The S + 5Ds: Spatial access to pedestrian environments and walking in Seoul, Korea

Cities xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

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

The S + 5Ds: Spatial access to pedestrian environments and walking in Seoul, Korea Chang-Deok Kang Dept. of Urban Planning and Real Estate, Chung-Ang University, 84 Heukseok-ro, Dongjak-gu, Seoul 06974, South Korea

A R T I C L E I N F O

A B S T R A C T

Keywords: Pedestrian environments Accessibility Centrality Walking volume Multilevel regression Seoul

Previous studies have confirmed that pedestrian environments affect walkers' behaviors in terms of density, diversity, design, destination accessibility, and distance to transit. However, street configuration connects walkers with other built environments. Few studies have yet determined the effects on walking volume by measuring street configurations combined with other walking environments. Thus, by capturing spatial accessibility and centrality to pedestrian environments, this study determines how street configuration, combined with pedestrian environments, affects pedestrian mobility. Our multilevel regression models verify the positive effects of higher access to primary destinations, such as neighborhood retail stores, schools, cultural facilities, and available public transit along the local street network on walking volume. Furthermore, while areas with a higher volume of pedestrian accidents and nonresidential land use are the main venues for walkers, access to residential density and parks has negative or no effects on pedestrian mobility. Finally, the effects of spatial access to pedestrian environments varies with multiple neighborhood scales. The key findings imply the significance of spatial access to pedestrian environments on walking volume. Thus, policies to create pedestrianfriendly neighborhoods should consider the relationship between spatial access to pedestrian environments and walking behaviors.

1. Introduction As auto-oriented cities face climate change, urban sprawl, traffic congestion, air pollution, and lower quality of life, creating walkable neighborhoods has been the core principle of sustainable and livable cities. Thus, many relevant studies have long been concerned with which urban settings create pedestrian-friendly cities and how the urban spatial structure should be reformed. Specifically, the diverse benefits of walkable cities justify identifying the determinants of walking choice and suggesting policy implications. These benefits include reducing the effects of climate change, pollution, and noise, enhancing public health and local economic performance, and higher social cohesion (Talen & Koschinsky, 2013). Previous studies have examined the determinants of walking behavior, the specific role of streets in walkable urban settings, and the interconnected effects of streets and other built environments on pedestrian mobility. Many studies have sought the main determinants of walking, including population and employment density, land use diversity, accessibility to destinations, and access to public transits and have verified that a higher density of residents, a higher diversity of land use, and easier access to land uses are strongly associated with higher pedestrian volume (Agrawal & Schimek, 2007). Higher density

means that more people live in a specific neighborhood. Dense neighborhoods tend to be safer due to more “eyes on the street” and more accessible services (Loo & Chow, 2006). A mix of residential, commercial, office, and other spaces is positively correlated with walking and cycling, as the local diversity of daily destinations promotes walking between destinations and lower automobile use (Koh & Wong, 2013; Manaugh & Kreider, 2013). Furthermore, easy access to the central business district (CBD), work, and retail stores decreases car use and increases the propensity to walk (Cervero, 2006). Finally, increased access to public transit creates a favorable urban setting that increases walking volume (Bento, Cropper, Mobarak, & Vinha, 2003; Rajamani, Bhat, Handy, Knaap, & Song, 2003). These pedestrian environments have been conceptualized as the 5Ds; namely density, diversity of land use, and design including safety and amenity (the 3Ds), and destination accessibility and distance to transit (the 2Ds) (Cervero & Kockelman, 1997; Ewing & Cervero, 2010). While previous studies have considered the various determinants of pedestrian-friendly environments, newer research has focused on the specific effects of street networks on walking mobility. The fundamental perspective is that streets are not equivalent to other built environments in terms of urban spatial structure. Design components comprise features such as street environments, amenity, safety, and street density.

E-mail address: [email protected]. https://doi.org/10.1016/j.cities.2018.01.019 Received 17 August 2017; Received in revised form 16 November 2017; Accepted 17 January 2018 0264-2751/ © 2018 Elsevier Ltd. All rights reserved.

