Cities 71 (2017) 30–40

Contents lists available at ScienceDirect

Cities
journal homepage: www.elsevier.com/locate/cities

Measuring the effects of street network configurations on walking in Seoul, MARK

Korea
Chang-Deok Kang
Dept. of Urban Planning and Real Estate, Chung-Ang University, 84 Heukseok-ro, Dongjak-gu, Seoul, 156-756, South Korea

A R T I C L E I N F O A B S T R A C T

Keywords: Many scholars have confirmed the close correlation between street network configurations and pedestrian be-
Street network havior. To elaborate on this discussion and to analyze the connection between street layout and walking using
Street configuration sophisticated methods, we attempt to capture the links and surrounding network features of urban street pat-
Walk terns. To this end, we employ the following four elements: closeness, betweenness, severance, and efficiency. Our
Multilevel regression
empirical study examines street network configurations via a multi-scale network radius. The results show that
Seoul
locations with closer destinations and areas where people pass more frequently throughout a given network are
associated with higher walking volume. Furthermore, a greater resemblance between straight-line and path
distances from origins to destinations tend to encourage walking mobility. A longer convex hull, with a max-
imum radius that covers a wider shape and structure of links, is the most favorable setting for walking within a
radius of up to 500 m. We confirm that a higher convex hull-shape index reveals more deviation from the
circularity of a street network's convex hull. This negatively affects walking within a radius of 1–2 km. We
conclude that link and network radius-based attributes of street formations have significant effects on the spatial
variations of walking within multi-scale walkable neighborhoods. The framework and results are intended to
help with the design of pedestrian-friendly street patterns and to create sustainable and livable cities.

1. Introduction
Walkable cities and neighborhoods are the core paradigm in urban
planning, urban design, and transport policy. A significant amount of
literature has addressed the advantages of walkable cities in terms of
urban economics, urban environments, health, and social cohesion
(Talen & Koschinsky, 2013). The central questions for urban planners
and designers are which settings benefit pedestrians, and how to im-
prove walkable cities. Street networks are the leading characteristics of
pedestrian-friendly cities. As many studies have confirmed, street at-
tributes such as density, connectivity, and proximity have a significant
effect on walking behaviors and on choosing whether to walk (Cervero,
Sarmiento, Jacoby, Gomez, & Neiman, 2009; Lamíquiz & López-
Domínguez, 2015).
Previous studies on walkable urban settings have identified various
factors, such as socioeconomic elements, the built environments, and
street configurations that affect walking activities and walking volume
in urban areas (Peiravian, Derrible, & Ijaz, 2014). A number of studies
have found that a greater density of residents and mixed high-rise de-
velopment, along with public transit, create substantially better con-
ditions for pedestrians (Agrawal & Schimek, 2007;
Cervero & Kockelman, 1997; Holtzclaw, 1994; Lopez-Zetina,
Lee, & Friis, 2005). Other studies highlight the blended growth of re-
sidential and neighborhood retail properties as the main drivers of in-
creased walking activities (Cervero, 1996; Ewing, 1995; Frank & Pivo,
1994). Prior empirical models have controlled for street features, and
have commonly noted the power of street patterns, which serve as
channels to help people reach their destinations.
Many studies on street configuration have found that diverse street
attributes generate higher walking volume. While early investigations
emphasized the physical features of street networks in relation to
walking behavior, later research confirmed the significance of street
density and link structures for walking activities (Cervero & Kockelman,
1997; Crane & Crepeau, 1998; Ewing & Cervero, 2010; Lee & Moudon,
2006; Song & Knaap, 2004). As analytic methods have advanced toward
catchment area concepts and the space syntax methodology, recent
research has discovered that pedestrians tend to be concentrated
around their main destinations and adjacent streets with higher in-
tegration and choice, as measured by space syntax tools (Handy,
Paterson, & Butler, 2003; Hess, Moudon, Snyder, & Stanilov, 1999;
Hillier, 1996; Peponis & Wineman, 2002; Rodríguez,
Khattak, & Evenson, 2006). Since local contexts produce spatial con-
nections between street designs and walking volume, many scholars
have controlled for other relevant aspects of walking environments,

E-mail address: cdkang@cau.ac.kr.

http://dx.doi.org/10.1016/j.cities.2017.07.005
Received 5 September 2016; Received in revised form 24 March 2017; Accepted 9 July 2017
Available online 13 July 2017
0264-2751/ © 2017 Elsevier Ltd. All rights reserved.
C.-D. Kang
such as the street density in given areas, connectivity, and block size
(Baran, Rodriguez, & Khattak, 2008).
A few recent studies have attempted to test the validity of newly
measured street patterns in terms of property prices, social cohesion,
and walking access to green spaces. Local and global accessibility along
thoroughfares substantially explain the spatial variations of residential
property prices. Specifically, a longer convex hull, with a maximum
radius that covers a wider shape and structure of links, tends to enhance
social interactions and cohesion in neighborhoods (Chiaradia, Hillier,
Schwander, & Barnes, 2013; Cooper, Fone, & Chiaradia, 2014). Fur-
thermore, a higher betweenness of streets indicates better network
connectivity and, combined with green spaces, encourages people to
walk (Sarkar et al., 2015).
While many scholars have tried to confirm that demographic traits,
land development features, and street configurations affect the spatial
patterns of pedestrian volume, few tests have used a sophisticated ap-
proach to show how heterogeneous attributes of urban layouts are
linked to walking volume. Our hypothesis is testable in that walking
volume and the choice to walk are associated with numerous traits of
street designs, as verified in prior research (Lee & Moudon, 2006;
Song & Knaap, 2004). Empirical tests result in specific policy implica-
tions and detailed guidance, which will enable improved pedestrian-
friendly street designs and urban environments.
Existing studies emphasizing the importance of street features have
made several significant contributions to urban studies and policy.
First, these previous studies have attempted to understand the effects of
fine-grained street layout on walking. Theories on central business
district-focused spatial economies have failed to explain polycentric
urban structure. As a result, small-scale layout analyses have increased
as an alternative perspective to capture local features of street config-
uration (Cuthbert, 2007). Second, analyzing spatial street layout sug-
gests rigorous knowledge of unexplored micro-level streets beyond the
level of over-simplified accessibility and centrality at the macro-level
(Chiaradia et al., 2013). In particular, individual walkers tend to re-
spond to the conditions and features of their nearest streets. Thus,
delving into street layout in small-scale neighborhoods effectively ex-
plains and predicts spatial variation in walking volume. Third, previous
studies have found a local connection between socioeconomic activities
and street networks. Because urban spatial structures consist of human
settlement and movement, these sophisticated studies have strength-
ened our understanding of the dynamic connection between land use
and urban mobility. Fourth, the diverse approaches of previous studies,
such as accessibility analyses, graph theories, space syntax, and mor-
phological measures, suggest that they provide valuable insights and
relevant policies on multiple features of street configurations (Sevtsuk,
2010). The accumulated findings have led to follow-up advanced stu-
dies and more effective policy and design for livable and sustainable
cities. Finally, empirical studies have suggested specific and detailed
policy implications for urban planning, land use planning, transporta-
tion policy, and urban design (Rashid, 2016). Specifically, diverse
empirical studies have emphasized the importance of fitting relevant
policies within local contexts, such as land use, street configuration,
and walking behaviors.
Notwithstanding the intense interest and aforementioned existing
empirical studies, we still need clear evidence of the multifaceted re-
lation between street configuration and walking behaviors. This arises
primarily because various features of street layout affect the spatial
variation of walking volume. This study expands the perspectives on
various street attributes by setting hypothetical notions, such as desti-
nation density and distribution, cognitive difficulty of street navigation,
ease of walking (flow), and navigability through network connections
and covered areas. These notions are conceptualized in four metrics and
quantified using seven sub-measurements.
This study applies a newly developed network analysis containing
multiple metrics (closeness, betweenness, severance, and efficiency) in the
continuum of the discussion of previous approaches. We expect that
31
Cities 71 (2017) 30–40
these link-centered approaches will capture the multi-dimensional
spatial features of street network arrangements (Cooper et al., 2014).
Street layouts are not simple, homogenous features of urban areas.
Thus, we need a fine-grained, micro-level analysis of street designs to
discover the links and surrounding network features of street layouts
(Chiaradia et al., 2013). The section titled “Measuring link-focused,
spatial network configurations” discusses the need for new methods in
further detail.
Thus, we aim to elaborate on the discussion of prior studies by
testing how various characteristics of street patterns change walking
activities in Seoul, Korea. Furthermore, we attempt to assess the effects
of the links and surrounding network features on street configurations.
In order to better understand the impacts of the new metrics of street
network layouts on walking volume, we measure closeness, betweenness,
severance, and efficiency, using the link as a basic unit. We use the
concept of a network radius to capture the surrounding networks. A
network radius is a type of pedestrian catchment area in which the
radius is set as the walkable distance. The noteworthy difference be-
tween the two concepts of network radius and pedestrian catchment
area is that we use the radius to measure the diverse metrics of street
configuration, rather than using the radius per se.
We theorize that the diverse attributes of built street patterns will be
correlated with spatial variations in walking volume. Thus, we test how
the four metrics of street configurations affect walking volume, while
controlling for other street attributes, location attributes, neighborhood
attributes, and population and employment density.
In summary, our contributions to the relevant studies and policies
are as follows. First, we measure street configurations in terms of clo-
seness, betweenness, severance, and efficiency at a local level of analysis in
order to capture various features of street networks. Second, we test
core hypotheses on the multifaceted effects of measured street config-
urations on walking, controlling for other explanatory variables. Third,
we compare the various effects among the metrics of street configura-
tion within several neighborhood scales in order to provide specific
ideas on urban policy and design. Fourth, we interpret key findings of
previous studies and within local contexts to aid our understanding of
the micro-level street configuration associated with walking behaviors.
This paper consists of four sections. The first section discusses the
study area and data sources that enabled us to determine how the
multiple elements of street layouts influence spatial variations in
walking. The second section describes the motivation behind measuring
link-focused spatial network configurations, the street metrics (and
relevant hypotheses), other variables, and the analytic methods used.
The third section describes and discusses the key findings of our em-
pirical analysis. Then, the final section summarizes the findings and
suggests relevant implications for urban policy and design.
2. Study context and data sources
With > 10 million residents, Seoul contains a high proportion of
mixed land use and dense street systems. In 2010, the population
density was 17,466 people per km2 (Urban Information Network,
2010). The large number of sidewalks and the high-quality public
transit system have created a favorable urban environment for pedes-
trians. The total volume of sidewalks in 2009 amounted to 2523 km2.
The public transit system comprises nine subway lines and > 290 sta-
tions, with routes that collectively span 316.8 km. The bus transit
system provides > 11,000 bus stops and several bus rapid transit (BRT)
lanes (Seoul Institute, 2015).
Since 2004, the Seoul Metropolitan Government (SMG) has estab-
lished key policy goals and taken steps to make the metropolis more
sustainable and pedestrian friendly. In 2013, the SMG announced its
plan “Pedestrian-Friendly City, Seoul.” The SMG is constantly enhan-
cing urban settings by creating pedestrian-friendly streets with wider
sidewalks, implementing lower speed limits for vehicles, and creating
connections between formerly separated streets (Seoul Metropolitan
C.-D. Kang Cities 71 (2017) 30–40

