How does the built environment at residential and work locations affect car ownership_ An application of cross-classified multilevel model

Contents lists available at ScienceDirect
Journal of Transport Geography
journal homepage wwwelseviercomlocatejtrangeo
How does the built environment at residential and work locations affect car
ownership An application of crossclassified multilevel model
Chuan Dingab Xinyu Caob⁎
a School of Transportation Science and Engineering Beijing Key Laboratory for Cooperative Vehicle Infrastructure System and Safety Control Beihang University Beijing
100191 China
b Beijing Advanced Innovation Center for Big Data and Brain Computing Beihang University Beijing 100191 China
ARTICLE INFO
Keywords
Auto ownership
Land use
Travel behavior
Spatial dependency
Random effect model
ABSTRACT
Although many studies investigate the connection between the residential built environment and car ownership
the literature offers limited evidence on the effect of work locations Using data from the Washington me
tropolitan area this study develops a crossclassified multilevel model to examine the influences of the built
environment at both residential and workplace locations on car ownership while controlling for spatial de
pendency arising from spatial aggregation We found that built environment characteristics at work locations
particularly bus stop density and employment density influence household car ownership They explain one
third of the total variation of car ownership across work locations The residential environment appears to
impose a stronger influence than the workplace environment Density diversity design transit access around
residences and distance from home to the city center affect car ownership
1 Introduction
During the past several decades a variety of land use and trans
portation policies have been implemented to mitigate the growth of car
ownership and car use (Jiang et al 2017 Araghi et al 2017 Huang
et al 2017) Car ownership has a dominant influence on mode choice
In the US when car ownership in a household increased from zero to 1
vehicle the share of trips by auto grew from 34 to 82 (Pucher and
Renne 2003) Accordingly planners are interested in understanding
the correlates of car ownership to alleviate the negative consequences
of car use on society such as greenhousegas emissions traffic safety
and congestion (Cao et al 2019 Li and Zhao 2017 Yang et al 2017)
As a mediumterm decision in a hierarchy of choices car ownership
mediates the relationship between residential and workplace location
choices (longterm decisions) and daily choices of travel mode and
activity destination (shortterm decisions) (BenAkiva and Atherton
1977 Van Acker and Witlox 2010) Because the built environment
influences the price of travel (including monetary temporal and psy
chological costs) (Boarnet and Crane 2001) location decisions and
associated built environment should affect car ownership and the
consumption of car travel Not surprisingly urban planners are in
trigued by the following set of questions To what extent do residential
and employment locations influence household car ownership choice
Which elements of the built environment have a significant impact The
answers to these questions will help policy makers and planners better
understand the effectiveness of using land use policies to influence car
ownership and hence car use In the literature however almost all
attention has been devoted to the residential environment (Maat and
Timmermans 2007 p 2) There is limited evidence on the connection
between car ownership and the built environment around work loca
tions
Furthermore in the studies of car ownership the built environment
is often measured at a clustered geographic scale such as census tract
census block group and traffic analysis zone (TAZ) The aggregation
leads to spatial dependence which may underestimate standard errors
of coefficients (Hong et al 2014 Bhat 2000) The literature offers
ample evidence on the advantages of using the multilevel modeling
approach to address the spatial analytic issue (Loo and Lam 2013 Ding
et al 2017a) However a limited number of studies examine the re
lationship between car ownership and the zonelevel built environment
at both residential location and work location which requires a novel
multilevel model to address the spatial dependence simultaneously
This study attempts to fill these two gaps Using data collected from
the Washington metropolitan area it employs a Bayesian crossclassi
fied multilevel ordered probit model to explore how built environment
factors at residential and work locations impact household car owner
ship This study enriches the literature in that (1) it examines the effect
of the built environment at work location on car ownership in addition
httpsdoiorg101016jjtrangeo201901012
Received 17 February 2018 Received in revised form 16 January 2019 Accepted 17 January 2019
⁎ Corresponding author
Email address cao@umnedu (X Cao)
Journal of Transport Geography 75 (2019) 37–45
09666923 © 2019 Elsevier Ltd All rights reserved
Tto the residential environment and (2) it conducts a crossclassified
spatial analysis to take the spatial dependence of built environment
measures at multiple locations into account
This paper is organized as follows The next section reviews the
literature on the relationship between the built environment and car
ownership Section 3 introduces the data and modeling approach
Section 4 presents the results The final section replicates the key
findings and offers implications for planning practice
2 Literature review
In this section we first discussed the importance of examining built
environment effects on car ownership and summarized empirical
findings Then we identified a critical gap in the literature overlooking
the built environment at work locations However filling this gap
brings about the issue of spatial dependency at multiple locations be
cause built environment variables at both residential and work loca
tions are often measured at an aggregate level This study employs a
crossclassified multilevel model to address the issue
To lower the externalities of automobiles urban planners hope to
reduce car use through land use strategies such as compact develop
ment and transitoriented development (Boarnet 2011 Jiang et al
