The effects of spatial accessibility and centrality to land use on walking in Seoul, Korea

Cities 46 (2015) 94–103

Contents lists available at ScienceDirect

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

The effects of spatial accessibility and centrality to land use on walking in Seoul, 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

Article history: Received 24 September 2014 Received in revised form 13 May 2015 Accepted 15 May 2015 Available online 22 May 2015 Keywords: Spatial accessibility Centrality Land use Walking Seoul

a b s t r a c t The debate on pedestrian-friendly urban structures has increased interest in the connections among land use, accessibility, and pedestrian volume. Most econometric models have focused on the individual and separate effects of density, land-use patterns, and street connectivity on the spatial variation of walking. This study investigates the effects of spatial accessibility and centrality by land-use types on pedestrian presence in Seoul in 2009. The model employs four newly developed accessibility indices and identifies the differentiated effects of land-use accessibility and centrality on pedestrian volume, controlling for street features, location and transportation characteristics, and neighborhood land-use attributes. The model results confirm that the effects of land-use accessibility and centrality vary with the spatial distribution of pedestrians. This analysis highlights the importance of investigating accessibility effects by land-use volume. Indeed, policies on pedestrian-friendly urban structures should consider local contexts as well as the complex relationship between land-use accessibility and walking. Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction As we experience climate change and a lower quality of life in auto-oriented cities, smart growth and new urbanism are dominating urban paradigms, reshaping urban spatial structures worldwide. Naturally, urban planning, policy, and design aim to create cities that have a low auto-use, high pedestrian-friendly urban form. Walkable cities have certain advantages such as fewer auto trips, more street activity for local retail outlets and the community, and robust health for citizens. Thus, we need to understand the determinants of walking activities to encourage the creation of pedestrian-friendly urban structures. Many studies have focused on the main determinants of walking: socioeconomic features, the built environment, and street layout. These studies specifically suggest that population and employment density, land-use patterns, and land-use mix determine the pedestrian volume in cities. Within a specific area, higher population density has been shown to be associated with more walking (Agrawal & Schimek, 2007). However, other conditions are also required. For instance, high-density development, mixed-use areas, and comfortable public transit services influence the link between population density and walking behavior (Cervero & Kockelman, 1997; Holtzclaw, 1994). Proponents of smart growth and new urbanism have even suggested that higher

E-mail address: [email protected] http://dx.doi.org/10.1016/j.cities.2015.05.006 0264-2751/Ó 2015 Elsevier Ltd. All rights reserved.

density development converts auto usage into public transit travel and walking (Lopez-Zetina, Lee, & Friis, 2005). Boarnet and Crane (2001), by contrast, maintained that a higher density and a greater mixture of land development fail to reduce long-distance vehicle travel. A number of studies have also confirmed that increased retail land use near residential areas generates more walking choices. Thus, mixed-use residential and retail environments increase walking as a transportation mode (Cervero, 1996; Ewing, 1995; Frank & Pivo, 1994). The key finding of previous studies is that accessible locations for walkers are associated with diverse land use and a convenient street layout. On the contrary, street-focused studies maintain that street conditions and connectivity generate walking alternatives. Early studies of street effects focused on the influence of topological and physical settings on walking choices (Crane & Crepeau, 1998; Ewing & Cervero, 2001). Many authors have confirmed that street density and its link with cul-de-sacs affect walking behaviors (Cervero & Kockelman, 1997; Lee & Moudon, 2006; Song & Knaap, 2004). A wider focus on overall street fabric and connections has also promoted studies of catchment area and space syntax approaches, finding that walking is more likely to be concentrated near key destinations and along corridors between origins and destinations (Handy, Paterson, & Butler, 2003; Hess, Moudon, Snyder, & Stanilov, 1999). Among measurements of space syntax, integration and choice attract increased walking (Hillier, 1996; Peponis & Wineman, 2002). One study of the effects of space syntax features showed that pedestrians prefer more accessible

C.-D. Kang / Cities 46 (2015) 94–103

streets to those with shorter paths. In addition, local contexts change the connection between space syntax features and walking volume (Baran, Rodriguez, & Khattak, 2008). In summary, while the findings of previous studies have confirmed that population density, land-use patterns, and street layout determine the spatial variations of walking, research has not thus far tested the hypothesis that different access to land use based on the characteristics of a street network affects the spatial distribution of pedestrian presence. This statement is testable in a sophisticated framework because most walkers are sensitive to the integrated urban setting of varied land use and street layout. The present study intends to bridge this gap in the body of knowledge on this topic by testing the various effects of spatial accessibility and centrality by land-use type on walking volume in Seoul, Korea. In particular, this study compares the effects of spatial accessibility and centrality discussed in previous studies. However, as previous studies have focused on how spatial accessibility and centrality separately affect walking, we have less understanding about how their effects differ. To improve our knowledge of the linked impacts of land-use density and street network layout, this study thus applies popular and widely used spatial accessibility and centrality indices to identify their different effects on walking under one research framework. The remainder of the paper is organized as follows. The first section introduces the local contexts of the study area and data sources used for the empirical tests. The second section describes the motivation behind measuring spatial accessibility and centrality as well as presents and defines the variables and multilevel regression models deployed. The third section discusses and interprets the results of the regression analyses. In the final section, the study summarizes the results and suggests policy implications.

2. Study context and data sources Seoul is a highly dense city with high land-use mix and a well-organized public transit system. The population density of Seoul was 17,466 people per km2 in 2010, almost twice that of New York and 3.6 times that of London (Urban Information Network, 2010). Seoul operates nine subway lines at a total distance of 316.8 km with more than 290 stations and 11,200 bus stops with dedicated bus lanes, high service quality, and a convenient fare system. The total sidewalk length measured 2523 km2 in 2009 (Seoul Institute, 2014). The Seoul Metropolitan Government (SMG) has constantly strived to make Seoul a more livable, pedestrian-friendly, and sustainable city. In particular, since 2004, it has implemented innovative policies such as replacing the freeway with the Cheonggyecheon urban stream,1 reforming public transit services, and expanding green spaces to promote pedestrian- and public transit-oriented urban structures (Kang, 2009). In 2012, the SMG further integrated pedestrian-friendly urban settings into the public transit system and introduced car-free streets and transit malls to promote walking. Thus, micro-level pedestrian survey and land-use data in Seoul provide a great opportunity to test the link between relative access to diverse land use and pedestrian volume. We compiled data from a number of key sources, namely (1) location-based pedestrian volume data; (2) street maps; (3) 1 Cheonggyecheon is the corridor-type stream in the urban core of Seoul that replaced a freeway in 2005. After taking office in 2002, Mayor Lee initiated the project of converting the elevated freeway into an urban greenway to provide a pedestrian-friendly park along the stream.

95

Korean census tract (Jipgyegu) boundary maps2; (4) major transit stations and route maps; (5) bus stop maps; (6) building maps and built volume by land use3; and (7) a population and employment density map. In terms of point (1), the SMG counted the number of pedestrians at 9848 survey spots across Seoul in 2009 and matched these data with street-level information such as the width of sidewalks, furniture on sidewalks, types of streets, and nearby road types (Seoul Metropolitan Government, 2010). Fig. 1 indicates the spatial patterns of walkers on weekdays across Seoul (the average walking patterns on Saturdays were similar). To clarify these spatial patterns, this study represents average walking volume in Seoul’s census tract units by summing the walking volume of each survey spot. As shown by the figure, a large number of walkers are concentrated in the CBD, sub-CBDs, and places near main roads.