Fig. 2. Example illustrating the limitations of intersection density.

Source: Cooper et al. (2014).

Fig. 1. Spatial distribution of survey locations.

Government, 2013).
We collected multiple geographical data: (1) walking volume; (2)
roads and streets; (3) Korean census tract maps; (4) subway station
locations; (5) bus stop locations; (6) building information by land use;
(7) location information on urban cores and sub-centers, roads and
streets, parks, schools, and main retail zones; and (8) number of re-
sidents and workers per Korean census tract.
The primary source of data in this study is the pedestrian survey
data from 2009, containing the number of walkers at 9848 survey lo-
cations, as well as other street attributes across Seoul (Seoul
Metropolitan Government, 2010). The survey locations cover all types
of neighborhood in Seoul, including low-density to high-density re-
sidential areas, commercial and office areas, industrial areas, among
others, as well as street and road networks. Fig. 1 shows the spatial
distribution of the survey locations. Fig. 3. The network radius and crow flight distance.
Source: Author

3. Methodology and models
3.1. Measuring link-focused, spatial network configurations
The essential component of our study methodology focuses on
network links as a unit of analysis rather than as node points. A few
significant advantages of link-focused analyses justify its application in
this study. First, previous analyses of intersections have ignored the
modifiable areal unit problem (MAUP), which causes statistically
biased results from different spatial units with aggregated point-based
values (Cooper et al., 2014). Thus, standardizing links along street
networks is an effective way to avoid the MAUP. Second, the point
(node)-based approach does not provide reliable measures. For ex-
ample, as shown in Fig. 2, different network patterns within the same
radius may contain the same intersection density (Cooper et al., 2014).
However, a network-focused approach uses network links as atomic and
indivisible units of everyday urban life. Network links are the sub-
stantial units of urban movement. For example, a higher link density is
associated with a higher density of jobs and housing (Chiaradia,
Cooper, & Wedderburn, 2014). Third, compared with access to central
business districts (CBDs), accessibility metrics based on topological
networks better capture the spatial complexity of polycentric urban
areas and explain the spatial variations in property values (Xiao,
Webster, & Orford, 2016). Thus, we expect that calculating street net-
work layouts based on a link structure will capture the features of street
configurations within small-scale neighborhoods. Fourth, given that the
network radius captures the features of surrounding networks for in-
dividual links, as shown in Fig. 3, we analyze the attributes within a
given radius. For example, a 600 m radius is defined as all reachable
points from the central point of each link, traveling 600 m throughout
the network. This approach is a flexible way to measure spatial network
characteristics via a multi-scaled radius. Finally, network links are ob-
jective entities used in diverse spatial analyses and policymaking. Thus,
the link-based approach is an effective way to share geographical data
and relevant studies. Additionally, the results of link-based studies are
easier to comprehend and communicate among urban scholars, urban
planners, and urban designers (Cooper & Chiaradia, 2015).
3.2. Description of hypotheses and metrics
3.2.1. Hypotheses
After establishing testable notions and their hypotheses, we identify
the most promising metrics that influence walking volume. We select
key factors from the results of a spatial design network analysis (sDNA)
in terms of their theoretical and empirical validation. An sDNA mea-
sures localized network accessibility, centrality, and navigability in-
dices pertaining to radius-based networks, destination density, detour
and flow, and link structure and shape (Sarkar et al., 2015). Table 1
describes our notions and hypotheses on density, density distribution,
twistiness, flow, and efficiency. These hypotheses cover multi-dimen-
sional aspects of street configurations that we expect will explain

32
C.-D. Kang Cities 71 (2017) 30–40

Table 1
Testable core hypotheses and descriptions.

Hypotheses Description

Density (D) Measuring the overall density of destinations, starting from an origin within a given radius.
Hypothesis: The higher the density of destinations, the greater the walking mobility, which helps people save on travel costs.
Density distribution (DD) Measuring the spatial distribution of destination density close to or far from an origin within a given radius.
Hypothesis: The optimal density distribution is positively associated with walking.
Twistiness (T) Measuring the cognitive difficulty of navigating a route within a given radius.
Hypothesis: The more twisted a street configuration, the greater the psychological barrier between each origin and reachable destination, which
results in less walking.
Flow (F) Measures estimated walking flow along streets.
Hypothesis: Places with a more predictable walking flow are associated with a higher walking volume.
Efficiency (E) Measures the efficiency of a street network of either space or distance in a neighborhood. Efficiency indicates navigability, which includes the shape
and structure of links and their connections.
Hypothesis: The greater the efficiency and navigability of a street network, the more people will choose to walk.