2017) If cities bring destinations closer people may drive a shorter
distance even if their activities still depend on cars Moreover if the
built environment is conducive to travel by alternative modes of
transport and discourages driving people may shed one or more of their
cars without a substantial influence on their daily life This change will
have a fundamental influence on individuals＇ travel and activity choices
because car ownership critically impacts car use (Pucher and Renne
2003) Given the potential many studies examined the impact of the
built environment on car ownership (Zegras 2010 Yin and Sun 2017)
Car ownership is largely determined by individuals＇ longterm de
cisions of residential location and work location although the avail
ability of cars also influences where people live and work over time
(BenAkiva and Atherton 1977) Thus the built environment at these
locations is expected to be associated with car ownership To offer
nuanced guidance on land use planning many studies discerned which
elements of the residential environment affect car ownership Most of
them focused on the built environment at the neighborhood level and
concluded that built environment attributes such as density diversity
street connectivity and transit accessibility have significant effects on
car ownership (Potoglou and Kanaroglou 2008 Cao and Cao 2014
Van Acker and Witlox 2010) Table 1 summarizes significant built
environment variables tested in some studies Among the variables
residential density and employment density are seen as the most im
portant correlates of car ownership and use Density a fundamental
element of land use not only impacts car ownership itself but also
serves as a proxy for other land use elements that go along with the
density (such as parking supply) High population density at the re
sidential location is correlated with low car ownership (Chen et al
2008) As the mixture of jobs and households increases the likelihood
of owning cars also decreases (Zegras 2010) In areas with wellde
signed street connectivity generally measured by intersection density
and average block size people tend to have fewer cars because the
areas facilitate the use of nonmotorized modes of transport (Zegras
2010 Ding et al 2016) Meanwhile as transit accessibility increases a
shift from carbased to transitbased travel may occur and owning a car
becomes unnecessary (Potoglou and Kanaroglou 2008 Shen et al
2016 Chen et al 2008)
Some studies however found that in terms of effect size neigh
borhoodscale built environment has a weaker effect on car ownership
than regional urban form For example using data from the
Copenhagen Metropolitan Area Næss (2009) found that the effects of
the metropolitanscale built environment are more influential than the
neighborhoodscale built environment effects Especially he concluded
that distance to the city center has a substantial influence on car
ownership Van Acker and Witlox (2010) and Ding et al (2016) also
substantiated the strong effect of distance to the city center
Although the residential environment is well studied the influence
of work location on car ownership is mostly overlooked in the litera
ture If work location is not conducive to alternative means of transport
a car becomes a necessity This helps explain the dominance of cars in
commuting mode choice in the US Therefore the built environment
around work locations should have an influence on car ownership and
use Empirically several studies found that commuting mode choice are
associated with the built environment around both residences and
workplaces (Cervero 2002 Ding et al 2014a Shiftan and Barlach
2002 Maat and Timmermans 2009) In terms of car ownership Chen
et al (2008 p 293) found that higher job accessibility at work via
transit would decrease the likelihood of owning more cars in the New
York metropolitan area However the relationship is not always con
sistent because of different local contexts Using the data from residents
in transitsupported suburban neighborhoods in Shanghai Shen et al
(2016) found that the connection between work location and car
ownership is insignificant
The omission of built environment attributes around workplaces
might bias the estimated effects of built environment elements at re
sidential locations and hence offer erroneous implications to planning
Table 1
Variables of built environment selection in existing car ownership studies
Measurements Variable description Studies
Density Residential density Shen et al 2016 (500 m buffer) Zegras 2010 (traffic analysis zone level) Li et al 2010 (subdistrict level) Giuliano and
Dargay (2006) (census tract level) Bhat and Guo 2007 (traffic analysis zone level) Hess and Ong 2002 (census tract level)
Chen et al 2008 (census tract level) Ding et al 2017b (census block group level)
Employment density Bhat and Guo 2007 (traffic analysis zone level) Chen et al 2008 (census tract level) Ding et al 2016 (traffic analysis zone
level) Ding et al 2017b (census tract level)
Builtup index Van Acker and Witlox 2010 (census tract level) Li et al 2010 (city level) Jiang et al 2017 (500 m buffer)
Diversity Land use diversity Van Acker and Witlox 2010 (census tract level) Zegras 2010 (traffic analysis zone level) Bhat and Guo 2007 (traffic analysis
zone level)
Land use entropy Jiang et al 2017 (1000 m buffer) Ding et al 2016 (traffic analysis zone level) Potoglou and Kanaroglou 2008 (traffic analysis
zone level) Shen et al 2016 (500 m buffer) Ding et al 2017b (1600 m buffer)
Jobhousing balance Potoglou and Kanaroglou 2008 (traffic analysis zone level) Jiang et al 2017 (parcel level) Hess and Ong (2002) (traffic
analysis zone level)
DesignConnectivity Intersection density Zegras 2010 (traffic analysis zone level) Ding et al 2016 (traffic analysis zone level) Ding et al 2017b (400 m buffer)
Average block size Ding et al 2016 (traffic analysis zone level) Bhat and Guo 2007 (traffic analysis zone level)
Access to transit Distance to transit Van Acker and Witlox 2010 Huang et al 2016 Zhang et al 2017 Hess and Ong (2002) Chen et al 2008 Shen et al 2016
Ding et al 2017b
Bus stops density Potoglou and Kanaroglou 2008 (500 m buffer)
Destination accessibility Job accessibility Chen et al 2008 (census tract level) Ding et al 2017b (traffic analysis zone level)
