Examining The Effects of The Built Environment On Auto Ownershipin Two Norwegian Urban Regions

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Transportation Research Part D 67 (2019) 464–474

Contents lists available at ScienceDirect

Transportation Research Part D


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

Examining the effects of the built environment on auto ownership


T
in two Norwegian urban regions

Xinyu (Jason) Caoa, Petter Næssb, , Fitwi Woldayb
a
Humphrey School of Public Affairs, University of Minnesota, Twin Cities, USA
b
Department of Urban and Regional Planning, Norwegian University of Life Sciences, Ås, Norway

A R T IC LE I N F O ABS TRA CT

Keywords: Based on a study in the Oslo and Stavanger metropolitan areas in Norway combining quantitative
Car ownership and qualitative methods, this paper sheds light on influences of built environment characteristics
Land use on auto ownership. In both metropolitan areas, the distance from the dwelling to the main city
Longitudinal design center, a measure of regional destination accessibility, has an important effect on auto owner-
Causality
ship. A quasi-longitudinal analysis shows that outward residential relocation tends to increase
Qualitative interview
auto ownership and inward relocation to reduce it. Neighborhood density variables including
population and employment also exert important influences. The analysis of qualitative interview
material substantiates the causal mechanism about the effects of the built environment char-
acteristics on auto ownership.

1. Introduction

During the past several decades, many policy proposals based on land use, transport alternatives, market-based incentives/
disincentives or a combination of them have been put forth to curb auto ownership and use as well as bring about sustainable means
of mobility (Dohyung et al., 2017; Hu et al., 2018; Wolday et al., 2018). Norway is one of the countries where such policies have been
in place for a long time, with a particular focus on coordinated land use and transport planning (Norwegian-Government, 1993,
2014). Moreover, since 2012, several central-government policy documents (including the National Transport Plan) have stated as a
goal that growth in travel demand in the larger Norwegian urban regions is to be captured by public and non-motorized transport
(Norwegian-Government, 2017). Given the predicted population growth, this will require a reduction in per capita driving in these
regions by 20–25 percent over the next couple of decades. The focus of this study is on the effects of local- and metropolitan-scale
built environment (BE) characteristics on auto ownership. One important drive for such a study goes along the line ‘a car owned is a
car driven’, as there is a strong positive influence of the number of vehicles in a household on total vehicle kilometers traveled (VKT)
(Chen et al., 2008; Van Acker and Witlox, 2010). At the same time, car owners show considerable differences in VKT, reflecting
differences in residential location and job location (Næss, 2005; Zegras, 2010).
This study employs cross-sectional and quasi-longitudinal survey data and qualitative interview material collected from two
metropolitan areas of different sizes in Norway in 2015 to examine the influence of residential location in the region and the BE at the
neighborhood level on auto ownership. This study enriches the literature in that (1) it examines the influence of the BE on auto
ownership between regions of different scales and with different center structures; and (2) it integrates cross-sectional analysis, quasi-
longitudinal analysis, and qualitative analysis to illustrate the causal relationship between the BE and auto ownership.


Corresponding author at: P.O. Box 5003 NMBU, N-1432 Ås, Norway.
E-mail addresses: cao@umn.edu (X. (Jason) Cao), petter.nass@nmbu.no (P. Næss), hamde.fitwi.wolday@nmbu.no (F. Wolday).

https://doi.org/10.1016/j.trd.2018.12.020

Available online 03 January 2019


1361-9209/ © 2019 Elsevier Ltd. All rights reserved.
X. (Jason) Cao et al. Transportation Research Part D 67 (2019) 464–474

This paper is organized as follows. The next section reviews the literature on relationships between the BE and travel behavior in
general, and auto ownership in particular and identifies the gaps. Section 3 introduces the data and variables. Section 4 present the
results of quantitative and qualitative analyses. The concluding section reiterates the key results and discusses their implications.

2. Literature review

Land-use policies such as transit-oriented development, densification and mixed-use development are favored policy interventions
aimed at reducing car dependence (Boarnet, 2011; Handy, 1996; Jiang et al., 2017). Change in the BE might lead to reduction in
amount of car travel in at least two ways. First, the BE may create more conducive conditions for alternative means of transport and
less conducive conditions for car travel, which can then induce a shift from auto-based to non-motorized and/or transit-based travel.
This may lead to a reduction in car ownership. Second, the BE that promotes higher population density and land-use mix may reduce
the overall amount of travel (VKT) by bringing origins and destinations closer together, although car ownership may not be ne-
cessarily reduced. Out of the two scenarios, the first one is more profound as it implies a strong shift in travel preferences.
Socioeconomic and demographic characteristics have long been recognized as the dominant predictors of auto ownership. For
example, Zegras (2010) found that owning the first car is more of an economic matter and BE attributes contribute when deciding on
additional vehicles. That being said, because of high affluence, many households own one or more cars to gain flexibility in their
mobility pattern: they may not need the car for daily travel but own it for some specific circumstances such as vacation.
The literature also shows that both local- and metropolitan-scale BE characteristics influence auto ownership. Neighborhood-scale
BE attributes such as density, diversity and street connectivity are consistently reported to be significantly related to auto ownership
(Chen et al., 2008; Ding et al., 2018; Van Acker and Witlox, 2010). In contrast, some researchers downplay the strength and hence the
significance of neighborhood attributes on auto ownership. For example, Næss (2009) found that neighborhood-scale and meso-scale
urban structural conditions had insignificant effects on auto ownership while distance to the city center had a substantial influence.
Differences in reported result notwithstanding, though, what is clearly evident from the literature is that the focus is largely
skewed towards local or neighborhood-scale BE attributes (e.g., Bhat and Guo, 2007; Cao and Cao, 2014; Macfarlane et al., 2015).
This is at least partly because land use planning is implemented by municipal governments instead of regional government in the U.S.
and neighborhoods are the testbeds of urban planners. This was for a long time the case in Norway as well, but since the 1990s a
number of regional plans (partial county plans, and in some cases regional plans across county borders) for land use and transport
have been adopted. Most recently, such a plan was adopted for the counties of Oslo and Akershus in 2016. Anyway, when exploring
the effect on mode choice by trip purpose, the share of car use appears to be larger when the distance to destination is longer such as
work commutes, which are typically carried out outside the local area. Moreover, the causal order of metropolitan-scale versus
neighborhood-scale built environment characteristics is such that metropolitan-scale BE attributes precede the local-scale attributes
(at least in Norway). The decision about where to build comes before the decision about how to build. For several economic (Alonso,
1960) and cultural (Fishman, 1996) reasons, higher densities are more likely to be accepted at central areas than at peripheral
locations. Deciding whether to densify or expand the city outward also largely determines whether to build apartments or single-
family houses. Therefore, distance to the city center has an indirect influence on travel behavior (and on auto ownership) via the