Source: Modified from Cooper et al. (2014).

walking behaviors. While density measures the spatial concentration of
reachable destinations, density distribution estimates how close an origin
is to destinations along street networks. We expect higher density and
density distribution to be favorable urban settings for pedestrians.
Twistiness reveals the cognitive difficulty of navigating while walking;
flow refers to the ease of walking along streets. Obviously, lower twis-
tiness and higher flow present the greatest probability of choosing to
walk. Finally, efficiency indicates spatial navigability, and includes the
shape and structure of links and their connections. This notion provides
an effective way to create a harmonious structure of street links and
surrounding areas. Conceptualizing and measuring these notions to test
their association with walking behaviors will be a useful resource when
designing pedestrian-friendly streets and sustainable cities.
3.2.2. Metrics of spatial network configurations
We determine the attributes of street layouts in terms of closeness,
betweenness, severance, and efficiency using sDNA software
(Cooper & Chiaradia, 2015; Cooper, Chiaradia, & Webster, 2016). We
set 2 km as the maximum radius, focusing on walkable neighborhoods
based on a previous relevant Korean study, which confirmed that
homemakers in Seoul tend to walk approximately 2.6 km per day, on
average (Park et al., 2016). Pedestrians tend to follow the Euclidean
shortest path, and continuous space preserves accuracy at the pedes-
trian level (Cooper et al., 2014). Thus, we apply Euclidean metrics and
continuous space to setting the network radii. To compare the effects
via a multi-scale radius, we calculate the four metrics for 100 m, 300 m,
and 500 m, as local radii, 1000 m as a meso radius, and 2000 m as a
global radius within walkable neighborhoods. These network radii
broadly cover reasonable walking distances.
Then, we analyze spatial network configurations in terms of close-
ness, betweenness, severance, and efficiency. Each measure contains sub-
measures to quantify the attributes of spatial network arrangements.
Although sDNA provides more indices, we strictly limit theoretical and
empirical interpretable measures using a variance inflation factor (VIF)
of < 4, after empirically testing the given metrics.
The first metric, closeness, is a popular benchmark in spatial network
analyses that measure how close the origin is to all destinations along
specific street networks. We calculate this metric using two measures:
mean Euclidean distance (MED), and network quantity, penalized by
distance, Euclidean (NQPDE). The MED represents the mean Euclidean
meter-based length between the origin and all reachable places within a
given radius. A higher value of MED means reduced closeness. The
NQPDE is calculated as the number of network links divided by the
Euclidean distance between the origin and all reachable destinations
within each radius. Thus, a higher value of NQPDE indicates that places
are relatively more accessible and closer to each other.
Betweenness estimates how each network link is populated, with
entities moving throughout the systems. Thus, betweenness assesses all
possible trips passing through a network link. We measure betweenness
within each radius, assuming that each trip takes place via the shortest
path, by Euclidean distance. Betweenness focuses on the intermediate
links in moving between specific locations and all reachable places
within a given radius.
The relative closeness between the crow flight distance and the
shortest path distance determines spatial variations in walking. This
notion is conceptualized as severance within network detour analyses.
Severance indicates the opposite of connectivity. We calculate the mean
crow flight (MCF) distance and the diversion ratio. The MCF distance
indicates the mean straight-line length between the centers of two links
as the origin and destination. The diversion ratio is calculated as the
mean value of the shortest network distance divided by straight-line
distance over all links. The diversion ratio quantifies how far the
shortest network distance strays from the straight-line distance (Cooper
et al., 2014).
Finally, the aforementioned measures focus on the spatial networks.
Spatial layouts cover specific zones. Efficiency measures the navigability
of the connected links covering space or distance in local areas (Cooper
et al., 2014). The principle of a convex hull applies to an area covered
by spatial networks. We calculate the convex hull maximum radius
(HullR) within each network radius, as well as the convex hull shape
index (HullSI). The HullR indicates the maximum radius of the convex
hull, as gauged from the center of the origin link within each radius, as
shown in Fig. 4. The HullSI represents how close the convex hull is to
forming a circle. Here, a larger value of the HullSI is less like a circle. In
other words, the higher the value of the HullSI, the greater the like-
lihood that the convex hull area is not the same as a circle.
Table 2 shows the four key metrics and describes the sub-mea-
surements associated with main hypotheses in given in Table 1. The
Appendix provides the mathematical formulae of the metric sub-mea-
surements.
Fig. 4. Illustration of convex hull maximum radius.
(Source: Cooper et al. (2014))

33
C.-D. Kang
Table 2
Metrics of street network configurations and relevant hypotheses.
Metric Measurement
Closeness Mean Euclidean distance (MED)
Network quantity penalized by distance, Euclidean
(NQPDE)
Betweenness Betweenness
Severance Mean crow flight (MCF) distance
Diversion ratio
Efficiency Convex hull maximum radius (HullR)
Convex hull shape index (HullSI)
Source: Partly modified from Cooper et al. (2014), with additional information by the author.
3.3. Additional variables
3.3.1. Street attributes
In the pedestrian survey, the SMG recorded the number of walkers
and the street conditions, including the physical characteristics of
streets, such as sidewalk types and width (meters), street facilities, the
slopes of surveyed locations, and the number of automobile lanes and
crosswalks near the streets. Street facilities include various installa-
tions, such as lamps, trees, and other facilities, while sidewalks are
divided into pedestrian streets with automobile roads, pedestrian streets
with cycling roads, and pedestrian streets (Kang, 2015). To convert the
nominal variables into dummy variables, we assign a value one if the
survey locations featured pedestrian streets with automobile roads, pe-
destrian streets with cycling roads, and pedestrian streets, crosswalks, and
slopes, and zero otherwise.
3.3.2. Neighborhood attributes
To evaluate how the characteristics of neighborhood land use mix
and transportation density affect walking behavior, we measure land
use, mixed uses, and the net density of seven Korean census tract units.
Because mixed land use is a popular index in urban studies, we calcu-
late the land-use index for seven units: residential, small retail, large
retail, offices, public, industrial, and other properties. Here, we use the
entropy index. When the value of the mix is zero, we find mono land
use, while a value of one represents an even land-use mix
(Cervero & Kockelman, 1997). Because the total area of Korean census
tracts includes non-developable zones, we calculate the net area, which
indicates the entire biotope land of housing, retail and office, industrial,
and infrastructure spaces. Our calculation is based on the Seoul Biotope
Map of 2010 (Seoul Metropolitan Government, 2013). We estimate the
net densities of the land uses by dividing the total building area of each
land use zone, the number of transit facilities (subway stations and bus
stops), and intersections of transportation patterns by the net area. Fi-
nally, the net density of parks is calculated as the total park area, based
on the Seoul Biotope Map, divided by the net area of Korean census
tract units.
3.3.3. Location attributes
To verify the influence of key locations and transportation attri-
butes, we measure the distance from the pedestrian-counted places to
the urban core and to the five sub-central business districts, as well as to
the nearest metro stations, bus stops, roads and streets, core retail
zones, schools, and parks. Prior relevant studies have confirmed that
proximity to the urban center and sub-centers significantly affects re-
sidents' location choice, firm location, and housing prices (e.g.,
Ewing & Cervero, 2010). Proximity to transportation networks and
main facilities, schools, retail clusters, and green spaces substantially
changes the spatial variations of walking mobility. Finally, we add the
straight-line distance to the classified land uses in order to control for
the effects of closeness (for each type of land use) on pedestrian pat-
terns.
Cities 71 (2017) 30–40
Description Hypothesis
Mean length in Euclidean metrics between origin and all reachable destinations in a D
radius
Number of links divided by Euclidean length along networks in a radius D, DD, T
Flow prediction based on Euclidean shortest path distance F
Mean crow flight distance between centers of two links in a radius T
Mean of path distance to crow flight length per route
Maximum radius of convex hull shape, covering street networks within a specific radius E
Circularity of convex hull shape, covering street networks within a specific radius
3.3.4. Population and employment density
Many existing studies have verified that a greater density of re-
sidents and workers is positively linked with spatial variations in
walking volume (Forsyth, Oakes, Schmitz, & Hearst, 2007). To control
for the effects of socioeconomic features within neighborhoods, the
models include population and employment density, indicating the
population and workers per total developable area of each census tract,
respectively.
3.4. Multilevel regression models
We employ various units of data, such as walker-counted locations
and census tract units. Thus, we chose multilevel regression models to
identify how the spatial features of street layouts affect walking vo-
lume. Most prior studies used the intra-class correlation to evaluate
whether a multilevel regression is necessary. The intra-class correlation
measures how the explanatory variables at the group level predict the
overall variation in the dependent variable at an individual level. We
found that the intra-class correlation (ICC) was higher than 0.05, in-
dicating the necessity of a multilevel regression. The multilevel re-
gression model addresses the problem of biased model results from
general regressions (Rabe-Hesketh & Skrondal, 2008). Based on model
selection and available data, we propose the following equation:
Pij = γ00 + β1 Aijk + β2 Sijk + β3 Cijk + β4 Lijk + β5 Eijk + μ0j + εij, (1)
where Pij is the number of walkers in location i in census tract j; βk are
the estimated coefficients of each explanatory variable; γ00 are con-
stants; Aijk are the street configuration features close to location i in
tract j; Sijk are the physical attributes of streets in location i in tract j; Cijk
are the neighborhood attributes in tract j; Lijk are the location attributes
in location i in tract j; Eijk are the resident and worker density in tract j;
and μ0j and εij are the error terms for tract j and location i, respectively
(Kang, 2015). Then, we add the vectors A, S, C, L, and E from Eq. (1) to
Tables 3 and 4 in order to link the vectors to individual variables.
There is no general theoretical consensus on how to select a type of
functional form (Duncan, 2011). After testing a few functional forms in
terms of theoretical and empirical considerations, this study chose a
log-log form, converting both the dependent and explanatory variables
with continuous values into natural logarithms. The log-log form pro-
vides additional merits, whereby the parameters can be interpreted as
mean elasticities, representing the relative response of walking volume
to variations in the explanatory variables (Shyr, Andersson, Wang,
Huang, & Liu, 2013). We apply the variance inflation factor (VIF) to
diagnose correlations among the explanatory variables. The empirical
models only use the explanatory variables when the VIF values are < 4,
based on the rule that a VIF value of > 10 indicates a high degree of
multicollinearity (Kleinbaum, Lawrence, & Muller, 1988). Table 3 de-
scribes the variables and the descriptive statistics.
34
C.-D. Kang Cities 71 (2017) 30–40