Regional location Distance to CBD Van Acker and Witlox 2010 Ding et al 2016 Zegras 2010 Li et al 2010 Zhang et al 2017 Næss 2009
C Ding X Cao Journal of Transport Geography 75 (2019) 37–45
38practice Furthermore identifying the influential built environment
elements at work locations will inform the design and redevelopment of
employment centers Without this knowledge planners may erro
neously generalize the effect of built environment attributes at re
sidences to workplaces
In the studies exploring the effects of both residential and work
locations on car ownership built environment variables are often
measured at an aggregated level instead of at the individual level For
example Chen et al (2008) used residential and workplace land use
data aggregated at the census tract level to examine built environment
effects on car ownership Individuals who live in the same tract share
the same built environment values Furthermore they may be alike as
they are sorted into the same neighborhoods That is they are spatially
dependent (Kim and Wang 2015) If these observations are treated as
independently and identically distributed we are likely to under
estimate the standard error of a coefficient Accordingly we may falsely
reject the null hypothesis that built environment elements have no ef
fect on car ownership or overstate the significance of the test (Hong
et al 2014) Recently the multilevel modeling approach has been
widely applied to address spatial dependency in the field of urban
planning and transportation (Ma et al 2018 Wu and Hong 2017 Ding
et al 2017a Zhang et al 2012) When the built environment at both
residential location and work location is considered simultaneously
however traditional multilevel models are unable to handle the two
different types of aggregation An alternative multilevel model is de
sirable to examine the effects of the built environment at multiple lo
cations on car ownership (Anowar et al 2014) This study employs a
Bayesian crossclassified multilevel ordered probit model to address the
issue
3 Research design
31 Data and variables
This study investigates the connections between car ownership and
the built environment at work location as well as residential location
Car ownership data are obtained from the regional household travel
survey in the Washington metropolitan area (Fig 1) conducted in
2007–2008 the most recent data in the region A total of 8051 com
muters older than 16 years of age are included in the final dataset after
removing the samples with missing data Household car ownership is
measured by an ordered scale zero one two and three (or more) cars
Besides the data include a list of sociodemographic variables and three
workrelated variables (Table 2) In terms of the built environment we
chose five Ds namely land use density diversity design distance to
transit and distance from CBD (central business districts) The density
measures include population density and employment density The
entropy index measures land use diversity The average block size is the
indicator of land use design Access to transit is represent by metro
station availability and bus stop density These variables are measured
at the TAZ level based on the geographical information of residences
and workplaces
32 Conceptual model and modeling approach
Fig 2 presents the conceptual relationships among sociodemo
graphic variables commuting programs the built environment and
household car ownership According to the household choice hierarchy
proposed by BenAkiva and Atherton (1977) household car ownership
is affected by employment location and residential location Therefore
we hypothesize that the built environment at both workplace and re
sidential locations influence car ownership Furthermore the influences
of sociodemographic characteristics on these choices reflect house
holds＇ taste variation and constraints Since the key interest of this study
is in the influence of the built environment on car ownership the in
direct influences of sociodemographics on car ownership through the
built environment are not considered This specification will not affect
the effects of built environment variables Moreover because com
mutingrelated travel demand management programs affect car use
(Zhou 2012 Ding et al 2018b) these programs are also expected to
influence car ownership
This study measures the built environment at residential and
workplace locations at the TAZ level Because households in the same
TAZ share the same values of built environment variables we employ
multilevel models to address the spatial dependency Furthermore
since car ownership is affected by the built environment at two different
locations this study applies a Bayesian crossclassified multilevel or
dered probit model to test the conceptual framework
The crossclassified multilevel model for discrete responses using
the Markov Chain Monte Carlo (MCMC) Bayesian technique has been
used recently in mode choice studies (Easton and Ferrari 2015 Ding
et al 2014a) Its great advantage of accounting for the unobserved
spatial heterogeneities and spatial dependences in crossclassified
neighborhoods attracts more attention in the field of urban planning
and transportation Since car ownership is measured as an ordered re
sponse in this study the crossclassified multilevel ordered probit
model is used based on the Bayesian estimator
Assume that an individual q (q 12 … Q) living in residential
zone h (h 12… H) and employed in work zone w (w 12… W)
is associated with car ownership i (i 1 2… I) as shown in Fig 3 The
latent car ownership propensity Uqhwi
∗ can be presented as follows
⎧
⎨
⎪
⎩⎪
+ + + +
+
+
<≤
∗
−
∗
UφααβXε
αμYξ
αγZδ
UiifδUδ
qhwi hi wi qhwi qhwi
hi h h
wi w w
qhw i qhw i
Τ
Τ
Τ
1
(1)
where Xqhwi is a vector of individual variables Yh and Zw are a list of
built environment factors at residential and work locations respec
tively φ is the intercept αhi and αwi are random effects at residential
zone level and workplace zone level β μ and γ are estimated para
meters εqhwi is a standard normally distributed error ξh and δw are
normally distributed random errors with standard deviations σh and σw
representing the unobserved variations across residential zones and
workplace zones respectively The effects of sociodemographic char