Fig. 1. Assumed causal relationships between different built environment characteristics and travel behavior. Characteristics included in the sta-
tistical analyses of this paper are shown with bold outline. (In the cross-sectional analyses, jobs-housing balance is also included as a variable. This
variable is not shown in the figure, as it is completely determined by the local-area job density and population density.)

465
X. (Jason) Cao et al. Transportation Research Part D 67 (2019) 464–474

neighborhood-scale characteristics, in addition to its direct effect (Fig. 1).


We contend that jointly examining local as well as metropolitan-scale BE variables can give a better understanding of BE effects on
car ownership decisions. Out of the few researchers that opted for such a line of enquiry, Næss (2009) and Zegras (2010) found that
metropolitan-scale effects are more influential than neighborhood-scale BE effects. Researchers such as Ding et al. (2016) and Van
Acker and Witlox (2010) also found that distance to the city center (CBD) has a strong influence on auto ownership.
In this paper, we examine the local- vs metropolitan-scale influences by investigating the effect on auto ownership of residential
proximity to three center hierarchies: the city center, second-order centers, and local centers, as well as neighborhood BE elements.
This will enable us to identify geographic scales where BE attributes exert the strongest influence on auto ownership. Besides, it will
also add robustness to our analysis by minimizing what Kwan and Weber (2008) describe as the Uncertain Geographic Context
Problem (UGCoP). The UGCoP refers to the situations where the delimitation of a geographic area from which travel data are
obtained may not capture the travel behavioral implications, for example, when the action space of an individual is wider than the
geography of the data source. This is especially important when considering motorized transport (public transit and personal ve-
hicles) which has a wider geographic reach.
Recently, many studies have explored the causal relationship between the BE and travel behavior (including auto ownership
(Jiang et al., 2017; Zhang et al., 2017)). To infer causality between two phenomena, at least four criteria ought to be met (Singleton
and Straits, 2010): causal mechanism, statistical association, non-spuriousness, and time precedence. Many statistical methods tackle
statistical association and non-spuriousness fairly well, but time precedence often proves demanding specially in terms of data
availability. This is especially true as many travel behavior studies depend on cross-sectional or repeated cross-sectional travel
surveys that do not include a temporal continuity across individuals. Mokhtarian and Cao (2008) conclude that longitudinal design
tends to have the strongest merits on causality inferences because of its capacity in addressing time precedence. Except for a few
studies (e.g., Cao and Cao, 2014; Huang et al., 2017), longitudinal or quasi-longitudinal designs are very limited in the literature of
BE effects on auto ownership. The data utilized in this research involves cohorts of movers and therefore will enable us to apply quasi-
longitudinal analysis. More importantly, this study offers evidence for causal mechanism by adding qualitative analysis of inter-
viewees and some examples of car dependency in different residential location situations. In fact, an integration of quantitative and
qualitative analyses is desirable for a better understanding of travel behavior (Clifton and Handy, 2003; Næss, 2015).

3. Data and variables

The data utilized in this study were obtained mainly from a survey based on a self-administered web questionnaire, and partly
from 33 qualitative in-depth interviews of survey respondents. The survey and interviews covered two regions. The Oslo metropolitan
area is the largest one in Norway. The continuous built up area in this region is home to 988 875 inhabitants1. The Stavanger
metropolitan area (North Jæren region) is a medium-sized one on the western coast of Norway with a population of 220 9432 in the
continuous built up area. Both regions extend farther than the continuous built up areas. The Oslo metropolitan area has a pre-
dominantly monocentric/hierarchical urban structure, with downtown Oslo as the main center, supplemented with a limited number
of second-order centers and many local centers. The Stavanger metropolitan area has a much more polycentric structure, reflecting
that the two previously separate cities of Stavanger and Sandnes have since several decades ago made up one continuous urbanized
area. In addition, a new employment center (called Forus), situated between the city centers of Stavanger and Sandnes but not along
the main transport corridor between the two centers, has developed since the 1970s. Forus now has the largest concentration of
workplaces, but has few other center functions. Downtown Stavanger is still the main metropolitan center, with nearly as many
workplaces in its vicinity as Forus and with a large number of stores, restaurants, public offices and cultural facilities. Thus, although
both city regions include a dominant center as well as second-order centers, the strength of the second-order centers compared to the
main city center is stronger in Stavanger than in Oslo, especially regarding workplaces. In our analyses, the distance from the
dwelling to the closest among two second-order centers were included in Stavanger and to the closest among eleven such centers in
Oslo.
Since this study is interested in change in travel behavior of recent movers, a total of 112 newly-built residential areas (73 in the
Oslo and 39 in the Stavanger region) were included in the survey, on top of a geographically stratified random sample from the rest of
the residential building stock in each of the two regions. Newly-built housing projects were first identified via a combination of
information from real-estate developers, real-estate brokers, and planning and building departments in the affected municipalities.
Individual addresses of newly-built homes were then identified using sources such as municipalities, development companies and
Yellow Pages. Finally, the names of residents in the newly-built housing projects were identified from the national population
register. The rest of the respondents were drawn from the national population register. Overall, each region included about 15,000
individual addresses.
An invitation letter was sent to the addressees in April 2015. The letter explained the purpose of the study and offered an
instruction on how to access the web survey. The letter also stated that a respondent could request a paper-based survey and a few
respondents did request. The incentive for the survey was a lottery for a gift card of 10,000 NOK. The questionnaire was pretested by
five prospective respondents and by colleagues. The survey was revised based on pretesters’ feedback.
The survey generated a total of 3365 observations: 1992 residents in Oslo and 1373 in Stavanger. The average response rate is