Please cite this article as: Kang, C.-D., Cities (2018), https://doi.org/10.1016/j.cities.2018.01.019

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

have established key hypotheses on the effects of spatial access to diverse built environments on walking volume. First, as previous studies have confirmed, the 5Ds combined with street configuration differently affect walking choice and behaviors. We expect destination accessibility, access to transit, and nonresidential land use diversity along the specific street layout to be positively associated with walking volume, while diversity of overall land use, residential density, access to parks, and pedestrian accidents will confer negative or no effects. In terms of public transit, we expect that access to bus-stops will be more associated with pedestrian volume than access to metro stations, because pedestrians tend to gather near bus-stops to transfer to metro transit (Kitamura, Mokhtarian, & Laidet, 1997; Targa & Clifton, 2005). Furthermore, many studies have verified that easier access to daily destination, mixing nonresidential land use, and access to public transit affect pedestrian volume (Handy, Cao, & Mokhtarian, 2006; Lee & Moudon, 2006). However, the existing literature has verified the inconsistent relation between resident density and diversity of overall land use and pedestrian presence (Cervero, Sarmiento, Jacoby, Gomez, & Neiman, 2009; Frank, Kavage, Greenwald, Chapman, & Bradley, 2009; Zhang, 2004). As walkers and traffic accidents tend to concentrate at specific spots, we expect a positive association between them (Mueller, Rivara, Lii, & Weiss, 1990; Rothman et al., 2014). Second, the effects on variations in pedestrian volume of spatial accessibility and centrality metrics combined with given built environments (such as Reach, Gravity Index, Betweenness, Straightness, and Closeness to pedestrian environments) vary with neighborhood scales. Comparing the metrics of accessibility and centrality provides the specific pattern of relation between pedestrian environment and walking, among other associations (Kang, 2016; Wang, Antipova, & Porta, 2011; Xiao, Webster, & Orford, 2016). This study anticipates that walkers are highly attracted to higher destination accessibility, access to transit, and nonresidential land use diversity along easily reachable, highly accessible, directly-routed, and highly dense street networks. However, locations with highly detoured passing tend to decrease walking presence, as Kang (2015) has confirmed. Third, the neighborhood scales of the effects suggest specific implications for more effective urban design. Thus, many previous studies have attempted to capture impact zones (Kang, 2017; Sarkar et al., 2015). We expect that neighborhood scales of the effects will vary with spatial access to diverse built environments. The remainder of this study consists of four sections. The first describes the features of the study case and data for the empirical models. The second provides background regarding measuring spatial access to pedestrian environments, variables for the empirical tests, and the multilevel regression models. Specifically, the Background section provides more detailed information. The third section interprets key findings from the empirical models. The fourth discusses the implications of the core results. The final section summarizes this study and suggests the further studies.