Table 3 square value. First, we simply utilized daily activities to reach a specific
Descriptive statistics. destination, excluding leisure or social activities (Gehl, 2011). Second,
our models miss individual preferences and behaviors that partially
Variable Description Mean Min Max
explain walking volume (Baran et al., 2008). Finally, the models are
Pedestrian volume used to analyze only 9848 locations in Seoul. Thus, we cannot expect
Average business day pedestrians 3061.14 6.00 106,186.00 robust models such that the independent variables accurately explain
Average Saturday pedestrians 2986.92 7.00 113,606.00
variations in walking volume. The intra-class correlation (ICC) of each
Metrics of street configuration (network radius = 1000 m, vector A) model ranges from 0.20 to 0.21, justifying our use of multilevel models.
Mean Euclidean distance (MED) 657.57 344.24 871.73 We only interpret the coefficients of explanatory variables within a
Network quantity penalized by distance, 0.98 0.003 3.40
Euclidean (NQPDE)
statistical significance of 5%.
Betweenness 5644.66 6.14 49,202.74
Mean crow flight distance (MCF) 516.07 249.76 762.22 4.1. Effects of spatial network configurations
Diversion ratio 1.31 1.01 2.33
Convex hull maximum radius (HullR) 6.85 6.57 6.91
Figs. 5 through 8 summarize the effects of the spatial network
Convex hull shape index (HullSI) 1.12 1.03 9.76
configurations on walking. The coefficients of closeness, betweenness,
Street attributes (vector S)
severance, and efficiency draw upon the model outputs (in multi-scale
Sidewalk width 3.959 1.00 24.30
Number of street ways 2.879 1.00 18.00 units), for walkable neighborhoods, at distances of 100 m, 300 m,
Street furniture (yes = 1, no = 0) 0.923 0.00 1.00 500 m, 1000 m, and 2000 m. At the end of each sub-metric shown in
Pedestrian streets with automobile 0.399 0.00 1.00 the figures, W indicates business days and WND represents Saturdays.
roads = 1, pedestrian streets = 0 With regard to the effects of closeness, it is notable that a lower
Pedestrian streets with cycling roads = 1, 0.053 0.00 1.00
MED value between origins and destinations within a radius, and a
pedestrian streets = 0
Crosswalk (yes = 1, no = 0) 0.462 0.00 1.00 higher NQPDE value both indicate greater closeness. Fig. 5 clearly
Street slope (yes = 1, no = 0) 0.250 0.00 1.00 shows that because a greater distance (MED) between origins and
Neighborhood attributes (vector C) destinations along specific streets tends to reduce walking, greater
Entropy index of seven land use types 0.41 0.00 0.89 closeness is associated with more walking within a radius of 100 m,
Net density of total floor area of industrial 7505.97 0.00 1,466,091.00 1000 m, and 2000 m. The positive effects appear stronger within 100 m
property and relatively weaker from 1000 m to 2000 m. Considering the nature
Density of total floor area of public 7212.16 0.00 588,172.60
property
of metrics, greater closeness implies a higher density of destinations
Density of total floor area of large retail 8878.85 0.00 700,692.20 within a given radius. These results confirm that people tend to walk
stores throughout street patterns with dense destinations because they can
Density of total floor area of office property 37,886.24 0.00 973,701.20 save on transport costs and their destinations can be accessed con-
Density of total floor area of residential 21,751.44 0.00 5,482,388.00
veniently. Then, as a further closeness metric, a higher network capa-
property
Density of total floor area of small retail 29,791.36 0.00 792,344.50 city—compared with the straight-line distances between origins and
stores destinations (NQPDE)—affects pedestrian volume positively, although
Density of metro stations 0.0000029 0.00 0.0032 the extent of the influence varies over the multi-scaled radii, except
Density of bus-stops 0.0000586 0.00 0.0034 within 100 m. The coefficients became weaker within radii of 300 m,
Park density 0.31 0.00 185.07
Density of street intersections 0.00 0.00 0.09
500 m, and 1000 m.
Density of road intersections 2.40 0.00 46.00 These patterns suggest that applying distance to destinations im-
pacts walking within most micro-level (as well as much wider) radii
Location attributes (vector L)
Distance to central business district 8501.94 21.67 17,647.34 because of heterogeneous local contexts, while the standardized index,
Distance to sub-CBD 4331.00 6.96 11,935.85 NQPDE, consistently changes walking mobility. The model outputs
Distance to metro stations 465.45 3.96 3729.61 suggest that street layouts with higher densities of destinations and
Distance to bus-stops 91.17 0.72 694.82 those requiring less cognitive difficulty to navigate are linked with
Distance to main retail-clustered areas 714.83 0.42 3633.39
Distance to schools 255.20 4.55 1685.50
more walking. As previous studies have shown, the greater the street
Distance to parks 535.84 2.75 2604.03 centrality (measured by closeness) to main destinations (e.g., schools
Distance to streets 10.92 0.00 428.79 and open spaces), the better the accessibility, which leads to greater
Distance to roads 38.55 0.00 599.34 numbers of pedestrians wanting to be in such places (Kim & Lee, 2016;
Distance to industrial property 214.11 1.79 1610.37
Koohsari, Karakiewicz, & Kaczynski, 2013).
Distance to large public property 187.77 3.19 1516.26
Distance to large retail stores 375.09 2.71 2798.12 Then, we test how betweenness within multiple network radii
Distance to office property 170.51 1.22 2127.64 changes the spatial patterns of walking. As shown in Fig. 6, walking
Distance to residential property 46.96 0.33 807.33 volume tends to be higher near spatial layouts with greater betweenness.
Distance to small retail stores 39.56 0.22 807.33 Links through which more entities pass are positively connected with a
Population and employment attributes (vector E) higher volume of walking. The effects are stronger as people moved
Population density 0.03 0.00 0.31 from 100 m to 1000 m, and then dropped within 2000 m. A link with a
Employment density 0.03 0.00 0.37
higher chance of movement within walking distance results in a greater
probability of walking. Moreover, betweenness captures positive ex-
4. Results and discussion ternalities of social opportunities to attract more walkers. Negative
externalities in neighborhoods, such as street crowdedness, congestion,
We identify the effects of street configurations on walking volume and obstacles, may explain the weaker effects within a 2000 m radius.
using multiple radius networks. The models contain 9848 cases within Prior research confirms that sites with higher betweenness have a strong
4264 census tract units. The model results for walking volume on effect on walking, housing prices, and the presence of economic ac-
business days and weekends show the overall R-squared (0.38 to 0.39), tivities (Porta et al., 2012; Wang, Chen, Xiu, & Zhang, 2014; Xiao et al.,
within-R-squared (0.22 to 0.24), and between-R-squared values (0.33 to 2016). Places that more frequently attract walkers lead to a greater
0.34). Owing to limited space, we provide only the results of the willingness to pay for property and provide promising venues for retail
1000 m radius (see Table 4). Three reasons explain the relatively low R- and other business activities.
Scholars argue that severance indicates greater barriers and a lower