3. Methodology 3.1. Motivation behind measuring spatial accessibility and centrality Previous studies have ignored the combined effects of land use and street conditions. Recent works examining the built environment and street design have thus captured the attention of urban planners and designers. Intuitively, pedestrians consider urban settings containing the built environment and street layout (Ozbil, Peponis, & Stone, 2011). For example, more people tend to walk along well-connected streets with a higher density of commercial and retail spaces (Peponis, Hadjinikolaou, Livieratos, & Fatouros, 1989). Still, we have little knowledge on how land-use accessibility generates spatial variations in pedestrian volume. The accessibility and network centrality approaches open up new perspectives on the complex connection between land use and street features. While accessibility measures the ease of reaching destinations along streets or road networks, network centrality identifies the relative importance of nodes in a network (Geurs & van Wee, 2004; Hansen, 1959; Newman, 2010). Key studies of accessibility and its effects have verified that higher accessibility to jobs and land use generates higher housing prices and fewer vehicle miles (Cervero, 2005; Osland & Thorsen, 2008; Srour, Kockelman, & Dunn, 2002). In particular, higher accessibility to retail outlets and universities confers higher premiums on residential property values (Adair, McGreal, Smyth, Cooper, & Ryley, 2000; Franklin & Waddell, 2003; Song & Sohn, 2007). Empirical studies of network centrality have mainly suggested that higher ‘‘Closeness’’ and ‘‘Betweenness’’ are associated with higher housing prices and rent, population density, and commuting (Barthélemy & Flammini, 2009; Chiaradia, Hillier, Schwander, & Barnes, 2013). Other studies have used the network centrality concept to measure the complex flow of commuting in spatial networks (Caschili & De Montis, 2013; Reggiani, Bucci, & Russo, 2011). However, few studies have investigated how accessibility to land use changes walking behavior in metropolitan areas. To compare the different effects of spatial accessibility and centrality to land use on walking, this study applies the Gravity Index, Betweenness, Straightness, and Closeness. The popular use of such measures in previous studies justifies selecting these indices. The Gravity Index is one of the most popular and applied indices to evaluate spatial accessibility (Cervero, 2005; Handy & Niemeier, 2 Similar to census tracts in the United States, Jipgyegu is defined as the smallest statistical and spatial units for collecting socioeconomic data. Statistics Korea designates these units to be relatively homogeneous with respect to their demographic attributes, economic status, and living conditions. Each unit has an average of about 500 inhabitants and Seoul had 16,471 units in 2010 (Statistics Korea, 2014). The database of Korean census tracts for 2010 provides information on boundary maps, population, and employment. 3 The Seoul Building Registry of 2009 is the key source for building location and built volume by land use.

96

C.-D. Kang / Cities 46 (2015) 94–103

Fig. 1. Spatial patterns of walkers in seoul (Weekdays). Source: Seoul Metropolitan Government (2010).

1997; Hansen, 1959; Sevtsuk, 2014). Betweenness, Straightness, and Closeness are the most widely used among various spatial centrality measures (Kuby, Tierney, Roberts, & Upchurch, 2005; Turner, 2007; Wang, Antipova, & Porta, 2011; Xiao, Webster, & Orford, 2014).

Table 1 Accessibility indices and calculation methods. Source: modified from Sevtsuk and Mekonnen (2012). Accessibility indices

Calculation methods

Gravity index

Gravityr ½i ¼

Betweenness

3.2. Description of the variables

Straightness Closeness

This study uses spatial data, the UNA toolbox installable for ArcGIS, and a multilevel model to capture how spatial accessibility to nearby buildings determines walking behavior. The dependent variables are the daily average pedestrian volume on weekdays and Saturdays in Seoul in 2009. This study classifies various independent variables: (1) four land-use accessibility points, (2) street conditions, (3) land-use and transportation attributes in Seoul’s census tract units, (4) location and transportation features in the survey spots, and (5) socioeconomic features (e.g., population and employment density) in the census tract units.

3.2.1. Land-use accessibility and centrality To identify the effects of land-use accessibility to residential, retail, office, and industrial property on walking volume, this study obtains accessibility and centrality by using the UNA tool originally developed by the MIT City Form Lab and information from the Seoul Building Register in 2009. As shown in Table 1, the UNA tool calculates four accessibility and centrality indices of land use: the ‘‘Gravity Index,’’ ‘‘Betweenness,’’ ‘‘Closeness,’’ and ‘‘Straightness.’’ Notably, this approach uses a third element, building volume, with nodes and edges in a typical spatial network. Thus, this tool provides more reliable and accurate results for urban studies, planning, and design.

P

W½j

j2Gfig;d½i;jr

P

r

ebd½i;j n ½i

Betweenness ½i ¼ j2Gfig;d½i;jr njkjk  W½j P r d½i;j Straightness ½i ¼ j2Gfig;d½i;jr d½i;j  W½j r

Closeness ½i ¼ P

1

j2Gfigd½i;jr

ðd½i;jW½jÞ

Note: i: survey spots. j: buildings for each land use. G: network. r: network radius (500 m). d[i, j]: shortest path distance between origin node i and destination node j (m). d[i, j]: Euclidian distance between origin node i and destination node j (m). njk[i]: number of paths that pass through node i with j and k in the network radius r from i. njk: number of paths between nodes j and k. Beta(b): 0.00217. W(j): building floor area of destination node j(m2).

The Gravity Index, first developed by Hansen (1959), adds a spatial impedance factor to the Reach concept from origin to destination. The Reach measures the total building floor area within a given radius along the shortest path distance between origin i and destination j. This concept is similar to cumulative opportunities-type accessibility (Bhat et al., 2000). The exponential beta controls distance friction based on a distance decay function. The beta value varies between walking, cycling, and driving. This study applies 0.00217 as the beta value, as suggested by Handy and Niemeier (1997). The beta value depends on the behaviors of walkers, cyclists, and drivers in specific local contexts. The

C.-D. Kang / Cities 46 (2015) 94–103

Gravity Index is increasingly sensitive to distance. Indeed, studies have confirmed that the index significantly predicts land-use patterns, retail location choice, and the spatial patterns of employment (Hansen, 1959; Huff, 1963; Waddell & Ulfarsson, 2003). The Betweenness centrality estimates the value from the ratio of total passing at building i located along the shortest path relative to the total available path between buildings j and k, multiplied by the total volume of building j. Previous studies have proven that Betweenness greatly determines the location of retail outlets and services in built-up areas (Porta et al., 2009; Sevtsuk, 2010). Straightness measures how the shortest path distance between origin i and destination j within a given radius is similar to the straight line distance (Porta, Crucitti, & Latora, 2005; Vragovi, Louis, & Diaz-Guilera, 2005). Some studies have supported the value of this concept for identifying the direct connection among the surrounding buildings and higher landmarks visible from afar. The final index, the Closeness centrality, calculates how close each origin location is to the surrounding locations within a given threshold path distance. This measurement is beneficial for indicating how much building floor area weighted by each of the shortest path distances is close to the origin location within a given radius. Closeness calculates the inverse value of the whole summed building area weighted by the shortest distance. This study notes that a lower Closeness value means more total building volume weighted by each path distance between the origin and destination within a given distance threshold. Thus, we should interpret negative Closeness coefficients as positive effects on walking. This study classifies land use into residential, commercial, office, industrial, and other types based on data taken from the Seoul Building Register in 2009 (Brueggeman & Fisher, 2011; Ministry of Land, Infrastructure & Transport, 2014; Song & Sohn, 2007). Residential land use contains single-family and multi-family housing, while commercial land use includes central and neighborhood retail outlets, cultural facilities, hotels and motels, and recreational spaces. Office land use represents private office buildings, while industrial land use indicates factories, transportation facilities, and warehouses. Other land uses include power plants and religious facilities not classified into the previous four land-use categories. 3.2.2. Street conditions The SMG’s Walking Report provides information on the number of walkers and sidewalk widths (meters) of the survey spots, number of road lanes near these spots, street facilities, sidewalk types, nearby crosswalks, and street slopes. Street facilities indicate diverse installations such as street lamps and trees, public phone booths, and road signs, whereas sidewalk type is classified as walking/car mixed street, walking/cycle mixed street, and walker-dedicated street. This study codes the relevant variables 1 when streets have street facilities, walking/car mixed streets, walking/cycle mixed streets, nearby crosswalks, and slopes and 0 otherwise. 3.2.3. Census tract attributes of land use and transportation To capture the land-use characteristics in neighboring census tract units, this study measures the balance indexes of residen tial–nonresidential (commercial, office, industrial, and other land uses), residential–commercial, residential–office, residential–industrial, and commercial–office space by using Seoul Building Register data from 2009. The balance index ranges from 0 to 1, indicating a higher value with a higher balance between pairs of land uses (Cervero & Duncan, 2003). The example formula for calculating the balance index (Balancei) for residential and nonresidential land use is as follows (the same logic is also applied to measure the other balance indices):