acteristics and workrelated factors on car ownership are described in
the microlevel function of βX and residential and workplace built
environment factors are included in the macrolevel function with a
crossclassified framework by the two varying intercepts αhi and αwi
Conditional on εqhwi terms the probability Pqhwi
(δi−1 < Uqhwi
∗ < δi) of an individual q living in residential zone h
employed in work zone w and owning i number of cars can be pre
sented as follows
⎛
⎝
⎜
−− − −
++
⎞
⎠
⎟
− ⎛
⎝
⎜
−− − −
++
⎞
⎠
⎟
−
P δφμYγZβX
σσ
δφμYγZβX
σσ
Φ
1
Φ
1
qhwi
ihwqhwi
hw
ihwqhwi
hw
ΤΤ Τ
22
1 ΤΤ Τ
22 (2)
The total variance can be decomposed into three components
microlevel variance macrolevel variance across residential zones and
macrolevel variance across workplace zones The correlation between
individuals in the same residential zone or the same workplace zone can
be expressed using the intrazone correlation (or called intraclass
correlation (ICC) in terms of the common terminology of multilevel
models) as shown in Eqs (3) and (4) respectively which is a measure
of the spatial dependence across residential or workplace zones (Ding
et al 2014a Bhat and Zhao 2002 Bhat 2000)
++
σ
σσintra‐zone correlation 1
h
hwresidential zone
2
22 (3)
C Ding X Cao Journal of Transport Geography 75 (2019) 37–45
39 ++
σ
σσintra‐zone correlation 1
w
hwworkplace zone
2
22 (4)
where σh
2 and σw
2 are the macrolevel variances across residential zones
and workplace zones respectively If intrazone correlation is zero the
betweenzone variance is zero and all observations within zones are
independent Therefore a multilevel model reduces to a traditional
singlelevel model However if intrazone correlation is 1 all ob
servations within zones have the same attributes A rule of thumb is
that a multilevel model is appropriate when ICC is larger than 010
(Snijders and Bosker 2012)
Compared to traditional models used in the literature the cross
classified multilevel model has several advantages (Ding et al 2014a
Bhat and Zhao 2002 Bhat 2000) First it can address the spatial issues
of heterogeneity dependency and heteroscedasticity Second it ac
commodates crossclassification (ie residential zones and work zones)
in the context of a multilevel analysis of an ordered response variable
Third it can test the effects of independent variables at different levels
thereby partly addressing the multicollinearity issue It also has some
weaknesses A critical one is that complex model estimation restricts its
application in the field of land use and travel behavior
4 Results
The final sample consists of 8051 commuters of which 308 (38)
do not own a car 2167 (269) own one car 3420 (425) own two
cars and 2156 (268) own more than two cars Table 3 shows the
results of the crossclassified multilevel ordered probit model for car
ownership using the Bayesian estimator in Mplus We applied Eqs (3)
and (4) to calculate intrazone correlation based on the estimated
variances of the two random effects (Table 3) The intrazone correla
tion between any two individuals within the same residential zone is
17 the intrazone correlation between any two individuals within the
same workplace zone is 33 and the microlevel variance accounts for
the remaining 50 of the total variance Because both intrazone cor
relations are larger than 01 the crossclassified multilevel model is
appropriate for the data The adjusted R2s for the car ownership model
at individual residential and workplace levels are 0450 0650 and
0333 respectively Thus 65 of the residentialzone level variation of
car ownership is explained by the built environment variables at re
sidential location and 33 of the workzone level variation is explained
by the built environment variables at work locations Therefore the
built environment at residential locations have a larger explanatory
power than those at work locations
All but three sociodemographic and workrelated factors are sig
nificantly related to household car ownership Specifically household
license ratio and number of working adults are positively correlated
with car ownership Household income also has an expected impact on
car ownership Car ownership increases as people get elder Men and
White people are likely to own more cars than others These findings
are generally consistent with previous studies (Huang et al 2016 Van
Acker and Witlox 2010 Ding et al 2016) For the workrelated factors
having free parking at work locations is positively associated with car
ownership By contrast if employees are provided transitvanpooling
subsidies they are likely to have fewer cars These findings imply that
travel demand management strategies at workplaces can help decrease
car ownership and thence promote travel by alternative means
(Coleman 2000 Ding et al 2014b) Having multiple jobs working
with flexible hours and being employed by the government do not have
significant effects on car ownership
Fig 1 Locations of the regional household travel survey area
C Ding X Cao Journal of Transport Geography 75 (2019) 37–45
40The literature suggests that residential built environment char
acteristics are correlated with car ownership (Cao and Cao 2014
Zegras 2010 Potoglou and Kanaroglou 2008 Cao et al 2007) and
this relationship is also applicable to the Washington metropolitan area
After controlling for other variables population density and employ
ment density are negatively associated with car ownership Further
more the negative association between car ownership and land use mix
is consistent with the findings of Zegras (2010) for Santiago de Chile
and Potoglou and Kanaroglou (2008) for Hamilton Canada Small
average block size representing a pedestrianfriendly design is sig
nificantly related to fewer cars Meanwhile metro station availability
and bus stop density representing access to transit show significantly
negative effects on car ownership As expected distance to CBD is po
sitively associated with car ownership In terms of effect size distance
to CBD and population density have the largest and comparable influ
ence on car ownership among the five built environment variables In
particular a onestandarddeviation increase of population density will
lower the propensity of car ownership by 0285 units all else equal To
summarize these findings suggest that individuals living in compact