35
C.-D. Kang Cities 71 (2017) 30–40

Table 4
Multilevel regression models for predicting the log of walking volume on business days (network radius = 1000 m, N = 9848, in 4264 census tracts).

Variables Closeness model Betweenness model Severance model Efficiency model

Coefficients VIF Coefficients VIF Coefficients VIF Coefficients VIF

Street configuration features (vector A)

log(Mean Euclidean distance) − 0.252⁎ 1.18
Network quantity penalized by distance, Euclidean 0.0731⁎⁎ 1.71
log(Betweenness) 0.0921⁎⁎⁎ 1.52
log(Mean crow flight distance) −0.354⁎⁎ 1.33
Diversion ratio −0.762⁎⁎⁎ 1.31
log(Convex hull maximum radius) − 0.24 1.31
Convex hull shape index − 0.273⁎⁎⁎ 1.14
Street attributes (vector S)
log(Sidewalk width) 0.270⁎⁎⁎ 1.3 0.265⁎⁎⁎ 1.3 0.268⁎⁎⁎ 1.3 0.271⁎⁎⁎ 1.3
log(Number of street ways) 0.109⁎⁎⁎ 2.55 0.105⁎⁎⁎ 2.55 0.105⁎⁎⁎ 2.56 0.113⁎⁎⁎ 2.63
Street furniture (yes = 1, no = 0) 0.113⁎⁎⁎ 1.03 0.112⁎⁎⁎ 1.03 0.115⁎⁎⁎ 1.03 0.114⁎⁎⁎ 1.03
Pedestrian streets with automobile roads = 1, pedestrian streets = 0 − 0.264⁎⁎⁎ 2.85 − 0.260⁎⁎⁎ 2.85 −0.265⁎⁎⁎ 2.85 − 0.262⁎⁎⁎ 2.85
Pedestrian streets with cycling roads = 1, pedestrian streets = 0 − 0.0934⁎ 1.12 − 0.0834⁎ 1.12 −0.0918⁎ 1.12 − 0.0966⁎⁎ 1.12
Crosswalk (yes = 1, no = 0) 0.144⁎⁎⁎ 1.58 0.136⁎⁎⁎ 1.58 0.142⁎⁎⁎ 1.58 0.144⁎⁎⁎ 1.58
Street slope (yes = 1, no = 0) − 0.0977⁎⁎⁎ 1.06 − 0.101⁎⁎⁎ 1.06 −0.0982⁎⁎⁎ 1.06 − 0.100⁎⁎⁎ 1.07
Neighborhood attributes (vector C)
Entropy index of seven land use − 0.388⁎⁎⁎ 3.41 − 0.399⁎⁎⁎ 3.4 −0.391⁎⁎⁎ 3.41 − 0.351⁎⁎⁎ 3.41
log(Net density of total floor area of industrial property) − 0.0100⁎ 2.2 − 0.00797⁎ 2.2 −0.0109⁎⁎ 2.19 − 0.0108⁎⁎ 2.19
log(Density of total floor area of public property) 0.0110⁎⁎⁎ 2.02 0.0124⁎⁎⁎ 2.02 0.0107⁎⁎⁎ 2.02 0.00988⁎⁎ 2.01
log(Density of total floor area of large retail stores) 0.0114⁎⁎ 1.95 0.0130⁎⁎ 1.95 0.0115⁎⁎ 1.95 0.0108⁎⁎ 1.95
log(Density of total floor area of office property) 0.00959⁎⁎ 3.3 0.00964⁎⁎ 3.3 0.00955⁎⁎ 3.3 0.00876⁎⁎ 3.3
log(Density of total floor area of residential property) 0.00699 1.85 0.00339 1.84 0.00821⁎ 1.81 0.00955⁎ 1.81
log(Density of total floor area of small retail stores) 0.0205⁎⁎⁎ 3.27 0.0176⁎⁎⁎ 3.25 0.0216⁎⁎⁎ 3.24 0.0220⁎⁎⁎ 3.24
log(Density of metro stations) − 0.000909 1.45 − 0.000771 1.45 −0.00104 1.45 − 0.000864 1.45
log(Density of bus-stops) 0.00228 1.6 0.00212 1.6 0.00268 1.6 0.00254 1.6
log(Park density) 0.0179⁎ 1.51 0.0136 1.51 0.0189⁎ 1.51 0.0192⁎ 1.51
log(Density of street intersections) − 0.0191⁎⁎ 1.11 − 0.0193⁎⁎ 1.09 −0.0167⁎ 1.09 − 0.0161⁎ 1.09
log(Density of road intersections) − 0.0975⁎⁎⁎ 2.23 − 0.0931⁎⁎⁎ 2.22 −0.100⁎⁎⁎ 2.22 − 0.097⁎⁎⁎ 2.22
Location attributes (vector L)
log(Distance to central business district) − 0.0163 1.78 − 0.0263 1.73 −0.0186 1.74 − 0.00311 1.72
log(Distance to sub-CBD) − 0.0218 1.15 − 0.0101 1.14 −0.0286 1.14 − 0.0207 1.14
log(Distance to metro stations) − 0.215⁎⁎⁎ 1.39 − 0.210⁎⁎⁎ 1.39 −0.214⁎⁎⁎ 1.39 − 0.216⁎⁎⁎ 1.41
log(Distance to bus-stops) − 0.139⁎⁎⁎ 1.44 − 0.135⁎⁎⁎ 1.44 −0.135⁎⁎⁎ 1.44 − 0.141⁎⁎⁎ 1.45
log(Distance to main retail-clustered areas) − 0.195⁎⁎⁎ 1.52 − 0.183⁎⁎⁎ 1.53 −0.193⁎⁎⁎ 1.52 − 0.194⁎⁎⁎ 1.53
log(Distance to schools) − 0.0620⁎⁎⁎ 1.22 − 0.0539⁎⁎⁎ 1.22 −0.0627⁎⁎⁎ 1.22 − 0.0586⁎⁎⁎ 1.22
log(Distance to parks) 0.0979⁎⁎⁎ 1.16 0.0974⁎⁎⁎ 1.15 0.0971⁎⁎⁎ 1.15 0.100⁎⁎⁎ 1.16
log(Distance to streets) − 0.0204⁎⁎ 1.17 − 0.0213⁎⁎ 1.15 −0.0222⁎⁎ 1.15 − 0.0228⁎⁎⁎ 1.15
log(Distance to roads) 0.0139⁎ 1.69 0.0225⁎⁎⁎ 1.71 0.0160⁎ 1.7 0.0128 1.72
log(Distance to industrial property) 0.0385⁎⁎ 1.75 0.0474⁎⁎⁎ 1.76 0.0388⁎⁎ 1.75 0.0376⁎⁎ 1.75
log(Distance to large public property) − 0.0709⁎⁎⁎ 1.51 − 0.0683⁎⁎⁎ 1.51 −0.0686⁎⁎⁎ 1.52 − 0.0712⁎⁎⁎ 1.51
log(Distance to large retail stores) − 0.107⁎⁎⁎ 1.75 − 0.0991⁎⁎⁎ 1.74 −0.108⁎⁎⁎ 1.73 − 0.106⁎⁎⁎ 1.74
log(Distance to office property) − 0.0498⁎⁎⁎ 2.24 − 0.0389⁎⁎ 2.23 −0.0527⁎⁎⁎ 2.2 − 0.0537⁎⁎⁎ 2.21
log(Distance to residential property) 0.125⁎⁎⁎ 2.27 0.129⁎⁎⁎ 2.24 0.121⁎⁎⁎ 2.24 0.122⁎⁎⁎ 2.24
log(Distance to small retail stores) − 0.148⁎⁎⁎ 1.97 − 0.138⁎⁎⁎ 1.96 −0.147⁎⁎⁎ 1.94 − 0.155⁎⁎⁎ 1.92
Population and employment density (vector E)
log(Population density) − 0.0207 2.07 − 0.0267⁎ 2.07 −0.0194 2.07 − 0.0212 2.07
log(Employment density) 0.0341⁎⁎⁎ 1.36 0.0334⁎⁎⁎ 1.36 0.0339⁎⁎⁎ 1.36 0.0337⁎⁎⁎ 1.36
Constant 12.73⁎⁎⁎ 10.07⁎⁎⁎ 14.45⁎⁎⁎ 12.99⁎⁎⁎
ICC 0.21 0.20 0.20 0.20
R-square
Within 0.22 0.23 0.22 0.22
Between 0.33 0.33 0.33 0.33
Overall 0.38 0.39 0.39 0.38