1
Statistics Norway, Population and land area in urban settlements, 1 January 2017.
2
Ibid.

466
X. (Jason) Cao et al.

Table 1
Comparison of demographic and socioeconomic characteristics of the survey respondents in the two urban regions, the participants in qualitative interviews, as well as averages for the groups of counties
to which each urban region belongs.
Variables Respondents of Oslo survey Respondents of Stavanger Interviewees Inhabitants of the counties Oslo and Inhabitants of the county Rogaland
(N = 1992) survey (N = 1373) (N = 33) Akershus (including Oslo metro area) (including Stavanger metro area)

Average number of persons per household 2.29 2.49 2.79 1.94 2.29
Average number of children aged 0–6 years 0.23 0.24 0.55 0.15 0.18
per household
Average number of children aged 7–17 years 0.35 0.43 0.36 0.13 0.15
per household

467
Average age of respondents/interviewees (all 49.3 46.4 49.4 45.5 45.5
aged 16 or more)
Gender (proportion female) 50.6% 51.6% 54.5% 50.3% 49.2%
Proportion of workforce participants among 68% 72% 73% 81% 83%
respondents/interviewees
Average annual household income (1000 1018 1053 1110 812 884
NOK)
Proportion with education at master level or 40% 34% 47% 16% 9%
higher
Average number of cars per household 1.23 1.45 1.39 1.18 1.49
Transportation Research Part D 67 (2019) 464–474
X. (Jason) Cao et al. Transportation Research Part D 67 (2019) 464–474

Table 2
Pattern matrix of factor analysis for residential preferences based on the data of both Oslo and Stavanger.
Exercise Amenity Children Transit Investment Shopping

Eigenvalues 0.880 3.385 0.977 1.915 1.292 1.172


Opportunities for physical exercise 0.622
Proximity to green areas 0.530 0.485
Nice view 0.595
Undisturbed location 0.515
Architecture 0.359
Private garden 0.343 0.609
Good school/kindergarten 0.674
Proximity to train/metro 0.705
Proximity to bus/streetcar 0.765
No social problems 0.403
Favorable investment object 0.661
Good property management 0.571
Proximity to shops 0.714
Easy access to shopping mall 0.625

Notes: 1. The number of factors was chosen based on conceptual interpretability. Because of low loadings or conceptual interpretability, the
following five items were dropped from the exploratory factor analysis: low housing costs, proximity to workplace, proximity to relatives and
friends, distance to major road/rail line, and familiar neighborhood.
2. Extraction Method: Principal Axis Factoring. Rotation: Varimax with Kaiser Normalization. The variance explained is 68.7%. Loadings smaller
than 0.333 were suppressed.

11.2%. The primary data obtained via the questionnaire include a range of individual characteristics on demographic and socio-
economic information, residential mobility (used to define movers), residential preferences, travel characteristics, physical activity
and health. In the quantitative analysis of this study, we used the first three sets of variables.
As shown in Table 1, the households to which the respondents and interviewees belong are on average bigger than for the
population in their counties, and households with more than one breadwinner are somewhat overrepresented. The respondents, and
particularly the interviewees, also have higher education than typical for the county populations. Together, these circumstances
contribute to household income levels considerably higher among the respondents and interviewees than among the inhabitants of
the respective counties. On the other hand, there is also a higher proportion of pensioners among the respondents and interviewees
than in the general population of the counties, which is reflected in higher age and lower proportions of workforce participants in our
samples. Auto ownership levels are similar to those of the county populations, and the same applies to the gender distribution among
the respondents. Since the objective of this study is not to describe the univariate distribution of auto ownership but to explore its
conditional relationship with built environment characteristics using multivariate analysis, the overrepresentation of certain groups
of people in the sample is not expected to materially affect the results (Babbie, 2007; Crano et al., 2015).
In the survey, respondents were asked to indicate the importance of 19 characteristics of housing and neighborhood when they
were looking for a place to live (or if they were to move to a new dwelling) on a four-point scale from “not at all important” (1) to
“highly important” (4). Because some of the characteristics are highly correlated, explanatory factor analysis was used to obtain six
underlying factors of residential preferences: exercise, amenity, children, shopping, transit, and investment (Table 2).
The survey data have been supplemented by several built environment attributes. Based on the addresses of residences and a set of
center coordinates (the main city center, second-order centers and local centers) in each urban region, a number of GIS-based
variables were added to the data file. These variables include distances from the dwelling to the three center hierarchies (the main
center, the closest second-order center and the closest local center), neighborhood population density, neighborhood job density, and
neighborhood job-housing balance. For movers, we measured distances from their former dwelling to the three center hierarchies.
In each of the two regions, qualitative interviews, each lasting for 1–1.5 h, were audio-recorded and transcribed. The interviewees
were chosen among questionnaire respondents volunteering for this task and included persons with different demographic and
socioeconomic characteristics and with their dwellings located in different parts of the urban area. The main purpose of the quali-
tative interviews was explanatory, although we also aimed to be open for new, previously overlooked aspects. An interpretation
scheme, designed for explanatory qualitative research, had been developed in our earlier studies (Næss, 2005, 2013) as an important
tool for interview analysis and was refined and developed further in the present study. The scheme comprised a number of research
questions that we, as researchers, tried to answer, based on the information given by the interviewees. Two project team members
interpreted each interview, and subsequent synthesizing of each group of research questions across the 33 interviews was also carried
out by at least two project team members.