Among street features, the density of street connections used as a proxy for street block size increases the propensity for walking (Jacobs, 1961). Street-focused empirical models have verified that street setting and network connection play different roles in promoting walking behaviors. Mainly, studies of topological and physical settings have found that higher street density and well-linked networks are positively associated with walking choices (Crane & Crepeau, 1998; Song & Knaap, 2004). Quantitative measurement of street networks by catchment area and Space Syntax methods have expanded the perspectives on street networks. The empirical models have discovered that pedestrians tend to concentrate along frequent-use corridors to reach their destinations (Handy, Paterson, & Butler, 2003). Specifically, higher integration and choice of street networks measured by Spatial Syntax methods are positively associated with attracting more walkers along shorter paths (Hillier, 1996; Peponis & Wineman, 2002). Since local features of street networks alter the link between street design and pedestrian volume, the street density, connectivity, and block size of given areas have been used to explain and predict variations in walking behavior (Baran, Rodríguez, & Khattak, 2008). Recent studies with more advanced perspectives have focused on the combined features of street layout and other built environments to generate variations in walking and cycling (Vale, Saraiva, & Pereira, 2015). The blended features of walking environments substantially affect pedestrian volume and choice. This new framework allows both street layout and other built environments to be considered as a single factor to explain and predict local walking behaviors. First, walking behavior tends to respond to accessibility, both implicitly and explicitly, including density, diversity, distance to destination, and route characteristics. Thus, diverse methods to measure accessibility utilize the relevant information of socioeconomic features, land use patterns, and street features (Dong, Ben-Akiva, Bowman, & Walker, 2006; ElGeneidy & Levinson, 2011; Miller, 2005). More advanced measurements, walkability, and the pedestrian environment index comprise the combined measurement of several built environments (Peiravian, Derrible, & Ijaz, 2014). Second, concurrently measuring both street configuration and other pedestrian environments generates more reliable factors affecting pedestrian mobility. Intuitively, built environments combined with street layout tend to influence walking behavior (Ozbil, Peponis, & Stone, 2011). Third, built environments of destinations along walkable street networks are substantially associated with pedestrians' spatial patterns. Specifically, higher access to commercial than to residential and office use are strongly correlated with walking volume (Kang, 2015). Thus, we need to consider two key factors of pedestrian environments, destinations and neighborhood scale, to capture more reliably the connection between spatial access to built environments and walking behaviors (Kang, 2017). To date, however, the effects of built environments have not adequately explained or predicted urban walking behavior. Furthermore, most studies have neglected the fact that street layout is not identical to other built environments in terms of walking choice. Rather, street configuration connects walkers with other factors related to pedestrian behaviors. Further, we have less understanding of the neighborhood scales from which effects occur. The accumulation of empirical tests on pedestrian environments and street-focused studies support new insights into how the combination of street configuration and other built environments affects variation in walking volume over multiple spatial scope. Thus, this study raises the unexplored research question of how the combined features of street configuration and other pedestrian environments affect walking volume. We expect this new approach to expand the existing discussion and provide more insight for academic and professional communities. This study also examines whether the research framework of Western studies can be generalized to East Asian cities. Particularly, the use of Seoul as a case allows us to investigate unexplored questions and hypotheses due to its diverse geographical information. Referring to previous studies and considering local contexts, we

2. Methods 2.1. Study case and data sources Seoul is the largest urban area and capital of South Korea. It is characterized by high population density and mixed land use (housing, retail, offices, manufacturing and warehouses, and other uses) that are well-served by public transit. In terms of pedestrian environments, the length of the sidewalk in Seoul in 2011 was 2789 km (10.25 km2), with 169 pedestrian overpasses and 88 pedestrian underpasses (Seoul Institute, 2017). Since the 1990s, the Seoul Metropolitan Government (SMG) has attempted to create livable, pedestrian-friendly, and sustainable urban settings to deal with traffic congestion and environmental pollution. The SMG has two policy directions: creating pedestrian-friendly streets and converting car space into pedestrian space. From 2012 to 2015, the SMG supported actions to improve districts and 2

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

local corridors for pedestrians, a public transit system combined with a pedestrian-friendly urban setting, local car-free zones, and a transit mall to promote walking (Lee, 2016). As a result, space previously used by automobiles was converted into space for pedestrians. From 2004 to 2005, the SMG began replacing freeway-channeling auto-flow in central areas with an urban greenway, including a linear park and walking paths (Kang & Cervero, 2009). This study collects key data to employ in empirical models, namely (1) spot-based walking volume data from the Pedestrian Survey of Seoul; (2) road and street network maps; (3) Korean census tract (Jipgyegu) data and boundary maps; (4) public transit maps for the subway station and bus stops; (5) land use data and maps; (6) residential units information; (7) park area data and maps; and (8) point maps of the CBD and sub-CBD. The Pedestrian Survey of Seoul in 2009 reported the number of walkers at specific spots according to street environments, including width of sidewalk; number of street lanes; street furniture; type of street (e.g., for use by pedestrians only, walkers and cars, or walkers and cyclists); presence of a crosswalk; and slope (Seoul Metropolitan Government, 2010). This study solely utilizes measurable and available geographical information since data on sense of place, subjective feeling of space, and familiarity of pedestrians are not available to determine walking choice. The SMG and Statistics Korea provided additional data and geographical information. Fig. 1 illustrates the spatial patterns of pedestrian-counted spots and street networks across Seoul. There were 9848 spots along micro-level street networks.

Fig. 2. Conceptual framework for the components of the S + 5Ds.

environments, including street environments, street configuration, and neighborhood built environments connect closely in real urban settings, influencing walkers' behaviors and choice of transport mode. Thus, capturing spatial access to the built environments of each neighborhood within a single framework is more reliable and accurate in explaining and predicting pedestrians' responses. After reviewing previous studies on the concepts and effects of built environments, the link between street conditions, street configuration, and neighborhood built environments, and spatial accessibility and centrality metrics, this study suggests a new model named the “S + 5Ds,” which combines street configuration with the 5Ds (Fig. 2). This model clarifies that the 5Ds affect pedestrian behaviors and choice of transport mode through street configuration.