⁎
p < 0.05.
⁎⁎
p < 0.01.
⁎⁎⁎
p < 0.001.

connectivity among roads and street networks (Grisolia, Lopez, & de
Dios Ortuzar, 2015; Quigley & Thornley, 2011). Thus, the greater the
severance of links, the more unfavorable the environment is for walking.
As shown in Fig. 7, a higher severance indicates a longer crow flight
distance between origins and destinations (MCF), which is correlated
with less walking. A higher MCF distance of 100 m, 1000 m, and
2000 m (i.e., network radius) resulted in a lower volume of walking.
Furthermore, a higher diversion ratio, representing a shortest path
distance that is greater than the crow flight distance, is consistently
correlated with less walking in all network radii.
These key findings imply that walkers prefer short, direct routes to
reachable destinations, as well as street networks that require less
cognitive effort in terms of navigation, as confirmed previously by si-
milar metrics that show the effects of straightness on walking (Jones
et al., 2016; Kang, 2015).
With regard to the effects of efficiency, as shown in Fig. 8, a higher
maximum radius of the convex hull (HullR) had a consistently positive
influence on walking within a radius of up to 500 m; however, the ef-
fects disappeared as the radius grew. This positive correlation implies
that walking volume tends to be higher in networks where faraway
points are reachable from the origin (Cooper et al., 2014). The convex
hull shape index (HullSI) shows the opposite effects between radii of

36
C.-D. Kang Cities 71 (2017) 30–40

Fig. 5. Effects of closeness on walking.

Fig. 6. Effects of betweenness on walking.

Fig. 7. Effects of severance on walking.