  Res  Nonresi   Balancei ¼ 1   Res þ Nonresi 

97

ð1Þ

Further, we measure two land-use mixes by using Seoul’s census tract and commonly used entropy index (Cervero & Kockelman, 1997). This index is applied to measure the mix of the five land-use types (residential, commercial, office, industrial, and other). A value closer to 0 means mono land use, whereas a value closer to 1 indicates mixed land use. We measure two indices of land-use mix: including all land use (mix) and excluding residential use (nmix). Furthermore, this study measures the straight line distance from the survey spots to the nearest residential, commercial, office, and industrial buildings and calculates the densities of road and street intersections, subway stations, bus stops, and park areas per net area of Seoul’s census tracts. Park area represents the total area of green and open spaces according to the Seoul Biotope Map in 2010 (Seoul Metropolitan Government, 2014). Net area represents the total biotope area of residential, commercial and office, industrial, and infrastructure space in each census tract. The land-use mix index (Mixi) uses the entropy index and calculates the value based on the following formula:

Pn Mixi ¼

i¼1 P i

lnðP i Þ lnðnÞ

ð2Þ

where n = number of land-use types (five for land-use mix containing all land uses and four for land-use mix excluding residential) and Pi = proportion of each land use i. 3.2.4. Location and transportation attributes To control for location and transportation attributes, this study applies space syntax methods to measure the street layout. Among the indices of space syntax such as control, connection, global integration, local integration, mean depth, and total depth, only the control variable is included in the model because of the low variance inflation factor (VIF). Furthermore, this study measures the straight line distance to the urban core (i.e., the CBD) and the five nearest sub-CBDs, subway stations, bus stops, arterial roads, streets, major commercial areas, schools, and parks. Many previous studies have argued that accessibility to the CBD and sub-CBDs changes the spatial variation of populations, households, economic activities, and property values (see, for example, Ewing & Cervero, 2010). In urban spatial structures, transportation networks and public transit services increase walker volume. Finally, pedestrian presence is sensitive to access to clusters of retail outlets, schools, and parks. 3.2.5. Socioeconomic features Previous empirical tests have confirmed that higher population and employment are associated with an increased presence of walking. To control for socioeconomic features in Seoul’s census tract, this study calculates population and employment density, which are represented by the total number of residents and employees per net area of each census tract, respectively. 3.3. Multilevel regression models The data structure determines the method for analysis. This study applies multilevel regression models to isolate the effects of accessibility to different land uses on walking, considering that the data are measured in two different units: pedestrian-surveyed spots and Seoul’s census tract units. Among the explanatory variables, the survey spot variables are land-use accessibility, street conditions, and location and transportation. The census tract features of land use and transportation and socioeconomic features are measured by using Seoul’s census tracts. Thus, the multilevel regression model helps fix the

98

C.-D. Kang / Cities 46 (2015) 94–103

over- or under-estimation of parameters when the general regression is used (Rabe-Hesketh & Skrondal, 2008). Our multilevel regression uses the following equation:

Pij ¼ c00 þ b1 Aijk þ b2 Sijk þ b3 C ijk þ b4 Lijk þ b5 Eijk þ l0j þ eij

ð3Þ

where Pi.j = number of pedestrians in survey spot i (Level 1) in census tract j (Level 2); bk = model coefficients of variables (k = 1, 2, 3,. . .,m); r00 = model constants; Aijk = a vector of land-use accessibility in survey spot i (Level 1) in census tract j (Level 2); Sijk = a vector of street conditions in survey spot i (Level 1) in census tract j (Level 2); Cijk = a vector of attributes of land use and transportation in census tract j (Level 2); Lijk = a vector of attributes of location and transportation in survey spot i (Level 1) in census tract j (Level 2); Eijk = a vector of socioeconomic attributes in census tract j (Level 2); u0j, eij = the residual error terms of census tract units and survey spots, respectively. In this study, the function of the models takes a log–log form that converts the independent and dependent variables into natural logarithms (except for the dummy and nominal variables). Generally, we have no theoretical ground to support a specific functional form (Cassel & Mendelsohn, 1985; Duncan, 2010). Most econometric models apply untransformed, semi-log, or log– log models. These forms are applied in this study, with similar results produced in terms of the coefficient sign of variables and statistical significance. As the log–log model seems to show a better degree of fit from theoretical and empirical perspectives, this study uses those models. An additional interpretative benefit of the log–log form is that the estimated coefficients can be understood as mean attribute elasticities (Shyr, Andersson, Wang, Huang, & Liu, 2013). Further, the intraclass correlation (ICC) justifies the use of a multilevel regression because its value is higher than 0.05. To check the correlations among the explanatory variables, this study tests the VIF and finds that the models contain only explanatory variables with VIF values of less than 4. Tables 2 and 3 describe the variables and descriptive statistics.

4. Results The multilevel regression models contain 9848 survey spots (Level 1) and 4264 census tract units (Level 2). The results show that the overall R-square ranges from 0.37 to 0.38, the within R-square from 0.21 to 0.22, and the between R-square is 0.32 (see Tables 4–7). Three main reasons support the relatively low R-square values shown herein. First, this study focuses on ‘‘necessary activities,’’ namely daily activities carried out to reach specific destinations, as defined by Gehl (2011). The presented empirical models do not explain optional (e.g., relaxing) or social (e.g., playing and meeting) activities. Second, walking is a complex phenomenon influenced by non-spatial factors such as personal preferences, which cannot be captured in the presented models (Baran et al., 2008). Finally, the number of cases in this study is merely the 9848 spots selected by the SMG. Thus, we cannot expect a robust relationship between walking volume and the explanatory variables. Further, the ICC(rho) spread of each model is 0.20–0.21, indicating the appropriate use of multilevel models in this study. The interpreting coefficients limited to the variables reach a statistical significance of 10%. 4.1. Land-use accessibility and centrality Regarding the four main land-use options, Figs. 2 and 3 show the different effects of land-use accessibility on walking volume. The coefficients are taken from the land-use accessibility variables in Tables 4–7. Among these four types of land uses (i.e., residential, commercial, office, and industrial), higher accessibility to commercial land use is associated with higher pedestrian volume on weekdays and Saturdays. The effects of residential and office land use are weaker than those of commercial land use, while industrial land-use effects are negatively linked with walking. Commercial functions attract residents and visitors because of their more frequent daily destination-related activities, such as trips to grocery stores, retail locations, and restaurants. While residents limit

Table 2 Descriptive statistics 1. Variable description

Variable Code

Mean

Min

Max

Walking volume Number of average weekday walkers Number of average Saturday walkers

log WeekdaysPedestrian log SaturdayPedestrian

3061.14 2986.92

6 7

106,186 113,606

Land-use accessibility and centrality Gravity index of residential Gravity index of commercial Gravity index of office Gravity index of industrial Betweenness of residential Betweenness of commercial Betweenness of office Betweenness of industrial Straightness of residential Straightness of commercial Straightness of office Straightness of industrial Closeness of residential Closeness of commercial Closeness of office Closeness of industrial

logre_gra logcom_gra logof_gra logind_gra logre_bet logcom_bet logof_bet logind_bet logre_st logcom_st logof_st logind_st logre_clo logcom_clo logof_clo logind_clo

