhee, S.M., Ahn, K.H., et al, 2006. The relationship between the characteristics of transportation energy consumption and urban form. Ann. Reg.
Sci. 40, 351–367.
Solomon, S., 2007. Climate Change 2007-the Physical Science Basis: Working Group I Contribution to the Fourth Assessment Report of the IPCC. Cambridge
University Press.
Sperling, D., Salon, D., 2011. Transportation in developing countries: an overview of greenhouse gas reduction strategies. Transport. Bus. J. 183.
Stevens, J.P., 2012. Applied Multivariate Statistics for the Social Sciences. Routledge.
Van Acker, V., Witlox, F., 2010. Car ownership as a mediating variable in car travel behaviour research using a structural equation modelling approach to
identify its dual relationship. J. Transp. Geogr. 18, 65–74.
Wang, D., Lin, T., 2013. Built environments, social environments, and activity-travel behavior: a case study of Hong Kong. J. Transp. Geogr. 31, 286–295.
Wang, D., Lin, T., 2014. Residential self-selection, built environment, and travel behavior in the Chinese context. J. Transport Land Use 7, 5–14.
Wang, Y., Yang, L., Han, S., et al, 2016. Urban CO2 emissions in Xi’an and Bangalore by commuters: implications for controlling urban transportation carbon
dioxide emissions in developing countries. Mitig. Adapt. Strat. Glob. Change, 1–27.
Wright, L., Fulton, L., 2005. Climate change mitigation and transport in developing nations. Transport Rev. 25, 691–717.
Wu, M., 2010. Structural Equation Modeling: The Operation and Application of AMOS. Chongqing University Press, Chongqing (in Chinese).
Xiao, Z., Chai, Y., Liu, Z., 2011. Quantitative distribution and related factors for household daily. Urban Stud. 18, 104–112 (in Chinese).
Yang, W., Li, T., Cao, X., 2015. Examining the impacts of socio-economic factors, urban form and transportation development on CO2 emissions from
transportation in China: a panel data analysis of China’s provinces. Habitat Int. 49, 212–220.
Zahabi, S.A.H., Miranda-Moreno, L., Patterson, Z., et al, 2012. Transportation greenhouse gas emissions and its relationship with urban form, transit
accessibility and emerging green technologies: a Montreal case study. Procedia – Soc. Behav. Sci. 54, 966–978.
Zhao, P., 2010. Sustainable urban expansion and transportation in a growing megacity: consequences of urban sprawl for mobility on the urban fringe of
Beijing. Habitat Int. 34, 236–243.
Zhao, P., Lü, B., Roo, G.D., 2011. Impact of the jobs-housing balance on urban commuting in Beijing in the transformation era. J. Transp. Geogr. 19, 59–69.
Zhou, B., Kockelman, K., 2008. Self-selection in home choice: use of treatment effects in evaluating relationship between built environment and travel
behavior. Transport. Res. Rec.: J. Transport. Res. Board, 54–61.

You might also like