37
C.-D. Kang
Fig. 8. Effects of efficiency on walking.
1000 m and 2000 m. A higher index indicates less similarity to a circle
and, thus, less walking. Note that the circular shape of a convex hull
benefits pedestrians, because such a street pattern covers multiple di-
rections. This finding confirms that circle-type network structures, with
their wider accessibility to other areas, lead to walkers being con-
centrated at the spatial level. Interestingly, the effects of the maximum
radius of a convex hull are concentrated more at the micro level, while
the circularity of a convex hull significantly affects the walking volume
within relatively wider radii. These results demonstrate that street
networks that cover wider areas and have higher navigability attract
more walkers.
4.2. Physical conditions of the street
With regard to the physical conditions of streets, highly spacious
streets with more lanes, facilities, and crosswalks are positively linked
with a greater walking volume (Cerin, Macfarlane, Ko, & Chan, 2007;
Cervero et al., 2009; Kang, 2015). In contrast, a mixed use of cars and
cycling on pedestrian streets creates an unfavorable urban setting for
walkers. These results are incompatible with those of Sung, Go, and
Choi (2013), but are compatible with those of other studies (Giles-Corti
et al., 2005). Owing to the combination of heterogeneous variables and
local contexts in prior research, we have not come to a decisive con-
clusion on the relationship between street conditions and walking vo-
lume.
4.3. Neighborhood attributes
The higher the land-use mix (among the seven types), the more
negative the association with walking volume. This contrasts with the
findings of previous studies, which conclude that a higher mixed land
use within buffers of 0.25–0.5 miles leads to more walking mobility
(McConville, Rodríguez, Clifton, Cho, & Fleischhacker, 2011). With re-
gard to the net density of each type of land use, more walkers con-
centrate in places with a greater number of retail stores, offices, and
public property, while fewer pedestrians tend to cluster near industrial
property. However, the net density of residential land use has no sta-
tistical significance. While a higher density of parks attracts more
walkers, a higher density of road and street intersections does not do so.
These patterns imply that people tend to walk near dense parks and
green spaces. However, more walkers are concentrated in pedestrian-
friendly, urban structures beyond highly dense intersection zones,
where roads and streets converge.
38
Cities 71 (2017) 30–40
4.4. Location attributes
Except for the distance to the urban center and sub-centers, the relative
location of transportation facilities, land use, and other urban functions are
found to change the walking volume at a 5% level of significance. The
models confirm that increased walking occurs closer to metro stations, bus
stops, streets, main retail areas, and education facilities. However, there
tend to be fewer walkers closer to roads and parks. Being closer to the
nearest heterogeneous type of land use affects walking volumes differently.
More people are likely to walk near small and large retail stores, offices, and
public facilities, whereas fewer people walk closer to housing, factories, or
warehouse properties.
4.5. Population and employment density
While worker density consistently creates a critical mass of pedes-
trians, but population density has an unstable statistical significance.
Daily travel near workplaces (e.g., commuting, business travel, face-to-
face meetings, and frequent pedestrian mobility between offices and
retail centers) is associated with a higher volume of walkers, a finding
that is, in general, supported by those of prior studies (Li, Fisher,
Brownson, & Bosworth, 2005).
5. Conclusion and policy implications
We isolated the effects of multiple features of street network con-
figurations on walking volume in Seoul, Korea. We used sDNA to ob-
jectively measure multiple metrics (closeness, betweenness, severance,
and efficiency) in order to estimate reliable and accurate links between
street configurations and walking volume. We used the effects of the
four metrics to capture street attributes that affect walking volume,
controlling for other relevant factors, within walkable neighborhoods.
Consistent with our core hypotheses, multiple features of street
layouts affect the spatial variations of walking volume in walkable
neighborhoods. Among our main findings, sites with closer destinations
and those with greater walking flow were associated with higher
walking volume, while there was less of a difference between the
straight-line and path distances from origins to destinations, which
tends to encourage greater walking mobility. A longer convex hull
maximum radius creates an advantageous urban setting for greater
walking within a radius of up to 500 m. However, a higher convex hull
shape index indicates a greater deviation from the shape of a circle,
which generated negative effects on walking within a radius of 1–2 km.
C.-D. Kang
In summary, we confirmed that the structure and design of street pat-
terns are associated with both walking behaviors and multiple dis-
tances.
Our core findings suggest a number of policy implications with regard to
the significant role of street designs in creating walkable, livable, and sus-
tainable cities. First, street configuration metrics are critical to identifying
the local relation between street layout and walking volume. These results
align with those of the existing literature, supporting the importance of the
spatial configuration of street networks in mediating between the spatial
interactions of urban form and mobility. Therefore, future urban planning
and urban design should consider macro-level accessibility analyses and the
micro-level characteristics of street configurations. In particular, it is ne-
cessary that the micro-level street layout be reflected in policies on walking
that are affected by micro urban settings. Second, we identified the urban
settings that benefit pedestrians within walkable neighborhoods. The em-
pirical models tell us that walkers usually prefer streets with dense desti-
nations, less difficult navigation, more frequent passing, and that cover
wider areas with convenient navigability. These key findings should be used
to create specific street designs and to measure their effects on walking
behaviors. Third, this study found that the features of street patterns at-
tractive to walkers varied according to neighborhood scales. Thus, we be-
lieve that customized street designs based on network radii that fit local
contexts would be more appealing to walkers. In future, if urban planners
and designers can determine the street configurations suitable for different
neighborhood scales, we will be better able to create pedestrian-friendly
streets. Finally, street layouts are not sufficient as walkable urban settings.
Thus, attractive road networks, mixed with neighborhood land use, would
be successful in shaping pedestrian-friendly urban structures (Kang, 2015).
Here, the ideal types of urban environments suggested by the empirical tests
are highly dense, with mixed retail and office land use, along streets with
dense destinations, convenient navigation, more frequent passing, and a
circular shape that covers wider areas.
Appendix A. Appendix
Metric Measurement
Closeness Mean Euclidean distance (MED)
Network quantity penalized by distance,
Euclidean (NQPDE)
Betweenness Betweenness
Severance Mean crow flight (MCF) distance
Diversion ratio
Efficiency Convex hull maximum radius (HullR)
Convex hull shape index (HullSI)
Notes
N: The set of links in street networks.
Rx: The set of links from link x in the given network radius.
P(y): The proportion of any link y within the radius. In continuous space, 0 ≤ P(y) ≤ 1.
dE(x, y): The Euclidean distance between an origin link x and a destination link y.
CFD(x, y): The crow flight distance between the centers of link x and y.
OD(y, z, x): The endpoints of the shortest path are y and z, not x when is measured.
Hullperimeter: The perimeter of convex hull covering street network within a radius.
Hullarea: The area of convex hull covering street network within a radius.
Source: Cooper (2016).
Cities 71 (2017) 30–40
In addition, we address whether the research framework can be applied
to other cases and research on urban studies and policies. First, we expect
the research framework and key findings of this study to be transferable to
cities in different contexts, because we have analyzed a diverse range of
neighborhoods in Seoul. If street network and walking-related data are
available, the cities can apply the analytic methods proposed here to capture
their relationship. Based on the main findings, it is also possible to design a
street policy suitable for a local context. In addition, comparative studies on
a global level will provide a useful reference when developing policies tai-
lored to local contexts. Second, we can apply the results of this study to
walking-relevant studies and policies. For instance, walking is closely linked
to citizen health, retail location, and public transportation. Therefore, street
features that attract walkers will enhance citizens' health and help in the
design of urban environments necessary for local economic prosperity. Our
results can also be applied to creating a walking environment that promotes
public transportation use. Thus, collaborative planning and implementation
by relevant planning authorities for land use, transportation, and urban
design are essential to creating livable and sustainable cities and neigh-
borhoods.
Lastly, this study has a few limitations. First, we fail to consider
street traffic, which contributes to features of neighborhoods. Traffic
volume along a street network affects walking behaviors. Thus, further
study of this topic would be helpful to better identify the complex links
among street networks, street traffic, and walking volume. Second, we
need to better understand the detailed causative mechanisms between
walking volume and the measured metrics of street configurations.
Finally, we utilized cross-sectional data to examine the effects of street