154,209 90,596 59,892 6411 8,266,839 3,643,277 1,165,861 98,692 233,127 137,231 90,686 10,044 0.000016 0.000004 0.0001 0.0002

0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.001 0.0000000001 0.0000000017 0.0000000018 0.0000000022

11,400,000 778,760 822,634 687,165 3,260,000,000 151,000,000 76,600,000 18,200,000 16,800,000 1,369,956 1,268,115 962,041 0.0095 0.0010 0.0010 0.0111

Street conditions Width of sidewalk Number of street lanes Presence of street furniture (yes = 1, no = 0) Walking/car mixed street = 1, pedestrian-only street = 0 Walking/bicycle mixed street = 1, pedestrian-only street = 0 Presence of nearby crosswalk (yes = 1, no = 0) Presence of street slope (yes = 1, no = 0)

logsidewalk_w loglane_no Street_Furniture Side_walk1 Side_walk2 Crosswalk Slope

3.959 2.879 0.923 0.399 0.053 0.462 0.250

1 1 0 0 0 0 0

24.3 18 1 1 1 1 1

99

C.-D. Kang / Cities 46 (2015) 94–103 Table 3 Descriptive statistics 2. Variable description

Variable Code

Mean

Min

Max

Census tract attributes of land-use and transportation Balance index for residential and commercial Balance index for residential and office Balance index for residential and industrial Balance index for commercial and office Entropy index of residential, commercial, office, industrial, and other land uses Entropy index of commercial, office, industrial, and other land uses Net density of road intersections, number of intersections per net area Net density of street intersections, number of intersections per net area Distance to nearest residential building Distance to nearest commercial building Distance to nearest office building Distance to nearest industrial building Net density of subway stations (number of stations per net area) Net density of bus stops (number of bus stops per net area) Total park area per net area

bal_resi_com bal_resi_off bal_resi_ind bal_com_off mix nmix logroad_den logst_den logne_resi_dis logne_com_dis logne_off_dis logne_ind_dis logsubw_den logbus_den logpark_den

0.510 0.274 0.152 0.323 0.501 0.396 2.398 0.002 46.956 38.251 170.610 214.073 0.000003 0.000059 0.309

0.00000022 0.00000004 0.00000004 0.00000029 0.00000711 0.00000623 0 0.000 0.186 0.097 1.230 1.473 0.000 0.000 0.000

1 1 1 1 1 1 46 0.088 807.426 807.426 2127.284 1609.852 0.003 0.003 185.066

Location and transportation attributes Control of space syntax Distance to CBD Distance to sub-CBD Distance to nearest subway stations Distance to nearest bus stops Distance to nearest arterial roads Distance to nearest streets Distance to main retail center Distance to nearest schools Distance to nearest park

logcontrol logcbd_dis logscbd_dis logsubw_dis logbus_dis logroad_dis logst_dis logcomz_dis logsch_dis logpark_dis

0.995 8501.940 4331.003 465.452 91.174 38.555 10.922 714.834 255.204 535.843

0.980 21.672 6.961 3.957 0.722 0.002 0.002 0.415 4.554 2.747

1.083 17647.340 11935.850 3729.608 694.818 599.344 428.786 3633.392 1685.502 2604.029

Socioeconomic features Net population density (people/net area) Net employment density (workers/net area)

logpop_den logemp_den

0.029 0.028

0.000 0.000

0.313 0.371

walking to accessing destinations such as retail outlets, workplaces, schools, and transit nodes, there are fewer walkers near office buildings because of time-dependent pedestrian flows. Most office workers tend to walk to reach their workplaces or specific destinations such as restaurants for lunch and business trips for face-to-face meetings. Among the four indices, Figs. 2 and 3 show the main finding that Closeness to commercial land is noticeably linked with higher walker volume. The positive and highest coefficients of the accessibility indices of commercial land indicate that more walkers are observed there. In particular, the higher Closeness of commercial land, indicating locations with higher built volumes within a threshold radius, generate the strongest effects on walking activities, followed by the Gravity Index, Straightness, and Betweenness in that order. Walkers may thus have a strong willingness to access local commercial opportunities that are reachable within a specific radius. These patterns show that pedestrians tend to cluster around retail outlets and other services with higher built volumes. Further, the next highest coefficients (i.e., for the Gravity Index) indicate that walking activities sensitively respond when spatial impedance is combined with built volume. Thus, walkers tend to be more concentrated near less spread-out areas with larger building volumes, denser street networks, and narrower gaps between buildings. Similarly, Franklin and Waddell (2003) used the gravity accessibility concept and found that greater access to retail outlets and universities increases housing prices. The parameters for Straightness imply that pedestrians prefer shorter direct routes to commercial activities. Wang et al. (2011) also verified that Closeness, Straightness, and Betweenness are highly associated with population and employment intensity. Finally, the effects of Betweenness are much weaker than those of the other accessibility indices, indicating fewer walkers along the main thoroughfares between two locations in general. Hence, walkers value direct accessibility and proximity to retail outlets

rather than geodesic paths for accessing the surrounding commercial destinations. However, previous studies have confirmed that retail and service establishments are more concentrated in areas with higher Betweenness (Porta et al., 2009; Sevtsuk, 2010). Controlling for the other key accessibility values in the same models may generate different outputs. Next, we find that the four indices of residential buildings in the weekdays and Saturdays models are positively linked with the spatial distribution of pedestrians but have relatively weaker marginal effects than those of commercial buildings. These patterns suggest that walkers might appear (although this is less likely than in commercial spaces) at places that are more accessible to people’s homes. Clearly, the variation in accessibility effects on walking volume is more modest compared with commercial buildings. These results confirm that residents value direct access to residential destinations instead of surrounding dwellings, street intensity, and accessibility to adjacent neighborhoods. By contrast, the marginal effects of accessibility to office buildings in the weekdays and Saturdays models are relatively flat, implying that accessibility to office-type workplaces and the surrounding built volume could be a weaker predictor of walking trips compared with the residential built volume. McConville, Rodriguez, Clifton, Cho, and Fleischhacker (2011) also found that higher accessibility to nonresidential land use containing offices and grocery stores is positively linked with increased walking activity. Finally, Closeness and the Gravity Index of industrial buildings in the weekdays models generate negative effects on pedestrian volume at a statistical significance of 10%. As people face negative externalities such as noise, dust, and freight movement near factories and warehouses, relatively few walkers tend to appear in those environments. In the Saturdays models, higher Betweenness is related to a higher spatial distribution of walkers, showing that pedestrians tend to appear near thoroughfares between industrial destinations. Most factories and warehouses tend to close at

100

C.-D. Kang / Cities 46 (2015) 94–103

Table 4 Multilevel models for predicting walker volume: gravity index. Variable

Model 3 (Weekdays)

Model 4 (Saturdays)

Coefficients

Coefficient

z

VIF

z

Land-use accessibility and centrality logre_gra 0.014⁄⁄⁄ logcom_gra 0.049⁄⁄⁄ logof_gra 0.007⁄⁄⁄ logind_gra 0.004⁄

3.620 6.790 3.300 1.670

0.012⁄⁄⁄ 0.05⁄⁄⁄ 0.006⁄⁄⁄ 0.003⁄⁄⁄

3.100 6.960 2.970 1.410

1.35 1.78 2.17 1.91

Street conditions logsidewalk_w loglane_no Street_Furniture Side_walk1 Side_walk2 Crosswalk Slope