4. Results and discussions

4.1. Cross-sectional analysis

In the survey, the respondents were asked to report the number of autos in their households. Following the literature (Anowar
et al., 2014), we employed ordered logit models with robust error. Three or more vehicles were recoded as three vehicles because a

468
X. (Jason) Cao et al. Transportation Research Part D 67 (2019) 464–474

Table 3
Ordered logit models for auto ownership in Oslo.
Oslo Model 1 Model 2 Final model

Regional location Local attributes Final

Variables Coefficient P-value Coefficient P-value Coefficient P-value

Built environment variables


Residential distance from city center 0.646 0.000 0.305 0.000
Residential distance from second order center 0.149 0.012 0.037 0.597
Residential distance from local center 0.169 0.001 0.053 0.420
Population density −0.605 0.000 −0.484 0.000
Job density −0.269 0.003 −0.199 0.033
Job-housing balance −0.242 0.001 −0.208 0.005
Residential preferences
Proximity to relatives and friends 0.183 0.001 0.133 0.023 0.141 0.016
Proximity to workplace −0.139 0.032 −0.153 0.021 −0.151 0.023
Exercise factor −0.105 0.063 −0.100 0.075 −0.102 0.071
Amenity factor 0.225 0.000 0.195 0.002 0.197 0.002
Children factor 0.182 0.009 0.109 0.134 0.099 0.173
Investment factor 0.094 0.107 0.119 0.046 0.121 0.042
Transit factor −0.307 0.000 −0.334 0.000 −0.308 0.000
Demographics
Age 0.300 0.000 0.215 0.006 0.206 0.008
Household size 0.387 0.000 0.353 0.000 0.357 0.000
Education −0.424 0.000 −0.413 0.000 −0.389 0.000
Household income 1.137 0.000 1.129 0.000 1.145 0.000
Threshold 1 −2.663 −2.774 −2.793
Threshold 2 1.114 1.161 1.171
Threshold 3 4.293 4.327 4.363
Number of observations 1520 1507 1507
Pseudo R2 0.249 0.264 0.268

All explanatory variables are standardized.

very small proportion of households own four or more vehicles. In this study, built environment characteristics are the explanatory
variables of interest whereas residential preferences and demographic characteristics are control variables. The built environment
attributes are related to three categories: distances to the city center and the second-order center represent metropolitan-scale lo-
cation; and distance to the local center, neighborhood densities and job-housing balance represent local-scale attributes. For each city
region, we developed three models: the first model includes metropolitan-scale variables; the second includes local-scale attributes;
and the final model includes all these built environment variables. For the control variables, we included those significant at the 0.1
level in any of the two final models to facilitate a comparison between the two models. That is, the two models have the same set of
explanatory variables. All continuous independent variables were standardized.
Table 3 illustrates the three models for Oslo. In Model 1, both distance variables are significant after controlling for demographics
and residential preferences. Those who live farther away from the city center tend to have more vehicles than those living closer to
the city center. This result is consistent with Zegras (2010) and Næss (2009). The distance to the city center has a much larger effect
than the distance to the second-order center. All four built environment variables at the neighborhood level are significant in Model
2. As we expected, the distance to the closest local center has a positive association with auto ownership and the other three variables
negatively affect auto ownership. Neighborhood population density has a much larger effect than the other three variables. In the
final model, the distance to the city center is the only significant among the three distance variables, suggesting the dominant role of
metropolitan-scale variables in affecting auto ownership. Moreover, the coefficient of the distance to the city center decreases from
0.646 in Model 1 to 0.305 in the final model. Therefore, most of its effect is absorbed by local-scale attributes. This is consistent with
the conceptual model.
In terms of demographic characteristics, age, household size and household income are positively associated with auto ownership
in the final model. These findings are generally consistent with previous studies (Huang et al., 2017; Pinjari et al., 2011; Potoglou and
Kanaroglou, 2008; Zegras, 2010). By contrast, education has a negative association with the number of autos. The affluence level in
Norway is so high that few people cannot afford a car if they need one. People with high education are more likely to have jobs that
are easily accessible by transit than those with low education. Therefore, it is plausible that their need for car is lower. In both data,
the dummy variable indicating the master’s degree or above is negatively (though weakly) associated with the distance from
workplace to the city center. Regarding residential preferences, two individual measures and four factors are significant. In parti-
cular, the preference for proximity to workplace is negatively associated with auto ownership whereas the preference for proximity to
relatives and friends has a positive association. The former is likely to be a reflection of disliking driving and the latter may reflect a
need for such a travel. Further, those who prefer exercise options tend to own fewer cars than others. This is likely because they like
active travel. The amenity and investment factors are positively associated with auto ownership. The latter may represent materi-
alism, whereas the former is often associated with suburban neighborhoods where the need for car travel is higher than in inner-city

469
X. (Jason) Cao et al. Transportation Research Part D 67 (2019) 464–474

Table 4
Ordered logit models for auto ownership in Stavanger.
Stavanger Model 1 Model 2 Final model