and mixeduse areas with pedestrianfriendly design and closer to
transit and downtown are able to manage their daily activities with
fewer cars
With regard to built environment characteristics at work locations
bus stop density employment density and distance from CBD show
significant effects on car ownership Bus stop density has a significantly
negative effect on car ownership suggesting that people working in a
place with more bus stops are likely to have fewer cars Metro station
availability also has a negative coefficient but it is insignificant at the
01 level The negative coefficient of employment density suggests that
those working in areas with more jobs per acre tend to own fewer cars
This finding is consistent with the studies on the relationship between
the built environment at work locations and mode choice employment
density at work locations is negatively associated with car use (Ding
et al 2014a Chen et al 2008 Zhang 2004) Workplace distance from
CBD is negatively associated with car ownership suggesting that those
working far away from CBD tend to own fewer cars This finding is
counterintuitive with unknown reasons Because population density
land use mix average block size and metro station availability are
insignificant bus stop density and employment density at workplaces
appear to play a dominant role in affecting car ownership
To assess the improvement of the crossclassified model we also ran
Table 2
Variable definitions and data summary of factors for car ownership
Variable name Variable description Mean St dev
Sociodemographic and workrelated factors at the individualhousehold level
Number of vehicles Number of vehicles in the household 205 108
Household license ratio Number of individuals with a driver＇s licensehousehold size 083 024
Household workers Number of workers in the household 177 069
Household gross income Income1 income is less than 40000 per year (1 yes dummy) 008 027
Income2 income is between 40000 and 125000 per year (1 yes dummy) 059 049
Income3 income is equal to or more than 125000 per year (1 yes dummy) 033 047
Age Age of the respondent in years 4433 1267
Male The respondent is male (1 yes dummy) 052 050
White The respondent is White (1 yes dummy) 074 044
Multiple jobs The respondent has more than one job (1 yes dummy) 006 025
Government employee The respondent works in a government agency (1 yes dummy) 038 049
Flexible work hours The respondent enjoys flexible work hours (1 yes dummy) 054 050
Free parking Employers provide free parking (1 yes dummy) 054 050
Transitvanpooling subsidies Employers provide subsidies for transitvanpooling (1 yes dummy) 020 040
Residential built environment factors at the TAZ level
Residential density Populationarea size (personsacre) 1058 1431
Employment density Employmentarea size (jobsacre) 675 2237
Land use mix (entropy) Mixture of residential service retail and other employment land use types 045 023
Average block size Average block size within the TAZ (sq mi) 012 019
Metro station availability Metro station is available within the TAZ (1 yes dummy) 008 027
Bus stop density Bus stoparea size (countsacre) 003 005
Distance from CBD Straight line distance from CBD (mile) 1763 1385
Workplace built environment factors at the TAZ level
Residential density Populationarea size (personsacre) 930 1590
Employment density Employmentarea size (jobsacre) 7968 14975
Land use mix (entropy) Mixture of residential service retail and other employment land use types 056 024
Average block size Average block size within the TAZ (sq mi) 010 022
Metro station availability Metro station is available within the TAZ (1 yes dummy) 020 040
Bus stop density Bus stoparea size (countsacre) 007 010
Distance from CBD Straight line distance from CBD (mile) 1328 1340
Note 8051 persons 1337 residential zones 1201 workplace zones
Sociodemographic
vairables
Built environment at
residential location
Built environment at
workplace location
Car ownership
Commuterelated
travel demand
management
Exogenous variables
Fig 2 Conceptual model describing the relationships between car ownership
and its determinant
Individual2 IndividualqIndividual1
TAZ1 TAZ2 TAZ3 TAZn
live work live work
Fig 3 Crossclassified multilevel membership of residential and work loca
tions
C Ding X Cao Journal of Transport Geography 75 (2019) 37–45
41a traditional singlelevel ordered probit model (Appendix Table A1) and
a traditional multilevel ordered probit model (Appendix Table A2) that
only captures the residential zone variation A comparison among the
three models shows some differences For example the effect of metro
station availability at work locations is significant at the 005 level in
the singlelevel model and it becomes significant at the 010 level in
the traditional multilevel model However it is insignificant at the 010
level in the Bayesian crossclassified multilevel model These differ
ences are consistent with Ding et al (2017a) and Hong et al (2014) the
traditional model is more likely to produce incorrect statistical in
ference because of the type I error In particular the traditional model
ignores the dependence of withinzone observations and hence under
estimates standard errors of coefficients Accordingly an insignificant
influence may become statistically significant The crossclassified
multilevel model provides an appropriate analytical framework to deal
with spatial dependence at both residential and workplace locations
5 Conclusions
Using data from the Washington metropolitan area this study ex
amines the influences of the built environment at both residential and
work locations on household car ownership after controlling for socio
demographic and workrelated factors Compared to most studies in the
literature it explores the influence of work locations and addresses
spatial dependency associated with both residential and work locations
using a crossclassified multilevel model
This study has a few limitations First the data were collected in
2007–2008 During the past decade transportation network companies
such as Uber and Lyft have become widely available Dockless bike
sharing programs also emerge in the Washington metropolitan area
These new services may influence commute mode choice However this
study cannot capture the effects of the evolving transportation systems