15.790 6.350 4.100 9.960 2.240 7.610 5.480

0.266⁄⁄⁄ 0.108⁄⁄⁄ 0.122⁄⁄⁄ 0.261⁄⁄⁄ 0.079⁄⁄ 0.146⁄⁄⁄ 0.111⁄⁄⁄

15.960 6.840 4.210 9.830 2.170 7.560 5.910

1.30 2.56 1.03 2.85 1.12 1.58 1.07

Census tracts attributes of land-use and transportation bal_resi_com 0.039 0.920 0.038 bal_resi_off 0.052 1.270 0.044 ⁄⁄ bal_resi_ind 0.082 2.060 0.084⁄⁄ bal_com_off 0.004 0.100 0.021 Mix 0.094 1.230 0.101 nmix 0.059 1.010 0.06 4.450 0.077⁄⁄⁄ logroad_den 0.076⁄⁄⁄ logst_den 0.016⁄⁄ 2.240 0.017⁄⁄ logne_resi_dis 0.14⁄⁄⁄ 13.020 0.139⁄⁄⁄ logne_com_dis 0.155⁄⁄⁄ 13.610 0.156⁄⁄⁄ logne_off_dis 0.042⁄⁄⁄ 3.330 0.033⁄⁄⁄ logne_ind_dis 0.046⁄⁄⁄ 3.570 0.047⁄⁄⁄ logsubw_den 0.002 0.430 0.002 logbus_den 0.001 0.390 0.001 logpark_den 0.017⁄⁄ 2.020 0.016⁄

0.920 1.100 2.120 0.490 1.320 1.030 4.510 2.290 13.000 13.750 2.690 3.650 0.460 0.340 1.840

2.17 2.00 1.66 2.72 3.98 3.64 2.02 1.09 2.16 1.81 2.32 1.88 1.43 1.52 1.48

Location and transportation logcontrol 5.884⁄ logcbd_dis 0.058⁄⁄⁄ logscbd_dis 0.022 logsubw_dis 0.229⁄⁄⁄ logbus_dis 0.135⁄⁄⁄ logroad_dis 0.016⁄⁄ logst_dis 0.018⁄⁄⁄ logcomz_dis 0.205⁄⁄⁄ logsch_dis 0.069⁄⁄⁄ logpark_dis 0.103⁄⁄⁄

0.264⁄⁄⁄ 0.01⁄⁄⁄ 0.119⁄⁄⁄ 0.265⁄⁄⁄ 0.082⁄⁄ 0.147⁄⁄⁄ 0.104⁄⁄⁄

Socioeconomic features logpop_den 0.03⁄⁄⁄ logemp_den 0.039⁄⁄⁄ Constant 10.053⁄⁄⁄ Rho 0.20 Number of cases 9848 Number of groups 4264 R-square Within Between Overall

0.21 0.32 0.38

1.780 3.090 1.470 17.320 12.040 2.360 2.600 15.440 4.740 8.180

5.369 0.059⁄⁄⁄ 0.016 0.232⁄⁄⁄ 0.135⁄⁄⁄ 0.015⁄⁄ 0.018⁄⁄⁄ 0.207⁄⁄⁄ 0.065⁄⁄⁄ 0.101⁄⁄⁄

1.630 3.170 1.040 17.600 12.050 2.320 2.630 15.670 4.450 8.100

1.08 1.68 1.14 1.35 1.43 1.69 1.15 1.56 1.16 1.16

2.780 6.230 36.170

0.023⁄⁄ 0.037⁄⁄⁄ 9.98⁄⁄⁄ 0.20

2.150 5.900 36.060

1.86 1.33

0.21 0.32 0.38

weekends and therefore more walkers appear near these industrial properties. The findings of Franklin and Waddell (2003) supported this result, confirming that higher accessibility to industrial space is more likely to be associated with lower housing prices. 4.2. Physical street conditions Among street conditions, wider streets, more lanes and facilities on streets, and crosswalks provide more favorable urban settings for walking. These results concur with the findings of previous studies that these street conditions are positively correlated with walking activities (Cerin, Macfarlane, Ko, & Chan, 2007; Cervero, Sarmiento, Jacoby, Gomez, & Neiman, 2009). However, less walking is present on car and cycling roads mixed with walking streets. These patterns contrast with those presented by Sung, Go, and

Choi (2013) for Seoul, but agree with the findings of other studies (Giles-Corti et al., 2005; Moudon, Hess, Snyder, & Stanilov, 1997). These contrasting findings may be explained by variable discrepancies controlling for spatial accessibility and centrality to the built volume of each land use and by the disparity in variable composition. 4.3. Census tract attributes of land use and transportation In the balance indices of land use, only those of residential and industrial space have negative effects on walking. The natural explanation for this finding is that neighborhoods with mixed residential and industrial spaces have fewer walking opportunities. The other balance indices do not reach statistical significance at the 10% level. The two indices of land-use mix are also not

Table 5 Multilevel models for predicting walker volume: betweenness. Variable

Land use accessibility logre_bet logcom_bet logof_bet logind_bet

Model 5 (Weekdays)

Model 6 (Saturdays)

Coefficient

Coefficient

z

VIF

z

0.007⁄⁄⁄ 0.012⁄⁄⁄ 0.007⁄⁄⁄ 0.002

2.710 3.560 4.350 1.490

0.006⁄⁄ 0.012⁄⁄⁄ 0.006⁄⁄⁄ 0.003

2.380 3.840 4.030 1.620

1.97 2.18 2.34 2.03

0.265⁄⁄⁄ 0.107⁄⁄⁄ 0.121⁄⁄⁄ 0.259⁄⁄⁄ 0.086⁄⁄ 0.141⁄⁄⁄ 0.101⁄⁄⁄

15.830 6.720 4.160 9.710 2.370 7.290 5.330

0.267⁄⁄⁄ 0.114⁄⁄⁄ 0.123⁄⁄⁄ 0.254⁄⁄⁄ 0.083⁄⁄ 0.14⁄⁄⁄ 0.109⁄⁄⁄

16.000 7.200 4.280 9.570 2.290 7.260 5.770

1.30 2.56 1.03 2.85 1.12 1.58 1.06

Census tracts attributes of land-use and transportation bal_resi_com 0.046 1.100 0.046 bal_resi_off 0.049 1.190 0.042 bal_resi_ind 0.078⁄⁄ 1.960 0.081⁄⁄ bal_com_off 0.003 0.070 0.014 mix 0.097 1.260 0.104 nmix 0.066 1.110 0.066 ⁄⁄⁄ logroad_den 0.077 4.470 0.078⁄⁄⁄ logst_den 0.018⁄⁄ 2.410 0.018⁄⁄ logne_resi_dis 0.138⁄⁄⁄ 12.720 0.137⁄⁄⁄ logne_com_dis 0.153⁄⁄⁄ 13.370 0.154⁄⁄⁄ logne_off_dis 0.042⁄⁄⁄ 3.320 0.033⁄⁄⁄ logne_ind_dis 0.065⁄⁄⁄ 5.070 0.064⁄⁄⁄ logsubw_den 0.002 0.490 0.002 logbus_den 0.001 0.430 0.001 logpark_den 0.018⁄⁄ 2.080 0.016⁄

1.100 1.030 2.050 0.320 1.350 1.130 4.520 2.460 12.700 13.490 2.660 5.090 0.520 0.370 1.890

2.16 2.00 1.66 2.73 3.98 3.64 2.01 1.09 2.24 1.83 2.36 1.89 1.43 1.52 1.48

Location and transportation logcontrol 5.886⁄ logcbd_dis 0.044 logscbd_dis 0.027⁄ logsubw_dis 0.228⁄⁄⁄ logbus_dis 0.128⁄⁄⁄ logroad_dis 0.017⁄⁄⁄ logst_dis 0.02⁄⁄⁄ logcomz_dis 0.219⁄⁄⁄ logsch_dis 0.072⁄⁄⁄ logpark_dis 0.095⁄⁄⁄

Street conditions logsidewalk_w loglane_no Street_Furniture Side_walk1 Side_walk2 Crosswalk Slope