Regional location Local attributes Final

Variables Coefficient P-value Coefficient P-value Coefficient P-value

Built environment variables


Residential distance from city center 0.257 0.000 0.176 0.046
Residential distance from second order center 0.028 0.618 0.012 0.873
Residential distance from local center 0.010 0.866 −0.047 0.513
Population density −0.256 0.002 −0.200 0.026
Job density −0.172 0.021 −0.161 0.021
Job-housing balance 0.099 0.190 0.113 0.142
Residential preferences
Proximity to relatives and friends 0.174 0.019 0.202 0.007 0.191 0.011
Proximity to workplace −0.093 0.234 −0.101 0.210 −0.097 0.226
Exercise factor −0.123 0.097 −0.106 0.160 −0.104 0.167
Amenity factor 0.160 0.052 0.130 0.119 0.134 0.109
Children factor 0.205 0.013 0.203 0.015 0.189 0.025
Investment factor 0.158 0.026 0.150 0.035 0.148 0.039
Transit factor −0.358 0.000 −0.354 0.000 −0.355 0.000
Demographics
Age 0.281 0.001 0.295 0.001 0.294 0.001
Household size 0.753 0.000 0.765 0.000 0.773 0.000
Education −0.337 0.000 −0.355 0.000 −0.345 0.000
Household income 1.075 0.000 1.079 0.000 1.084 0.000
Threshold 1 −3.958 −3.981 −4.003
Threshold 2 0.459 0.464 0.469
Threshold 3 3.645 3.699 3.706
Number of observations 1000 992 992
Pseudo R2 0.197 0.204 0.206

All explanatory variables are standardized.

locations. Further, the preference for proximity to transit facility is negatively related to the number of autos, as we expected.
When it comes to Stavanger (Table 4), the final model shows that the distance to the city center still has a more important role in
affecting auto ownership than the other two distance variables. Both neighborhood population density and job density are significant.
However, job-housing balance becomes insignificant. All control variables have the same signs as those in the final model for Oslo.
However, the preference for workplace proximity, the exercise factor, and the amenity factor become insignificant at the 0.1 level and
the children factor becomes significant.
To compare whether explanatory variables influence auto ownership differently in Oslo and Stavanger, we pooled the data of
these two regions. A dummy variable “Oslo” was used to indicate whether a respondent is from Oslo. The interaction term between
“Oslo” and each of all the explanatory variables is included in the model. If an interaction term is significant at the 0.1 level, it is kept
in the final model. Table 5 presents model results. Overall, the “Oslo” dummy and four interaction terms are significant at the 0.05
level. In particular, Oslo residents tend to own fewer vehicles than Stavanger residents. Furthermore, household size has a smaller
positive effect on auto ownership in Oslo than in Stavanger. Moreover, population density has a larger negative impact on auto
ownership in Oslo than in Stavanger; so has job-housing balance. On the other hand, distance to local center has a larger positive
effect in Oslo than in Stavanger. The differences may be attributable to two reasons. First, the predominantly polycentric urban
structure of Stavanger may undermine the effect of the built environment. Second, large metropolitan areas are more likely to have
strong employment centers and extensive transit infrastructure, and experience more severe congestion than medium or small me-
tropolitan areas. Accordingly, the influence of built environment attributes on auto ownership could be stronger in Oslo than in
Stavanger.

4.2. Quasi-longitudinal analysis

This subsection explores the impacts of inward or outward relocation on change in auto ownership, using an ordered logit model.
In the survey, the respondents who moved within the last two years (called movers) were asked to report their number of automobiles
in the household and residential location when they lived at current and previous residences, respectively. Change in auto ownership
was recoded as a three-level ordinal scale: shedder (getting rid of one or more vehicles), no changer, and acquirer (obtaining one or
more vehicles). For change in the distance from their residence to the city center (a key explanatory variable), a positive value means
that the respondent moves away from the city center (outward relocation) and a negative value indicates that she moves toward the
city center (inward relocation). In Oslo, 54 out of 494 movers shed at least one vehicle and 56 obtained one or more vehicles; 227
movers relocated away from the city center and 267 relocated toward the center. In Stavanger, 32 out of 277 movers shed one vehicle
and 55 obtained one or more vehicles; 124 movers relocated away from the city center and 153 relocated toward the center. Based on

470
X. (Jason) Cao et al. Transportation Research Part D 67 (2019) 464–474

Table 5
Pooled Model for Oslo and Stavanger.
Variables Coefficient Robust SE P-value

Oslo residents −0.811 0.087 0.000


Built environment variables
Residential distance from city center 0.251 0.060 0.000
Residential distance from second order center 0.046 0.050 0.357
Residential distance from local center −0.087 0.060 0.150
Population density −0.148 0.081 0.068
Job density −0.176 0.058 0.002
Job-housing balance 0.137 0.071 0.053
Interaction terms with Oslo residents
Residential distance from local center 0.156 0.073 0.033
Population density −0.408 0.108 0.000
Job-housing balance −0.376 0.085 0.000
Household size −0.335 0.150 0.026
Residential preferences
Proximity to relatives and friends 0.158 0.046 0.001
Proximity to workplace −0.123 0.051 0.015
Exercise factor −0.106 0.045 0.018
Amenity factor 0.174 0.050 0.001
Children factor 0.133 0.054 0.014
Investment factor 0.133 0.045 0.004
Transit factor −0.333 0.046 0.000
Demographics
Age 0.238 0.057 0.000
Household size 0.713 0.149 0.000
Education −0.369 0.048 0.000
Household income 1.111 0.058 0.000
Threshold 1 −3.715 0.111
Threshold 2 0.412 0.072
Threshold 3 3.623 0.122
Number of observations 2499
Pseudo R2 0.253

change in the distance to the city center, we created two dummy variables to indicate whether respondents moved toward or away
from the city center as indicated above. Then we multiplied the two dummy variables with change in the distance to the city center to
compare the different impacts of inward and outward relocations on auto ownership. In the survey, we also asked respondents to
report the number of household members at current and previous residences, respectively. We computed change in household size by
taking the difference between the two variable. A positive value indicates that the number of household members increases after
residential relocation. This variable is used to control for the influence of change in household structure on change in auto ownership.
In the survey, we should have captured changes in income, which may affect auto ownership.
Because some respondents moved from other metropolitan areas to the current residence, they are not movers within a me-
tropolitan area. In this study we modeled the Oslo respondents who moved less than 50 km and the Stavanger residents whose
relocation was less than 30 km. Because the farthest outward relocation is less than 50 km in Oslo and 30 km in Stavanger, these two
values were chosen as the thresholds. Furthermore, these distances make sense conceptually based on the size of the metropolitan
areas.
Table 6 presents the model for the pooled data of the two metropolitan areas. First, both inward moving and outward moving are