2.2. Background to measuring spatial access to pedestrian environments As noted, many previous studies have confirmed that the 3Ds—density, diversity, and design—and the additional 2Ds—destination accessibility and distance to public transit—change the spatial variation of walking. While the 5D model is useful in identifying the specific built environments that affect pedestrian mobility, more robust models are required to isolate the features of pedestrian environments along street networks. The models of Moudon and Lee (2003) and Seo (2006) highlight that neighborhood pedestrian

Fig. 1. Spatial patterns of pedestrian-counted spots and street networks in Seoul.

3

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

Fig. 3. Conceptual framework of spatial access to pedestrian environments.

and three centrality metrics, namely Betweenness, Straightness, and Closeness, to analyze urban street layout (Turner, 2007; Wang et al., 2011). UNA is an alternative method of Space Syntax methodology. While Space Syntax methodology focuses on topological features such as the number of connections, rather than distance between connected lines (called dual representation), UNA uses metric distance to measure the connection between graph elements (called primal representation) (Hillier, 1996; Sevtsuk & Mekonnen, 2012). Both methods commonly focus on nodes and edges as the unit of analysis. The remarkable characteristic of UNA is that it can be employed to weigh the features of neighborhood pedestrian environments, such as density, diversity, design, building area of destination, and transit ridership, while capturing the features of street configuration. Two reasons support the application of the UNA methodology (Sevtsuk & Mekonnen, 2012). First, the variation and features of pedestrian environments lead to uneven walking distribution. For instance, higher density and diversity, better design, voluminous built areas at the destination, and higher transit ridership are associated with higher walking volume. Second, previous approaches, namely edge and node-focused network analyses, ignored the connection between the features of pedestrian environments and pedestrian presence. Thus, most studies have failed to capture the link between pedestrian environments and walking along the street configuration. In contrast, measuring street layout by weighting each attribute of pedestrian environments creates more accurate and reliable accessibility and centrality metrics to explain and predict spatial variations in walking volume. Measuring spatial access to pedestrian environments through the UNA requires three components: (1) straight-line and shortest path distance between origin i and destination j, (2) network radius, and (3) features of destinations. Fig. 3 illustrates the spatial access to pedestrian environments within a specific network radius. This study sets pedestrian-counted spots as origins and pedestrian environments of the census tract or specific areas as destinations. We selected 500 m, 1 km, and 2 km as the network radii to capture the association between pedestrian environments and walking in walkable neighborhoods. A Korean study by Park et al. (2016) justifies the neighborhood ranges. They confirmed that homemakers in Seoul walk up to 2.6 km per day on average. Finally, this study applied the UNA to measure the spatial access to pedestrian environments listed in Table 1. The features of

Street network analysis measuring spatial accessibility and centrality metrics enables a new perspective to capture the close link between street layout and pedestrian environments using a single framework. While accessibility metrics identify the ease of approaching diverse destinations along streets, centrality metrics measure the relative significance of origins within street networks (Geurs & Van Wee, 2004; Hansen, 1959). Some empirical studies have verified that improved accessibility to retail and job destinations is associated with higher residential property prices (Adair, McGreal, Smyth, Cooper, & Ryley, 2000; Cervero, 2005; Franklin & Waddell, 2003; Osland & Thorsen, 2008; Song & Sohn, 2007). Measuring street centrality is a promising perspective for understanding the urban street fabric and its relationship with urban phenomena such as residential value and rent, demographic concentration, and travel flows (Caschili & De Montis, 2013; Chiaradia, Hillier, Schwander, & Barnes, 2013; Reggiani, Bucci, & Russo, 2011). For the empirical tests, this study connected each concept of the built environment and street configuration with available data, with reference to existing studies (Table 2). 2.3. Description of the variables 2.3.1. Spatial access to pedestrian environments Most studies disregard the integrated effects of street configuration and pedestrian environments on walking. However, recent studies have highlighted how the urban setting connecting street layout with neighborhood built environments changes pedestrian behaviors (Ozbil et al., 2011). Specifically, well-linked streets with dense retail spaces attract more pedestrians (Peponis, Hadjinikolaou, Livieratos, & Fatouros, 1989). Furthermore, some studies have attempted to identify the association between the 3Ds and walking routes to retail and school destinations, by comparing the effects of straight-line and network distance to destinations with the walking response to a buffer range (Lee & Moudon, 2006). However, few studies have investigated the combined effects of street configuration and pedestrian environments. This study employs an Urban Network Analysis (UNA), developed by Sevtsuk, Mekonnen, and Kalvo in the City Form Lab, to identify spatial access to neighborhood built environments along a specific street network (Sevtsuk & Mekonnen, 2012). The method produces two accessibility indexes, the Reach and Gravity Index, to measure spatial accessibility (Cervero, 2005; Handy & Niemeier, 1997; Hansen, 1959), 4