Socioeconomic features logpop_den 0.031⁄⁄⁄ logemp_den 0.039⁄⁄⁄ Constant 10.384⁄⁄⁄ Rho 0.21 Number of Cases 9848 Number of Groups 4264 R-square Within Between Overall

0.22 0.32 0.37

1.770 2.350 1.820 17.170 11.310 2.620 2.980 16.710 4.900 7.580

5.389 0.046⁄⁄ 0.021 0.23⁄⁄⁄ 0.127⁄⁄⁄ 0.017⁄⁄⁄ 0.02⁄⁄⁄ 0.221⁄⁄⁄ 0.066⁄⁄⁄ 0.094⁄⁄⁄

1.630 2.470 1.370 17.440 11.320 2.590 3.000 16.920 4.570 7.530

1.08 1.65 1.13 1.35 1.45 1.69 1.15 1.51 1.15 1.16

2.840 6.190 39.73

0.024⁄⁄ 0.036⁄⁄⁄ 10.299⁄⁄⁄ 0.21

2.190 5.860 39.590

1.86 1.34

0.22 0.32 0.37

101

C.-D. Kang / Cities 46 (2015) 94–103

density alone is not always a favorable urban setting for walking. Hence, walkers desire a pedestrian-friendly urban structure.

Table 6 Multilevel models for predicting walker volume: straightness. Variable

Model 7 (Weekdays)

Model 8 (Saturdays)

Coefficient

Coefficient

z

VIF

z

4.4. Location and transportation attributes

Land use accessibility logre_st 0.015⁄⁄⁄ logcom_st 0.044⁄⁄⁄ logof_st 0.007⁄⁄⁄ logind_st 0.003

3.840 6.150 3.460 1.450

0.013⁄⁄⁄ 0.044⁄⁄⁄ 0.007⁄⁄⁄ 0.003

3.330 6.290 3.130 1.180

1.34 1.72 2.10 1.83

Street conditions logsidewalk_w loglane_no Street_Furniture Side_walk1 Side_walk2 Crosswalk Slope

15.830 6.320 4.090 9.960 2.260 7.600 5.500

0.267⁄⁄⁄ 0.108⁄⁄⁄ 0.121⁄⁄⁄ 0.261⁄⁄⁄ 0.08⁄⁄ 0.146⁄⁄⁄ 0.112⁄⁄⁄

16.000 6.810 4.200 9.830 2.190 7.550 5.930

1.30 2.56 1.03 2.85 1.12 1.58 1.07

Census tracts attributes of land-use and transportation bal_resi_com 0.04 0.960 0.04 bal_resi_off 0.052 1.270 0.044 ⁄⁄ bal_resi_ind 0.082 2.060 0.084⁄⁄ bal_com_off 0.005 0.120 0.022 mix 0.097 1.260 0.103 nmix 0.06 1.030 0.061 4.460 0.077⁄⁄⁄ logroad_den 0.077⁄⁄⁄ logst_den 0.017⁄⁄ 2.270 0.017⁄⁄ logne_resi_dis 0.139⁄⁄⁄ 12.990 0.139⁄⁄⁄ logne_com_dis 0.157⁄⁄⁄ 13.810 0.158⁄⁄⁄ logne_off_dis 0.043⁄⁄⁄ 3.480 0.035⁄⁄⁄ logne_ind_dis 0.047⁄⁄⁄ 3.780 0.048⁄⁄⁄ logsubw_den 0.002 0.430 0.002 logbus_den 0.001 0.360 0.001 logpark_den 0.017⁄⁄ 2.020 0.016⁄

0.950 1.090 2.130 0.510 1.350 1.050 4.520 2.330 12.970 13.960 2.840 3.850 0.460 0.300 1.850

2.17 2.00 1.66 2.72 3.98 3.64 2.02 1.09 2.16 1.79 2.27 1.81 1.43 1.52 1.48

Location and transportation logcontrol 5.877⁄ logcbd_dis 0.058⁄⁄⁄ logscbd_dis 0.022 logsubw_dis 0.23⁄⁄⁄ logbus_dis 0.135⁄⁄⁄ logroad_dis 0.016⁄⁄ logst_dis 0.018⁄⁄⁄ logcomz_dis 0.207⁄⁄⁄ logsch_dis 0.068⁄⁄⁄ logpark_dis 0.103⁄⁄⁄

0.265⁄⁄⁄ 0.1⁄⁄⁄ 0.119⁄⁄⁄ 0.265⁄⁄⁄ 0.082⁄⁄ 0.147⁄⁄⁄ 0.104⁄⁄⁄

Socioeconomic features logpop_den 0.03⁄⁄⁄ logemp_den 0.039⁄⁄⁄ Constant 10.093⁄⁄⁄ Rho 0.20 Number of Cases 9848 Number of Groups 4264 R-square Within 0.22 Between 0.32 Overall 0.37

1.770 3.090 1.480 17.350 12.050 2.390 2.630 15.580 4.670 8.170

5.364 0.059⁄⁄⁄ 0.016 0.232⁄⁄⁄ 0.135⁄⁄⁄ 0.015⁄⁄ 0.018⁄⁄⁄ 0.219⁄⁄⁄ 0.064⁄⁄⁄ 0.101⁄⁄⁄

1.620 3.170 1.050 17.630 12.060 2.350 2.650 15.820 4.380 8.090

1.08 1.68 1.14 1.35 1.43 1.69 1.15 1.56 1.16 1.16

2.790 6.280 36.300

0.023⁄⁄ 0.037⁄⁄⁄ 10.024⁄⁄⁄ 0.20

2.160 5.940 36.210

1.86 1.33

0.22 0.32 0.37

statistically significant. The greater explanatory power of land-use accessibility and other variables may generate no estimated effects of the land-use mix. McConville et al. (2011), however, confirmed that higher land-use mix within 0.25- and 0.5-mile buffers generates more walking behavior. In addition, proximity to the nearest land use produces different effects on walking. More walkers tend to appear near commercial and office buildings, whereas fewer are near residential and industrial buildings. These model outputs indicate that walkers tend to appear near retail and office buildings. Higher park density is positively associated with more walking, while higher road and street density negatively affect walking. These results show that more people tend to walk in neighborhoods that have a higher volume of parks and green spaces. However, roads for automobiles are not conducive to walking, and the model finds that higher street

By using the space syntax approach, we find that higher control over the street layout is linked to more walking, consistent with the findings of Baran et al. (2008) that confirmed that higher control and global integration are positively linked with increased leisure walks, while higher local integration negatively affects these trips. Except for distance to sub-CBDs, all location and transportation attributes change the pedestrian volume. In particular, consistent with the theory and research, increased walking occurs around the CBD, subway stations, bus stops, streets, major retail clusters, and schools. Numerous studies have found that access

Table 7 Multilevel models for predicting walker volume: closeness. Variable

Model 9 (Weekdays)

Model 10 (Saturdays)