Table 6
Quasi-longitudinal comparisons of change in auto ownership for inward and outward movers.
Coefficient P-value

Change in distance to city center for inward relocation 0.076 0.000


Change in distance to city center for outward relocation 0.070 0.000
Change in the number of household members after relocation 0.753 0.011
Oslo residents −0.386 0.066
Oslo X Change in the number of household members after relocation −0.438 0.071
Threshold 1 −2.675
Threshold 2 1.848
Number of observation 635
Pseudo R-square 0.079

Dependent variable: change in auto ownership: shredder, no changer, and acquirer.


The interactions between Oslo and inward moving and between Oslo and outward moving are insignificant at the 0.1 level and hence
are dropped out of the final model.

471
X. (Jason) Cao et al. Transportation Research Part D 67 (2019) 464–474

Fig. 2. Proportion of respondents living at different distances from the city centers of Oslo and Stavanger who consider themselves as highly
dependent on car travel in order to reach daily activities. N = 1907 in the Oslo metropolitan area and 1337 in the Stavanger metropolitan area.

associated with change in auto ownership. Outward moving tends to increase auto ownership whereas inward moving has the
opposite effect. Their effect size is similar. Oslo residents are less likely to increase or more likely to reduce auto ownership than
Stavanger residents (significant at the 0.1 level). Moreover, an increase in household size has a smaller effect on change in auto
ownership in Oslo than in Stavanger. The difference in the last two variables between the two regions is likely due to the difference in
urban form.

4.3. Discussion on causal mechanism

Theoretically, when residential location influences auto ownership, the main causal mechanism of such an influence is plausibly
that the need for car travel depends on where in the metropolitan area the dwelling is located. From a time–geographical perspective,
for example, the location of the residence will influence people’s need to own private motor vehicles (Hägerstraand, 1970). If an
individual lives far away from daily trip destinations and is compelled to travel by foot, bike or public transport, these trips will
consume a large proportion of the time budget. The time tied up in the necessary everyday travel may then easily supersede other,
desired doings, for example, being together with the children, participating in organized leisure activities, or managing full-time
employment. By providing oneself with a car (or possibly a second car), higher travel speeds are obtained, and more time will be
available for other everyday activities.
Our survey material supports the above lines of reasoning. As discussed in Section 4.2, outward movement is associated in both
city regions with more need for driving or better conditions for driving, which in turn is associated with an increase in auto own-
ership, and inward movers, especially in Oslo, are more likely to feel less need for cars. The survey respondents’ answers to questions
about car dependence provide additional evidence. There is a clear tendency of higher perceived car dependency among suburbanites
than among inner-city dwellers (Fig. 2). Among respondents living less than 6 km from the city center of Oslo, only 12% consider
themselves to be highly dependent on car driving, compared to 51% among those living more than 22 km from the city center. In
Stavanger metropolitan area, 18% of the respondents living less than 5 km from the city center of Stavanger perceive themselves as
highly car dependent, compared to 36% among those living more than 15 km from the city center. In particular, in the Oslo me-
tropolitan area but also in the Stavanger region, there is a significant correlation between perceived car dependency and actual auto
ownership (Pearson’s r = 0.339, p = 0.000 in the Oslo and 0.231 and 0.000, respectively, in the Stavanger case).
The lower availability of parking space in inner-city areas may also play a role. If parking is expensive and/or it is difficult to find
vacant parking spaces close to the dwelling, this may discourage residents of such areas from buying a car (or a second car). In the
suburbs, parking opportunities are usually much more ample. However, among the Oslo and Stavanger respondents who have
changed their auto ownership after moving, parking and driving conditions are still not mentioned by many as an important reason
for increasing or reducing their auto ownership.
Our qualitative interviews show several examples of perceived high car dependence when living in the outskirts of either of the
two metropolitan areas. Among the car-owning interview households, those living at inner-city locations would generally encounter
fewer problems in sustaining their present activity pattern than interviewees living close to a second-order center or at non-central
locations if they did no longer have access to a private car. For the car-owning inner-city households, being without a car would still
make some activities more cumbersome (especially going to cabins). For most activities, inner-city residents would still be much less
vulnerable if they did not have a car available than suburban residents. In particular, this applies to those interviewees living in
suburbs outside the main suburban public transport nodes. For the six non-centrally residing interviewee households in each