networks on walking volume. Thus, we have no knowledge of how
changes in street networks affect walking volume. If panel data on
street networks and walking volume are available, we might be able to
decipher dynamic correlations between multiple features of street
patterns and walking volume.
Mathematical Form
∑ dE (x , y ) P (y )
y ∈ RX
∑ P (y )
y ∈ RX
∑ P (y )
y ∈ RX
∑ dM (x , y )
y ∈ RX
∑ ∑ P (z ) OD (y, z , x )
y = N y ∈ RX
∑ CFD (x , y ) P (y )
y ∈ RX
∑ P (y )
y ∈ RX
∑
dM (x , y )
P (y )
CFD (x , y )
y ∈ RX
∑ P (y )
y ∈ RX
Maximum radius of the convex hull (with the center defined as the center of
the original link)
(Hullperimeter )2
4π (Hullarea)
39
C.-D. Kang
References
Agrawal, A. W., & Schimek, P. (2007). Extent and correlates of walking in the USA.
Transportation Research Part D: Transport and Environment, 12, 548–563.
Baran, P. K., Rodriguez, D. A., & Khattak, A. J. (2008). Space syntax and walking in a new
urbanist and suburban neighborhoods. Journal of Urban Design, 13, 5–28.
Cerin, E., Macfarlane, D. J., Ko, H.-H., & Chan, K.-C. A. (2007). Measuring perceived
neighbourhood walkability in Hong Kong. Cities, 24, 209–217.
Cervero, R. (1996). Mixed land-uses and commuting: Evidence from the American
Housing survey. Transportation Research Part A: Policy and Practice, 30, 361–377.
Cervero, R., & Kockelman, K. (1997). Travel demand and the 3Ds: Density, diversity, and
design. Transportation Research Part D: Transport and Environment, 2, 199–219.
Cervero, R., Sarmiento, O. L., Jacoby, E., Gomez, L. F., & Neiman, A. (2009). Influences of
built environments on walking and cycling: Lessons from Bogotá. International Journal
of Sustainable Transportation, 3, 203–226.
Chiaradia, A., Cooper, C., & Wedderburn, M. (2014). Network geography and accessibility.
Chiaradia, A., Hillier, B., Schwander, C., & Barnes, Y. (2013). Compositional and urban
form effects on residential property value patterns in Greater London. Proceedings of
the ICE-urban design and planning. vol. 166. Proceedings of the ICE-urban design and
planning (pp. 176–199).
Cooper, C. (2016). Spatial design network analysis (sDNA) version 3.4 Manual. Cardiff
University. Available, http://www.cardiff.ac.uk/sdna/software/documentation.
Cooper, C. H., & Chiaradia, A. J. (2015). sDNA: How and why we reinvented Spatial Network
Analysis for health, economics and active modes of transport. GIS Research UK
(GISRUK).
Cooper, C. H., Fone, D. L., & Chiaradia, A. J. (2014). Measuring the impact of spatial
network layout on community social cohesion: A cross-sectional study. International
Journal of Health Geographics, 13, 1–14.
Cooper, C., Chiaradia, A., & Webster, C. (2016). Spatial design network analysis software,
version 3.4. Cardiff Universityhttp://www.cardiff.ac.uk/sdna/.
Crane, R., & Crepeau, R. (1998). Does neighborhood design influence travel?: A beha-
vioral analysis of travel diary and GIS data. Transportation Research Part D: Transport
and Environment, 3, 225–238.
Cuthbert, A. (2007). Urban design: Requiem for an era – Review and critique of the last
50 years. Urban Design International, 12(3), 177–223.
Duncan, M. (2011). The impact of transit-oriented development on housing prices in San
Diego, CA. Urban Studies, 48, 1–27.
Ewing, R. (1995). Beyond density, mode choice, and single-purpose trips. Transportation
Quarterly, 49, 15–24.
Ewing, R., & Cervero, R. (2010). Travel and the built environment. Journal of the American
Planning Association, 76, 265–294.
Forsyth, A., Oakes, J. M., Schmitz, K. H., & Hearst, M. (2007). Does residential density
increase walking and other physical activity? Urban Studies, 44, 679–697.
Frank, L. D., & Pivo, G. (1994). Impacts of mixed use and density on utilization of three
modes of travel: Single-occupant vehicle, transit, and walking. Transportation
Research Record, 1466, 44–52.
Gehl, J. (2011). Life between buildings: Using public space. Island Press.
Giles-Corti, B., Broomhall, M. H., Knuiman, M., Collins, C., Douglas, K., Ng, K., ...
Donovan, R. J. (2005). Increasing walking: How important is distance to, attrac-
tiveness, and size of public open space? American Journal of Preventive Medicine, 28,
169–176.
Grisolia, J. M., Lopez, F., & de Dios Ortuzar, J. (2015). Burying the highway: The social
valuation of community severance and amenity. International Journal of Sustainable
Transportation, 9, 298–309.
Handy, S., Paterson, R. G., & Butler, K. (2003). Planning for street connectivity: Getting from
here to there. American Planning Association.
Hess, P. M., Moudon, A. V., Snyder, M. C., & Stanilov, K. (1999). Site design and pe-
destrian travel. Transportation Research Record, 9–19.
Hillier, B. (1996). Space is the machine. Space Syntax.
Holtzclaw, J. (1994). Using residential patterns and transit to decrease auto dependence and
costs. CA: Natural Resources Defense Council San Francisco.
Jones, T., Horton, D., Mullen, C., Pooley, C., Scheldeman, G., Strano, E., & Tight, M.
(2016). The role of street network connectivity and access to everyday facilities in
shaping everyday walking and cycling in English cities. Strategies for Sustainable
Mobilities: Opportunities and Challenges (pp. 15–38). Routledge.
Kang, C.-D. (2015). The effects of spatial accessibility and centrality to land use on
walking in Seoul, Korea. Cities, 46, 94–103.
Kim, H. J., & Lee, C. (2016). Does a more centrally located school promote walking to
school? Spatial centrality in school-neighborhood settings. Journal of Physical
40
Cities 71 (2017) 30–40
Activity & Health, 13, 481–487.
Kleinbaum, D., Lawrence, L., & Muller, K. (1988). Applied regression analysis and other
multivariate methods. PWS-Kent Publishing Company.
Koohsari, M. J., Karakiewicz, J. A., & Kaczynski, A. T. (2013). Public open space and
walking the role of proximity, perceptual qualities of the surrounding built en-
vironment, and street configuration. Environment and Behavior, 45, 706–736.
Lamíquiz, P. J., & López-Domínguez, J. (2015). Effects of built environment on walking at
the neighborhood scale. A new role for street networks by modeling their config-
urational accessibility? Transportation Research Part A: Policy and Practice, 74,
148–163.
Lee, C., & Moudon, A. V. (2006). The 3Ds + R: Quantifying land use and urban form
correlates of walking. Transportation Research Part D: Transport and Environment, 11,
204–215.
Li, F., Fisher, K. J., Brownson, R. C., & Bosworth, M. (2005). Multilevel modeling of built
environment characteristics related to neighborhood walking activity in older adults.
Journal of Epidemiology and Community Health, 59, 558–564.
Lopez-Zetina, J., Lee, H., & Friis, R. (2005). The link between obesity and the built en-
vironment. Evidence from an ecological analysis of obesity and vehicle miles of
Travel in California. Health & Place, 12, 656–664.
McConville, M. E., Rodríguez, D. A., Clifton, K., Cho, G., & Fleischhacker, S. (2011).
Disaggregate land uses and walking. American Journal of Preventive Medicine, 40,
25–32.
Park, S., Choi, Y., Seo, H., Moudon, A. V., Bae, C.-H. C., & Baek, S.-R. (2016). Physical
activity and the built environment in residential neighborhoods of Seoul and Seattle:
An empirical study based on Housewives' GPS walking data and travel diaries.
Journal of Asian Architecture and Building Engineering, 15(3), 471–478.
Peiravian, F., Derrible, S., & Ijaz, F. (2014). Development and application of the
Pedestrian Environment Index (PEI). Journal of Transport Geography, 39, 73–84.
Peponis, J., & Wineman, J. (2002). Spatial structure of environment and behavior.
Handbook of Environmental Psychology, 271–291.
Porta, S., Latora, V., Wang, F., Rueda, S., Strano, E., Scellato, S., ... Cormenzana, B.
(2012). Street centrality and the location of economic activities in Barcelona. Urban
Studies, 49, 1471–1488.
Quigley, R., & Thornley, L. (2011). Literature review on community cohesion and community
severance: Definitions and indicators for transport planning and monitoring. Report to New
Zealand Transport Agency. Wellington: Quigley and Watts Ltd.
Rabe-Hesketh, S., & Skrondal, A. (2008). Multilevel and longitudinal modeling using STATA.
STATA Press.
Rashid, M. (2016). The geometry of urban layouts: A global comparative study. Springer.
Rodríguez, D. A., Khattak, A. J., & Evenson, K. R. (2006). Can new urbanism encourage
physical activity? Comparing a new urbanist neighborhood with conventional sub-
urbs. Journal of the American Planning Association, 72, 43–54.
Sarkar, C., Webster, C., Pryor, M., Tang, D., Melbourne, S., Zhang, X., & Jianzheng, L.
(2015). Exploring associations between urban green, street design and walking:
Results from the Greater London boroughs. Landscape and Urban Planning, 143,
112–125.
Seoul Metropolitan Government (2010). Seoul pedestrian volume report.
Seoul Metropolitan Government (2013). Vision for pedestrian - Friendly city.
Seoul Institute (2015). The Seoul research data service. http://data.si.re.kr/.
Sevtsuk, A. (2010). Path and place: A study of urban geometry and retail activity in Cambridge
and Somerville, MA. Department of Urban Studies and Planning, Massachusetts
Institute of Technology (Doctoral Dissertation).
Shyr, O., Andersson, D. E., Wang, J., Huang, T., & Liu, O. (2013). Where do home buyers
pay most for relative transit accessibility? Hong Kong, Taipei and Kaohsiung com-
pared. Urban Studies, 50, 2553–2568.
Song, Y., & Knaap, G. J. (2004). Measuring urban form: Is Portland winning the war on
sprawl? Journal of the American Planning Association, 70, 210–225.
Sung, H.-G., Go, D.-H., & Choi, C. G. (2013). Evidence of Jacobs's street life in the great
Seoul city: Identifying the association of physical environment with walking activity
on streets. Cities, 35, 164–173.
Talen, E., & Koschinsky, J. (2013). The walkable neighborhood: A literature review.
International Journal of Sustainable Land Use and Urban Planning (IJSLUP), 1, 42–63.
Urban Information Network (2010). World city information. http://ubin.krihs.re.kr.
Wang, F., Chen, C., Xiu, C., & Zhang, P. (2014). Location analysis of retail stores in
Changchun, China: A street centrality perspective. Cities, 41, 54–63.
Xiao, Y., Webster, C., & Orford, S. (2016). Identifying house price effects of changes in
urban street configuration: An empirical study in Nanjing, China. Urban Studies, 53,
112–131.