Coefficient

Coefficient

Land use accessibility logre_clo 0.021⁄⁄⁄ logcom_clo 0.06⁄⁄⁄ logof_clo 0.012⁄⁄⁄ logind_clo 0.009⁄⁄

z

VIF

z

3.670 6.500 3.160 2.330

0.017⁄⁄⁄ 0.061⁄⁄⁄ 0.01⁄⁄⁄ 0.009⁄⁄

3.110 6.670 2.710 2.240

1.47 2.14 2.44 1.75

15.810 6.320 4.090 10.060 2.280 7.570 5.590

0.267⁄⁄⁄ 0.108⁄⁄⁄ 0.121⁄⁄⁄ 0.264⁄⁄⁄ 0.081⁄⁄ 0.145⁄⁄⁄ 0.113⁄⁄⁄

15.980 6.820 4.200 9.930 2.210 7.530 6.030

1.30 2.56 1.03 2.85 1.12 1.58 1.06

Census tracts attributes of land-use and transportation bal_resi_com 0.04 0.970 0.04 bal_resi_off 0.055 1.360 0.047 ⁄ bal_resi_ind 0.067 1.700 0.071⁄ bal_com_off 0.006 0.150 0.023 mix 0.096 1.260 0.104 nmix 0.054 0.920 0.055 logroad_den 0.078⁄⁄⁄ 4.520 0.078⁄⁄⁄ logst_den 0.016⁄⁄ 2.190 0.016⁄⁄ logne_resi_dis 0.141⁄⁄⁄ 13.130 0.14⁄⁄⁄ logne_com_dis 0.157⁄⁄⁄ 13.910 0.159⁄⁄⁄ logne_off_dis 0.04⁄⁄⁄ 3.290 0.033⁄⁄⁄ logne_ind_dis 0.045⁄⁄⁄ 3.710 0.045⁄⁄⁄ logsubw_den 0.002 0.430 0.002 logbus_den 0.001 0.340 0.001 logpark_den 0.016⁄ 1.910 0.015⁄

0.970 1.170 1.790 0.540 1.360 0.950 4.570 2.260 13.110 14.090 2.700 3.680 0.460 0.290 1.740

2.17 2.00 1.66 2.72 3.98 3.64 2.02 1.09 2.16 1.79 2.24 1.73 1.43 1.52 1.49

Location and transportation logcontrol 5.727⁄ logcbd_dis 0.058⁄⁄⁄ logscbd_dis 0.02 logsubw_dis 0.227⁄⁄⁄ logbus_dis 0.137⁄⁄⁄ logroad_dis 0.015⁄⁄ logst_dis 0.017⁄⁄ logcomz_dis 0.198⁄⁄⁄ logsch_dis 0.069⁄⁄⁄ logpark_dis 0.103⁄⁄⁄

Street conditions logsidewalk_w loglane_no Street_Furniture Side_walk1 Side_walk2 Crosswalk Slope

0.265⁄⁄⁄ 0.1⁄⁄⁄ 0.119⁄⁄⁄ 0.268⁄⁄⁄ 0.083⁄⁄ 0.147⁄⁄⁄ 0.106⁄⁄⁄

Socioeconomic features logpop_den 0.031⁄⁄⁄ logemp_den 0.038⁄⁄⁄ Constant 9.239⁄⁄⁄ Rho 0.20 Number of cases 9848 Number of groups 4264 R-square Within Between Overall

0.21 0.32 0.38

1.730 3.090 1.320 17.210 12.150 2.340 2.460 14.770 4.690 8.240

5.257 0.059⁄⁄⁄ 0.014 0.23⁄⁄⁄ 0.136⁄⁄⁄ 0.015⁄⁄ 0.017⁄⁄ 0.201⁄⁄⁄ 0.064⁄⁄⁄ 0.102⁄⁄⁄

1.590 3.160 0.910 17.490 12.160 2.300 2.490 15.040 4.410 8.150

1.08 1.73 1.15 1.35 1.44 1.70 1.15 1.60 1.16 1.16

2.920 6.050 28.120

0.024⁄⁄ 0.036⁄⁄⁄ 9.225⁄⁄⁄ 0.20

2.270 5.730 28.180

1.86 1.34

0.21 0.32 0.38

102

C.-D. Kang / Cities 46 (2015) 94–103

0.07

Marginal Effects

0.06 0.05 0.04

Residential

0.03

Commercial

0.02

Office

0.01

Industrial

0 -0.01 -0.02 Gravity Index Betweenness

Straightness

Closeness

Fig. 2. Marginal effects of the four accessibility types on walker volume (Weekdays).

0.07

Marginal Effects

0.06 0.05 0.04 Residential

0.03

Commercial

0.02

Office

0.01

Industrial

0 -0.01 -0.02 Gravity Index Betweenness Straightness

Closeness

Fig. 3. Marginal effects of the four accessibility types on walker volume (Saturdays).

to urban centers and transportation nodes is associated with walking (Ewing & Cervero, 2010). However, proximity to roads and parks leads to less walking, indicating that relatively few pedestrians are present near roads for automobiles and green spaces. 4.5. Socioeconomic features Employment density is positively associated with higher neighborhood walking. Population density, however, generates the opposite effect in contrast to a previous study that did not consider the employment factor in a nearby neighborhood (Agrawal & Schimek, 2007). The positive coefficients of employment density indicate that more walkers tend to be observed near workplaces (e.g., for commuting, business trips, and face-to-face interactions) than near residential neighborhoods.

residential buildings. Accessibility, measured by using Closeness and Gravity Index values for industrial buildings on weekdays, has a negative effect on pedestrian volume because of the increased nuisance from industrial properties. In the Saturday models, Betweenness for industrial space was found to be positively associated with walking activities. The key findings of this study are relevant to urban planners and designers for creating pedestrian-friendly urban structures. The econometric models show that the spatial distribution of walkers is strongly associated with commercial activities. Thus, a well-organized configuration of commercial land use along street networks would improve the design of pedestrian-friendly urban settings. The model outputs imply that a larger built volume along a denser street network is a critical factor for encouraging walking. By contrast, there is less spatial variation in access to office and residential buildings for walkers across Seoul. However, although accessibility to commercial buildings has remarkable effects on pedestrian patterns, a sophisticated link among diverse land uses still needs to be discovered in order for urban planners and designers to create a harmonious setting among various land uses. In addition, the four accessibility indices considering built density and street networks were shown to influence walking activities to differing degrees. For all land uses, the Closeness centrality was ranked first, followed by the Gravity Index, Straightness, and Betweenness. These accessibility effects indicate which urban setting (i.e., land-use volumes and street networks) within the surrounding neighborhood favors the promotion of walking in cities. Indeed, communication and cooperation between institutions related to land use and street configuration help improve the reliability of urban design (Cervero, 2005). As shown in this study, close integration among land-use volume, urban form, and street networks decisively changes walking behavior. Finally, urban designers should turn their attention to an accessibility concept that combines land-use density and street fabrics. As the model results indicate, these complex features fusing land-use patterns and street connectivity substantially determine walking behavior. In conclusion, this study confirmed the effects of access to the built volume of various land uses along street networks on walking activities. If the results are firmly tested and generalized, they could be applied to create more effective urban designs. Further, by including the considered preferences of pedestrians, the key findings could contribute to shifting auto-oriented urban structures toward pedestrian-friendly urban settings. In terms of future research directions, further studies could examine how reinforcing pro-pedestrian policy relieves urban issues such as traffic congestion, energy overuse, and climate change.

5. Conclusion and policy implications References This study identified how land-use accessibility is associated with walking patterns in Seoul, Korea, by using data from 2009. The presented network analysis containing built density and street features estimated more reliable and accurate features, generating new perspectives on the close connections among land use, street networks, and walking. We identified the effects of four accessibility indices of four land uses, controlling for street conditions, the census tract features of land use, location and transportation at the survey spots, and socioeconomic variables. The results of the current study support the initial hypothesis that combining the features of land-use density, urban form, and street configuration determines walking activities. The model results of our multilevel regression suggest that walkers prefer urban settings in order to access commercial buildings, such as large building volume, densely spaced areas, and dense street networks; however, these effects are relatively weak for office and

Adair, A., McGreal, S., Smyth, A., Cooper, J., & Ryley, T. (2000). House prices and accessibility: The testing of relationships within the Belfast Urban Area. Housing Studies, 15, 699–716. 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 neighbourhoods. Journal of Urban Design, 13, 5–28. Barthélemy, M., & Flammini, A. (2009). Co-evolution of density and topology in a simple model of city formation. Networks and Spatial Economics, 9, 401–425. Bhat, C., Handy, S., Kockelman, K., Mahmassani, H., Chen, Q., & Weston, L. (2000). Urban accessibility index: Literature review. Austin: Texas Department of Transportation. Boarnet, M. G., & Crane, R. (2001). Travel by design: The influence of urban form on travel. NC, USA: Oxford University Press. Brueggeman, W., & Fisher, J. (2011). Real estate finance and investments. New York: McGraw-Hill. Caschili, S., & De Montis, A. (2013). Accessibility and complex network analysis of the U.S. commuting system. Cities, 30, 4–17. Cassel, E., & Mendelsohn, R. (1985). The choice of functional forms for hedonic price equations: Comment. Journal of Urban Economics, 18, 135–142.