on car ownership Second this study does not explicitly address the
issue of residential selfselection individuals choose their residential
locations based on their demographics and propositions towards travel
and land use (Mokhtarian and Cao 2008) On the other hand it cap
tures the effect of sociodemographic characteristics and hence at least
partly accounts for the selfselection effect Some scholars argued that
controlling for sociodemographics could capture most of the selfse
lection effect (Brownstone and Golob 2009) Nevertheless this study
offers new insights to the literature
The built environment at residential locations plays an important
Table 3
Bayesian crossclassified multilevel ordered probit model for car ownership
Variables Estimate 95 credible interval 90 credible interval
Mean Posterior SD Lower 25 Upper 25 Lower 5 Upper 5
Sociodemographic and workrelated factors at the individualhousehold level
Household license ratio 0080 0010 0060 0099 0064 0099
Household workers 0553 0009 0535 0571 0538 0568
Household income1 −0146 0010 −0167 −0126 −0164 −0130
Household income3 0150 0012 0127 0173 0131 0169
Age 0040 0010 0020 0059 0024 0056
Male 0068 0010 0048 0088 0051 0084
White 0025 0011 0004 0047 0007 0044
Multiple jobs −0010# 0011 −0032 0011 −0028 0007
Government employee −0011# 0011 −0033 0011 −0029 0008
Flexible work hours −0011# 0011 −0032 0009 −0029 0005
Free parking 0073 0011 0052 0094 0055 0092
Transitvanpooling subsidies −0060 0012 −0083 −0038 −0079 −0041
Residential built environment factors at the TAZ level
Residential density −0285 0031 −0347 −0229 −0339 −0239
Employment density −0090 0033 −0156 −0024 −0144 −0034
Land use mix −0128 0026 −0178 −0078 −0169 −0085
Average block size 0099 0038 0018 0167 0030 0157
Metro station availability −0081 0028 −0136 −0027 −0128 −0036
Bus stop density −0163 0038 −0234 −0081 −0220 −0095
Distance from CBD 0329 0036 0262 0405 0271 0392
Workplace built environment factors at the TAZ level
Residential density −0021# 0083 −0166 0166 −0141 0135
Employment density −0143 0065 −0266 −0011 −0244 −0027
Land use mix 0043# 0133 −0152 0424 −0121 0331
Average block size −0021# 0122 −0247 0251 −0208 0197
Metro station availability −0114# 0080 −0295 0029 −0254 0007
Bus stop density −0191 0091 −0372 −0010 −0342 −0041
Distance from CBD −0606 0139 −0839 −0292 −0811 −0348
Model threshold values
τ1 −0401 0090 −0536 −0292 −0512 −0300
τ2 1232 0096 1115 1340 1142 1331
τ3 2600 0105 2482 2709 2507 2695
Spatial dependence parameter across zones
σh
2 0351 0027 0298 0403 0306 0397
σw
2 0667 0118 0414 0858 0460 0839
Model fit information
R2 at individual level 0450 0009 0432 0469 0440 0464
R2 at residential zone level 0650 0027 0597 0702 0601 0689
R2 at workplace zone level 0333 0118 0142 0586 0164 0571
Note 8051 persons 1337 residential zones 1201 workplace zones The coefficients of all explanatory variables are standardized A variable is statistically significant
at the 95 level if the 95 credible interval does not include zero # the variable is insignificant at the 01 level All other variables are significant at the 005 level
C Ding X Cao Journal of Transport Geography 75 (2019) 37–45
42role in determining car ownership It accounts for 64 of the total
variation of car ownership across residential zones In particular people
living in dense mixeduse and walkable areas tend to have fewer cars
and individual living far away from CBD tend to own more cars These
findings suggest that planning strategies such as compact development
pedestrianoriented development and urban growth boundary have the
potential to lower car ownership Second some built environment
elements at work locations also affect car ownership They collectively
explain 23 of the workzone level variation of car ownership Among
the five elements tested employment density and proximity to CBD are
significant whereas residential density land use mix and average block
size are insignificant
The residential built environment appears to be more relevant to car
ownership than the workplace built environment As presented above
the residential environment is associated with a larger Rsquared than
the workplace environment Furthermore the residential environment
has more significant land use variables than the workplace environ
ment Moreover as the model reports standardized coefficients built
environment variables at residences collectively have a larger influence
than those around work locations This result is consistent with our
expectation because one anchor of most trips (including commuting) is
home
Besides workrelated factors are significantly associated with car
ownership Specifically free parking at work locations tends to increase
car ownership On the other hand if employers provide transit or
vanpool subsidies employees tend to opt for fewer cars Accordingly
transportation management organizations could advocate for travel
demand management strategies which can play an active role in re
ducing auto travel hence lowering employee car ownership
Planners are interested in using land use policies to shape travel
behavior Since car ownership is a mediating variable between the built
environment and car use (Van Acker and Witlox 2010 Ding et al
2018a) future studies should examine the relationships among the built
environment at residential and workplace locations car ownership and
commute mode choice simultaneously using crossclassified structural
equations models In this way we could have a comprehensive un
derstanding of built environment effects on car ownership and use
Acknowledgements
This work is supported by the National Natural Science Foundation
of China (71874010 61773040 and U1764265) and Young Elite
Scientist Sponsorship Program by the China Association for Science and
Technology (2017QNRC001)
Appendix A Appendix
Table A1
Traditional singlelevel ordered probit model for car ownership
Variables Estimate 95 credible interval 90 credible interval
Mean Posterior SD Lower 25 Upper 25 Lower 5 Upper 5
Sociodemographic and workrelated factors at the individualhousehold level
Household license ratio 0066 0009 0049 0083 0052 0080
Household workers 0458 0008 0442 0474 0444 0471
Household income1 −0117 0009 −0134 −0100 −0131 −0103
Household income3 0124 0009 0106 0143 0109 0140
Age 0047 0008 0030 0063 0033 0060