472
X. (Jason) Cao et al. Transportation Research Part D 67 (2019) 464–474

metropolitan area that have one or more private cars at their disposal, the present activities would generally still be possible to carry
out in the absence of car availability. However, depending on the activity types, alternatives to car travel for reaching the relevant
facilities would be more time-consuming, expensive, inconvenient and/or physically exhausting. Because of time constraints in the
households’ daily schedules, it seems likely that some activities (the least necessary ones) would be carried out less often, or maybe
dropped.
The qualitative interviews also include some examples of interviewees who have increased their auto ownership due to moving
from the inner city to a suburb or reduced their auto ownership as a result of having moved from a suburb to the inner city. In
Stavanger metropolitan area, the interviewee ID 35894 moved from an inner-city apartment 2.5 km from the city center of Stavanger
to a single-family house 6.1 km away from the city center. Because of the move, she has increased her travel distances for daily
purposes and changed from walking to driving for trips to virtually all intra-metropolitan activities. She has also changed her auto
ownership as a result of moving. ID 35894 did not have a car when she lived at her previous residential address, and she tried to live
without having her own car the first year after moving to her present dwelling. However, she bought a car after experiencing how
cumbersome it was to reach her relevant destinations by public transport from her new residential location. She found the bus service
to be unreliable, often canceled or full so she had to wait for the next one.
Conversely, one interview household that moved from an outer-suburban, non-central location in the Oslo metropolitan area to
Oslo’s inner-city (ID 50136 and her husband) sold one of the two cars they used to have because they now travel much less by car
than they did before the move. Another interviewee household (ID 53940) had moved to the inner city of Stavanger one and a half
year earlier from a suburb in another Norwegian city (Trondheim). This household had owned two cars since long before the move
but had now stopped using one of them and intended to store it away. When living in Trondheim, they used both cars, reflecting the
location of key trip destinations beyond acceptable non-motorized travel distance when living in their previous suburban neigh-
borhood.
These examples of course do not provide any statistical evidence of the influence of residential location on travel behavior, but
they do show some of the causal mechanisms at work. The way that the interviewees’ auto ownership has developed after moving
partly reflects that the need for car travel is different in their present residential situation from what it used to be. There are no
examples among the 33 interviews of households reducing their auto ownership due to moving outward or increasing it due to
moving closer to the city center. Admittedly, one Stavanger metropolitan area interviewee who had moved further away from the city
center, ID 39693, intended to sell one of the household’s two cars. However, in this case, the move was from a single-family house
near the city center to a new high-density former industrial site close to a train stop and the major bus route. Besides, her intention to
sell the car was equally much for other reasons (her youngest child will soon move out of home) as the good public transport
accessibility of her new dwelling.
This last example illustrates that auto ownership may also change due to changes in household situation accompanying the move.
For example, ID 12137 became without a car when she divorced, but became part of a car-owning household again when her present
partner moved in. Another example is ID 52271 and her ex-husband, who had a car together when they lived together but now have
one car each. This increase is due to the split-up and not 52271’s move to another dwelling. There is also considerable inertia in auto
ownership. For example, some interviewees who do not any longer need a car to reach daily destinations (ID 37424, ID 53783 and ID
52703) still depend on car travel to go to their cabins. Several interviews illustrate that a wish for contact with nature – not only when
going to cabins, but also for weekend trips to natural areas for long-duration outdoor recreation activities – often induces activity
patterns leading to car-dependency among households who would otherwise be able to carry out their activities without car travel.

5. Conclusions

This study used both quantitative and qualitative data to analyze the influence of the built environment on auto ownership in two
urban regions in Norway. Cross-sectional models for both Oslo and Stavanger show that the distance to the city center is positively
associated with auto ownership and it has a more important effect than the distance to other centers. The examples from the
qualitative interviews about changes in auto ownership due to moving and the statements by survey respondents living in different
parts of the metropolitan areas about their perceived car dependency illustrate important causal mechanisms: suburbanites tend to
own more vehicles than inner-city dwellers. The interviewees’ hypothetical considerations about car dependency if living in a dif-
ferent urban context also support the plausibility of causal influences of residential location on auto ownership. Therefore, quanti-
tative and qualitative analyses reach a consistent conclusion. Furthermore, the cross-sectional models also demonstrate the role of
neighborhood-level variables. In both regions, population density and job density are negatively associated with auto ownership, and
in Oslo, job-housing balance also has a small but significant effect.
Quasi-longitudinal analyses show that in both Oslo and Stavanger, moving inward tends to reduce auto ownership whereas
moving outward has the opposite effect. This further corroborates the causal effect of the built environment on auto ownership.
Moreover, residents living in Oslo are more likely to reduce or less likely to increase auto ownership than those in Stavanger. This
difference may be attributable to the fact that many areas in Oslo are developed densely and transit-friendly and hence can support
lifestyle without cars or with fewer cars.
This study has a few limitations in terms of variable measurements. Street design was not investigated separately, but in the
qualitative interviewee this was not at all mentioned as a built environment characteristic that matters to travel. We did not include
any variable for transit access in the analyses. Although most residents in the two city regions can reach a transit stop within a
moderate distance from the dwelling, there are great geographical differences in the level of transit service offered at these stops. It
would therefore not make sense just to measure the distance to the closest transit stop. We could have developed a more complex

473
X. (Jason) Cao et al. Transportation Research Part D 67 (2019) 464–474

measure for transit access in Oslo, but the necessary data for doing so were not available for Stavanger. However, in the context of
Norwegian larger cities, transit accessibility is strongly correlated with proximity to the city center, population density and job
density3, i.e. variables that are already included in the analysis.
Overall, the distance to the city center, a measure of regional destination accessibility, has a strong influence on auto ownership in
both Oslo and Stavanger. Neighborhood density variables including population and employment impose comparable effects.
Furthermore, job-housing balance, a measure of neighborhood diversity, affects auto ownership in Oslo.
For city and regional planning aiming at decreasing car travel and promoting sustainable transport modes, reducing the need for
auto ownership is a key measure. Our studies show that urban containment, particularly through densification in the central parts of
the metropolitan area, is crucial to this end. By reducing travel distance and encouraging the use of modes of transport other than the
private car, densification close to the center of the metropolitan area decreases not only car dependency but also actual levels of car
ownership. In addition to new inner-city dwellings, a high proportion of jobs (except freight-intensive and area-demanding ones)
should also be located close to the city center instead of at suburban or exurban sites if we aim at reducing car dependency and auto
ownership rates.

Acknowledgements

Research Council Norway funded the study on which this paper is based (grant no. 230313).