C.-D. Kang / Cities 46 (2015) 94–103 Cerin, E., Macfarlane, D. J., Ko, H. H., & Chan, K. C. A. (2007). Measuring perceived neighbourhood walkability in Hong Kong. Cities, 24(3), 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. (2005). Accessible cities and regions: A framework for sustainable transport and urbanism in the 21st century. UC Berkeley Center for Future Urban Transport. Cervero, R., & Duncan, M. (2003). Walking, bicycling, and urban landscapes: Evidence from the San Francisco Bay Area. American Journal of Public Health, 93, 1478–1483. 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(4), 203–226. 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, 166, 176–199. Crane, R., & Crepeau, R. (1998). Does neighborhood design influence travel?: A behavioral analysis of travel diary and GIS data. Transportation Research Part D: Transport and Environment, 3, 225–238. Duncan, M. (2010). The impact of transit-oriented development on housing prices in San Diego, CA. Urban Studies, 48, 1–25. Ewing, R. (1995). Beyond density, mode choice, and single-purpose trips. Transportation Quarterly, 49, 15–24. Ewing, R., & Cervero, R. (2001). Travel and the built environment: A synthesis. Transportation Research Record, 87–113. Ewing, R., & Cervero, R. (2010). Travel and the built environment. Journal of the American Planning Association, 76, 265–294. 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. Franklin, J. P., & Waddell, P. (2003).A hedonic regression of home prices in King County, Washington, using activity-specific accessibility measures. In Proceedings of the transportation research board 82nd annual meeting. Washington, DC. Gehl, J. (2011). Life between buildings: Using public spaces. Island Press. Geurs, K. T., & van Wee, B. (2004). Accessibility evaluation of land-use and transport strategies: Review and research directions. Journal of Transport Geography, 12, 127–140. Giles-Corti, B., Broomhall, M. H., Knuiman, M., Collins, C., Douglas, K., Ng, K., et al. (2005). Increasing walking: How important is distance to, attractiveness, and size of public open space? American Journal of Preventive Medicine, 28(2), 169–176. Handy, S., Paterson, R. G., & Butler, K. (2003). Planning for street connectivity: Getting from here to there. American Planning Association. Handy, S., & Niemeier, A. D. (1997). Measuring accessibility: An exploration of issues and alternatives. Environment and Planning A, 29, 1175–1194. Hansen, W. G. (1959). How accessibility shapes land use. Journal of the American Institute of Planners, 25, 73–76. Hess, P. M., Moudon, A. V., Snyder, M. C., & Stanilov, K. (1999). Site design and pedestrian 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. San Francisco, CA: Natural Resources Defense Council. Huff, D. L. (1963). A probabilistic analysis of shopping center trade areas. Land Economics, 39, 81–90. Kang, C. (2009). Land market impacts and firm geography in a green and transitoriented city: The case of Seoul, Korea, Dissertation, Department of City and Regional Planning, University of California, Berkeley. Kuby, M., Tierney, S., Roberts, T., & Upchurch, C. (2005). A comparison of geographic information systems, complex networks, and other models for analyzing transportation network topologies. NASA/CR-2005-213522. 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. Lopez-Zetina, J., Lee, H., & Friis, R. (2005). The link between obesity and the built environment. Evidence from an ecological analysis of obesity and vehicle miles of travel in California. Health Place, 12, 656–664.

103

McConville, M. E., Rodriguez, D. A., Clifton, K., Cho, G., & Fleischhacker, S. (2011). Disaggregate land uses and walking. American Journal of Preventive Medicine, 40(1), 25–32. Ministry of Land, Infrastructure, and Transport, Korea. (2014). Architectural law. Moudon, A. V., Hess, P. M., Snyder, M. C., & Stanilov, K. (1997). Effects of site design on pedestrian travel in mixed-use, medium-density environments. Transportation Research Record: Journal of the Transportation Research Board, 1578(1), 48–55. Newman, M. (2010). Networks: An introduction. Oxford: Oxford University Press. Osland, L., & Thorsen, I. (2008). Effects on housing prices of urban attraction and labor-market accessibility. Environment and Planning A, 40, 2490–2509. Ozbil, A., Peponis, J., & Stone, B. (2011). Understanding the link between street connectivity, land use and pedestrian flows. Urban Design International, 16, 125–141. Peponis, J., Hadjinikolaou, E., Livieratos, C., & Fatouros, D. A. (1989). The spatial core of urban culture. Ekistics, 56, 43–55. Peponis, J., & Wineman, J. (2002). Spatial structure of environment and behavior. Handbook of Environmental Psychology, 271–291. Porta, S., Crucitti, P., & Latora, V. (2005). The network analysis of urban streets: A primal approach. Environment and Planning B, 35, 705–725. Porta, S., Latora, V., Wang, F., Strano, E., Cardillo, A., Scellato, S., et al. (2009). Street centrality and densities of retail and services in Bologna, Italy. Environment and Planning B: Planning and Design, 36, 450–465. Rabe-Hesketh, S., & Skrondal, A. (2008). Multilevel and longitudinal modeling using STATA, STATA Press. Reggiani, A., Bucci, P., & Russo, G. (2011). Accessibility and impedance forms: Empirical applications to the german commuting network. International Regional Science Review, 34, 230–252. Seoul Institute (2014). Seoul INFO, . Seoul Metropolitan Government (2010). Seoul pedestrian volume report. Seoul Metropolitan Government (2014). Seoul biotope map. Sevtsuk, A. (2010). Path and place: A study of urban geometry and retail activity in cambridge and somerville. Cambridge, USA: MIT. Sevtsuk, A. (2014). Location and agglomeration: The distribution of retail and food business in dense urban environments. Journal of Planning Education and Research, 34(4), 374–393. Sevtsuk, A., & Mekonnen, M. (2012). Urban network analysis. A new toolbox for ArcGIS. Revue Internationale de Géomatique, 22, 287–305. 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 Compared. 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. Song, Y., & Sohn, J. (2007). Valuing spatial accessibility to retailing: A case study of the single family housing market in Hillsboro, Oregon. Journal of Retailing and Consumer Services, 14, 279–288. Srour, I., Kockelman, K., & Dunn, T. (2002). Accessibility indices: Connection to residential land prices and location choices. Transportation Research Record: Journal of the Transportation Research Board, 1805, 25–34. Statistics Korea (2014), Statistics Korea Website, . 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. Turner, A. (2007). From axial to road-centre lines: A new representation for space syntax and a new model of route choice for transport network analysis. Environment and Planning B: Planning and Design, 34, 539–555. Urban Information Network (2010), World City Information, . Vragovi, I., Louis, E., & Diaz-Guilera, A. (2005). Efficiency of information transfer in regular and complex networks. Physics Review E, 71(036122), 1–9. Waddell, P., & Ulfarsson, G. F. (2003). Accessibility and agglomeration: Discretechoice models of employment location by industry sector. In 82nd annual meeting of the transportation research board, Washington, DC, Citeseer. Wang, F., Antipova, A., & Porta, S. (2011). Street centrality and land use intensity in Baton Rouge, Louisiana. Journal of Transport Geography, 19(2), 285–293. Xiao, Y., Webster, C., & Orford, S. (2014). Identifying house price effects of changes in urban street configuration: An empirical study in Nanjing, China. Urban Studies, in press (doi: 10.1177/0042098014560500), 1–20.