Male 0059 0009 0042 0076 0045 0073
White 0019 0009 0002 0037 0005 0034
Multiple jobs −0004# 0009 −0021 0014 −0018 0010
Government employee −0005# 0009 −0023 0013 −0020 0010
Flexible work hours −0010# 0009 −0027 0007 −0024 0004
Free parking 0063 0009 0045 0081 0049 0078
Transitvanpooling subsidies −0057 0010 −0075 −0037 −0073 −0040
Residential built environment factors at TAZ level
Residential density −0159 0012 −0184 −0134 −0180 −0139
Employment density −0032 0010 −0053 −0012 −0050 −0015
Land use mixture −0056 0009 −0074 −0038 −0071 −0040
Average block size 0037 0011 0015 0059 0019 0056
Metro station availability −0032 0009 −0050 −0014 −0048 −0017
Bus stop density −0073 0013 −0098 −0049 −0095 −0053
Distance from CBD 0189 0017 0155 0222 0161 0216
Workplace built environment factors at TAZ level
Residential density −0004# 0009 −0022 0014 −0019 0012
Employment density −0034 0012 −0057 −0010 −0053 −0014
Land use mixture 0013# 0009 −0004 0030 −0002 0027
Average block size 0001# 0010 −0019 0021 −0016 0018
Metro station availability −0024 0010 −0042 −0005 −0039 −0008
Bus stop density −0029 0012 −0053 −0004 −0050 −0008
Distance from CBD −0080 0016 −0112 −0049 −0108 −0054
Model threshold values
τ1 −0218 0046 −0283 −0099 −0271 −0117
τ2 1110 0044 1045 1221 1058 1205
τ3 2216 0042 2153 2322 2165 2309
Model fit information
R2 0567 0008 0553 0583 0562 0579
Note 8051 persons 1337 residential zones 1201 workplace zones All coefficients of explanatory variables are standardized A variable is statistically significant at
the 95 level if the 95 credible interval does not include zero
# The variable is insignificant at the 005 level as well as at the 01 level All other variables are significant at the 005 level
C Ding X Cao Journal of Transport Geography 75 (2019) 37–45
43Table A2
Residential zonebased Bayesian multilevel ordered probit model for car ownership
Variables Estimate 95 credible interval 90 credible interval
Mean Posterior SD Lower 25 Upper 25 Lower 5 Upper 5
Sociodemographic and workrelated factors at the individualhousehold level
Household license ratio 0079 0011 0057 0099 0061 0096
Household workers 0546 0010 0525 0566 0529 0562
Household income1 −0147 0011 −0169 −0127 −0166 −0131
Household income3 0149 0012 0126 0171 0129 0167
Age 0037 0011 0014 0058 0018 0055
Male 0068 0011 0048 0088 0051 0085
White 0024 0011 0001 0046 0006 0043
Multiple jobs −0010# 0010 −0031 0009 −0028 0007
Government employee −0012# 0011 −0033 0010 −0030 0007
Flexible work hours −0013# 0011 −0034 0008 −0031 0005
Free parking 0073 0011 0050 0095 0055 0092
Transitvanpooling subsidies −0061 0012 −0085 −0037 −0081 −0040
Residential built environment factors at the TAZ level
Residential density −0293 0032 −0353 −0231 −0344 −0240
Employment density −0090 0033 −0152 −0024 −0142 −0033
Land use mix −0131 0027 −0188 −0079 −0178 −0086
Average block size 0095 0036 0022 0167 0033 0154
Metro station −0082 0028 −0138 −0028 −0128 −0036
Bus stop density −0163 0039 −0243 −0089 −0229 −0101
Distance from CBD 0322 0037 0245 0394 0259 0380
Workplace built environment factors at the TAZ level
Residential density −0001# 0011 −0024 0020 −0021 0017
Employment density −0037 0014 −0064 −0007 −0059 −0013
Land use mix 0008# 0011 −0014 0030 −0010 0026
Average block size −0002# 0012 −0026 0021 −0023 0018
Metro station availability −0022⁎ 0012 −0046 0001 −0041 −0003
Bus stop density −0033 0015 −0062 −0002 −0058 −0008
Distance from CBD −0078 0020 −0119 −0040 −0112 −0045
Model threshold values
τ1 −0434 0092 −0662 −0317 −0635 −0330
τ2 1202 0086 0995 1310 1012 1300
τ3 2557 0084 2365 2677 2383 2663
Spatial dependence parameter across zones
σh
2 0350 0028 0298 0407 0305 0397
Model fit information
R2 at individual level 0445 0010 0424 0464 0438 0453
R2 at residential zone level 0650 0028 0593 0702 0602 0687
Note 8051 persons 1337 residential zones 1201 workplace zones The coefficients of all explanatory variables are standardized A variable is statistically significant
at the 95 level if the 95 credible interval does not include zero
# The variable is insignificant at the 01 level
⁎ The variable is insignificant at the 005 level but significant at the 01 level All other variables are significant at the 005 level
References
Anowar S Eluru N MirandaMoreno LF 2014 Alternative modeling approaches
used for examining automobile ownership a comprehensive review Transp Rev 34
(4) 441–473
Araghi Y Kroesen M van Wee B 2017 Identifying reasons for historic car ownership
and use and policy implications an explorative latent class analysis Transp Policy
56 12–18
BenAkiva M Atherton TJ 1977 Methodology for shortrange travel demand pre
dictions analysis of carpooling incentives Journal of Transport Economics and
Policy 11 224–261
Bhat C 2000 A multilevel crossclassified model for discrete response variables
Transp Res B Methodol 34 (7) 567–582
Bhat CR Guo J 2007 A comprehensive analysis of built environment characteristics
on household residential choice and auto ownership levels Transp Res B Methodol
41 (5) 506–526
Bhat C Zhao H 2002 The spatial analysis of activity stop generation Transp Res B
Methodol 36 (6) 557–575
Boarnet MG 2011 A broader context for land use and travel behavior and a research
agenda J Am Plan Assoc 77 197–213
Boarnet MG Crane R 2001 Travel by Design The Influence of Urban Form on Travel
Oxford University Press New York
Brownstone D Golob TF 2009 The impact of residential density on vehicle usage and
energy consumption J Urban Econ 65 91–98
Cao J Cao X 2014 The impacts of LRT neighbourhood characteristics and self
selection on auto ownership evidence from MinneapolisSt Paul Urban Stud 51
(10) 2068–2087
Cao X Mokhtarian PL Handy SL 2007 Crosssectional and quasipanel explorations
of the connection between the built environment and auto ownership Environ Plan A
39 (4) 830–847
Cao X Næss P Wolday F 2019 Examining the effects of the built environment on
auto ownership in two Norwegian urban regions Transp Res Part D Transp
Environ 67 464–474
Cervero R 2002 Built environment and mode choice toward a normative framework
Transp Res Part D Transp Environ 7 (4) 265–284
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 (3)
285–299
Coleman C 2000 Green commuter plans and the small employer an investigation into
the attitudes and policy of the small employer towards staff travel and green com