References

Alonso, W., 1960. A theory of the urban land market. Pap. Reg. Sci. 6, 149–157.
Anowar, S., Eluru, N., Miranda-Moreno, L.F., 2014. Alternative modeling approaches used for examining automobile ownership: a comprehensive review. Transp. Rev. 34,
441–473.
Babbie, E.R., 2007. The Practice of Social Research, 11th ed. Thomson Wadsworth, Belmont, CA.
Bhat, C.R., Guo, J.Y., 2007. A comprehensive analysis of built environment characteristics on household residential choice and auto ownership levels. Transp. Res. Part B 41,
506–526.
Boarnet, M.G., 2011. A broader context for land use and travel behavior, and a research agenda. J. Am. Plann. Assoc. 77, 197–213.
Cao, J., Cao, X., 2014. The impacts of LRT, neighbourhood characteristics, and self-selection on auto ownership: evidence from Minneapolis-St. Paul. Urban Stud. 51, 2068–2087.
Chen, C., Gong, H., Paaswell, R., 2008. Role of the built environment on mode choice decisions: additional evidence on the impact of density. Plann. – Pol. – Res. – Pract. 35,
285–299.
Clifton, K.J., Handy, S.L., 2003. Qualitative methods in travel behaviour research. In: Stopher, P., Jones, P. (Eds.), Transport Survey Quality & Innovation. Elsevier Science,
Pergamon, Oxford UK, pp. 283–302.
Crano, W.D., Brewer, M.B., Lac, A., 2015. Principles and Methods of Social Research, third ed. Routledge, New York.
Ding, C., Wang, Y., Tang, T., Mishra, S., Liu, C., 2018. Joint analysis of the spatial impacts of built environment on car ownership and travel mode choice. Transp. Res. Part D 60,
28–40.
Dohyung, K., Jiyoung, P., Andy, H., 2017. The role of destination’s built environment on nonmotorized travel behavior: a case of Long Beach, California. J. Plann. Edu. Res.
https://doi.org/10.1177/0739456X16688765.
Fishman, R., 1996. Bourgeois utopias: visions of Suburbia. In: Fainstein, S., Campbell, S. (Eds.), Readings in Urban Theory. Blackwell, Malden, Mass./Oxford, UK, pp. 23–71.
Handy, S., 1996. Methodologies for exploring the link between urban form and travel behavior. Transp. Res. Part D: Transp. Environ. 1, 151–165.
Hu, L., Sun, T., Wang, L., 2018. Evolving urban spatial structure and commuting patterns: a case study of Beijing, China. Transp. Res. Part D: Transp. Environ. 59, 11–22.
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.
Hägerstraand, T., 1970. What about people in regional science? Pap. Reg. Sci. 24, 7–24.
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.
Kwan, M.-P., Weber, J., 2008. Scale and accessibility: implications for the analysis of land use–travel interaction. Appl. Geogr. 28, 110–123.
Macfarlane, G.S., Garrow, L.A., Mokhtarian, P.L., 2015. The influences of past and present residential locations on vehicle ownership decisions. Transp. Res. Part A: Pol. Pract. 74,
186–200.
Mokhtarian, P.L., Cao, X., 2008. Examining the impacts of residential self-selection on travel behavior: a focus on methodologies. Transp. Res. Part B: Methodol. 42, 204–228.
Norwegian-Government, 1993. Rikspolitiske retningslinjer for samordnet areal og transportplanlegging. Oslo Ministry of the Environment.
Norwegian-Government, 2014. Statlige planretningslinjer for samordnet bolig-, areal- og transportplanlegging. (Central-governmental planning guidelines for coordinated re-
sidential, land us and transport planning).
Norwegian-Government, 2017. Nasjonal transportplan 2018–2029. Meld. St. 33 (2016–2017). The Norwegian Government, Oslo.
Næss, P., 2005. Residential location affects travel behavior – but how and why? The case of Copenhagen Metropolitan Area. Plann. Prog. 63, 167–257.
Næss, P., 2009. Residential self-selection and appropriate control variables in land use: travel studies. Transp. Rev. 29, 293–324.
Næss, P., 2013. Residential location, transport rationales and daily-life travel behavior: the case of Hangzhou Metropolitan Area, China. Prog. Plann. 79, 1–50.
Næss, P., 2015. The explanatory qualitative-quantitative method. In: Pinho, P., Silva, C. (Eds.), Mobility Patterns and Urban Structure. Ashgate, Farnham, pp. 101–120.
Pinjari, A., Pendyala, R., Bhat, C., Waddell, P., 2011. Modeling the choice continuum: an integrated model of residential location, auto ownership, bicycle ownership, and
commute tour mode choice decisions. Transportation 38, 933–958.
Potoglou, D., Kanaroglou, P.S., 2008. Modelling car ownership in urban areas: a case study of Hamilton, Canada. J. Transp. Geogr. 16, 42–54.
Scheurer, J., 2014. A comparison of public transport and land use integration in 12 European cities. Appendix slides for Oslo sowing closeness centrality and degree centrality of
the transit system. Presentation at Urban Sustainability research group seminar, September 11, 2014. Ås: Norwegian University of Life Sciences.
Singleton, R.A., Straits, B.C., 2010. Approaches to Social Research. Oxford University Press, New York.
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.
Wolday, F., Cao, J., Næss, P., 2018. Examining factors that keep residents with high transit preference away from transit-rich zones and associated behavior outcomes. J. Transp.
Geogr. 66, 224–234.
Zegras, C., 2010. The built environment and motor vehicle ownership and use: evidence from Santiago de Chile. Urban Stud. 47, 1793–1817.
Zhang, Y., Zheng, S., Sun, C., Wang, R., 2017. Does subway proximity discourage automobility? Evidence from Beijing. Transp. Res. Part D: Transp. Environ. 52B, 506–517.

3
This is evident, among others, from a mapping of the ‘closeness centrality’ and ‘degree centrality’ accessibility indices of 97 transit nodes in
Greater Oslo (Scheurer, 2014).

474

You might also like