muter plans Transp Policy 7 (2) 139–148
Ding C Lin Y Liu C 2014a Exploring the influence of built environment on tour
based commuter mode choice a crossclassified multilevel modeling approach
Transp Res Part D Transp Environ 32 230–238
Ding C Liu C Lin Y Wang Y 2014b The impact of employer attitude to green
commuting plans on reducing car driving a mixed method analysis Promet Traffic&
Transportation 26 (2) 109–119
Ding C Wang Y Yang J Liu C Lin Y 2016 Spatial heterogeneous impact of built
environment on household auto ownership levels evidence from analysis at traffic
analysis zone scales Transportation Letters 8 (1) 26–34
Ding C Mishra S Lu G Yang J Liu C 2017a Influences of built environment
C Ding X Cao Journal of Transport Geography 75 (2019) 37–45
44characteristics and individual factors on commuting distance a multilevel mixture
hazard modeling approach Transp Res Part D Transp Environ 51 314–325
Ding C Wang D Liu C Zhang Y Yang J 2017b Exploring the influence of built
environment on travel mode choice considering the mediating effects of car owner
ship and travel distance Transp Res A Policy Pract 100 65–80
Ding C Wang Y Tang T Mishra S Liu C 2018a Joint analysis of the spatial im
pacts of built environment on car ownership and travel mode choice Transp Res
Part D Transp Environ 60 28–40
Ding C Cao X Wang Y 2018b Synergistic effects of the built environment and
commuting programs on commute mode choice Transp Res A Policy Pract 118
104–118
Easton S Ferrari E 2015 Children＇s travel to schoolthe interaction of individuals
neighbourhood and school factors Transp Policy 2015 (44) 9–18
Giuliano G Dargay J 2006 Car ownership travel and land use a comparison of the US
and Great Britain Transp Res A Policy Pract 40 (2) 106–124
Hess DB Ong PM 2002 Traditional neighborhoods and auto ownership
Transportation Research Record Journal of the Transportation Research Board 1805
35–44
Hong J Shen Q Zhang L 2014 How do builtenvironment factors affect travel be
havior A spatial analysis at different geographic scales Transportation 41 (3)
419–440
Huang X Cao X Cao J 2016 The association between transit access and auto
ownership evidence from Guangzhou China Transp Plan Technol 39 (3)
269–283
Huang X Cao X Yin J Cao X 2017 Effects of metro transit on the ownership of
mobility instruments in Xi＇an China Transp Res Part D Transp Environ 52
495–505
Jiang Y Gu P Chen Y He D Mao Q 2017 Influence of land use and street
characteristics on car ownership and use evidence from Jinan China Transp Res
Part D Transp Environ 52 518–534
Kim C Wang S 2015 Empirical examination of neighborhood context of individual
travel behaviors Appl Geogr 60 230–239
Li S Zhao P 2017 Exploring car ownership and car use in neighborhoods near metro
stations in Beijing does the neighborhood built environment matter Transp Res
Part D Transp Environ 56 1–17
Li J Walker JL Srinivasan S Anderson WP 2010 Modeling private car ownership
in China investigation of urban form impact across megacities Transportation
Research Record Journal of the Transportation Research Board 2193 76–84
Loo BPY Lam WWY 2013 A multilevel investigation of differential individual
mobility of working couples with children a case study of Hong Kong
Transportmetrica A Transport Science 9 (7) 629–652
Ma J Chen Y Dong G 2018 Flexible spatial multilevel modeling of neighborhood
satisfaction in Beijing Prof Geogr 70 (1) 11–21
Maat K Timmermans HJP 2007 Household car ownership in relation to residential
and work location In Transportation Research Board 86th Annual Meeting CD
ROM Washington DC
Maat K Timmermans HJP 2009 Influence of the residential and work environment
on car use in dualearner households Transp Res A Policy Pract 43 654–664
Mokhtarian PL Cao X 2008 Examining the impacts of residential selfselection on
travel behavior a focus on methodologies Transp Res B Methodol 42 204–228
Næss P 2009 Residential selfselection and appropriate control variables in land use
travel studies Transp Rev 29 (3) 293–324
Potoglou D Kanaroglou PS 2008 Modelling car ownership in urban areas a case
study of Hamilton Canada J Transp Geogr 16 (1) 42–54
Pucher J Renne JL 2003 Socioeconomics of urban travel evidence from the 2001
NHTS Transp Q 57 (3) 49–77
Shen Q Chen P Pan H 2016 Factors affecting car ownership and mode choice in rail
transitsupported suburbs of a large Chinese city Transp Res A Policy Pract 94
31–44
Shiftan Y Barlach Y 2002 Effect of employment site characteristics on commute mode
choice Transportation Research Record Journal of the Transportation Research
Board 1781 19–25
Snijders TAB Bosker RJ 2012 Multilevel Analysis An Introduction to Basic and
Advanced Multilevel Modeling second edition Sage Publishers London
Van Acker V Witlox F 2010 Car ownership as a mediating variable in car travel
behaviour research using a structural equation modeling approach to identify its dual
relationship J Transp Geogr 18 65–74
Wu W Hong J 2017 Does public transit improvement affect commuting behavior in
Beijing China A spatial multilevel approach Transp Res Part D Transp Environ
52 (B) 471–479
Yang Z Jia P Liu W Yin H 2017 Car ownership and urban development in Chinese
cities a panel data analysis J Transp Geogr 58 127–134
Yin C Sun B 2017 Disentangling the effects of the built environment on car owner
ship a multilevel analysis of Chinese cities Cities 2017 (published online)
Zegras C 2010 The built environment and motor vehicle ownership and use evidence
from Santiago de Chile Urban Stud 47 (8) 1793–1817
Zhang M 2004 The role of land use in travel mode choice evidence from Boston and
Hong Kong J Am Plan Assoc 70 (3) 344–360
Zhang L Hong J Nasri A Shen Q 2012 How built environment affects travel be
havior a comparative analysis of the connections between land use and vehicle miles
traveled in US cities Journal of Transport and Land Use 5 (3) 40–52
Zhang Y Zheng S Sun C Wang R 2017 Does subway proximity discourage auto
mobility Evidence from Beijing Transp Res Part D Transp Environ 52 (B)
506–517
Zhou J 2012 Sustainable commute in a cardominant city factors affecting alternative
mode choices among university students Transp Res A Policy Pract 46 (7)
1013–1029
C Ding X Cao Journal of Transport Geography 75 (2019) 37–45
45

《香当网》用户分享的内容，不代表《香当网》观点或立场，请自行判断内容的真实性和可靠性！
该内容是文档的文本内容，更好的格式请下载文档