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

Table 1 Description of pedestrian environments. Pedestrian environment

Feature of destinations (weight)

Spatial units

Relevant studies

Density Diversity

Residential density Entropy index of all land use Entropy index of nonresidential land use Number of pedestrian accidents Park areas Total built areas of main destination Transit ridership

Census tracts Census tracts

Cervero et al. (2009) Targa and Clifton (2005) Frank, Kerr, Chapman, and Sallis (2007) Cervero et al. (2009) Kitamura et al. (1997) Lee and Moudon (2006) Næss (2005) Kitamura et al. (1997) Targa and Clifton (2005)

Design: access to amenities Destination accessibility Distance to transit

Pedestrian accident spots Locations of parks Census tracts Metro stations Bus stops

for convenience shopping within walkable neighborhoods (Kang, 2016). The first centrality index, Betweenness, quantifies the total number of passing at pedestrian-counted spots i per all potential number of paths between destinations j and other potential locations k, weighted by the features of destinations j. Because pedestrian-counted spots i require the value of detour passing to explain the variation in pedestrian volumes, Betweenness assumes that locations j and k are origindestination pairs in measuring all potential numbers of paths within the given network radius. Previous studies have verified that venues with more detours are favorable locations for retail stores while higher detouring access to land use tends to reduce walking volume (Kang, 2015; Sevtsuk, 2014). Straightness indicates how the shortest path distance resembles straight-line distance, weighted by pedestrian environments (Porta, Crucitti, & Latora, 2006). Higher Straightness implies a more direct route to destinations and distinctive visibility of origins (Kang, 2016). Finally, Closeness is the inverse sum of the shortest distance multiplied by the value of each pedestrian environment. A lower value for Closeness implies a higher value for pedestrian environments within the given radius. Thus, we must interpret the negative coefficients of Closeness as positively associated with walking volume.

Table 2 Mathematical forms for measuring spatial access to pedestrian environments. Metrics

Calculation forms

Reach

Reachr[i] = ∑j ∈ G − {i}; d [i, j] ≤ r W [j]

Gravity Index Betweenness Straightness Closeness

W [j]

Gravityr[i] = ∑j ∈ G − {i}; d [i, j] ≤ r

Betweennessr[i] = Straightnessr[i] =

Closenessr[i] =

ε β ∙ d [i, j] njk [i] ∑j ∈ G − {i}; d [i, j] ≤ r ∙W njk

δ [i, j] ∑j ∈ G − {i}; d [i, j] ≤ r ∙W d [i, j]

[j]

[j]

1 ∑j ∈ G − {i}; d [i, j] ≤ r (d [i, j] ∙ W [j])

Notes: i: Pedestrian-counted spot (origins). j: Pedestrian environments (destinations). G: Network. r: Network radius (500 m, 1 km, 2 km). d[i, j]: Shortest-path distance between spot i and destination node j (meter). δ[i, j]: Straight-line distance between spot i and destination node j (meter). njk[i]: Number of paths passing through spot i with j and k in the network radius r from i. njk: Number of paths between nodes j and k. Beta(β): 0.00217. W(j): Features in pedestrian environments j (housing density, land use mix, transit ridership, built areas of main destination, park areas, and pedestrian accidents). Source: Modified from Sevtsuk and Mekonnen (2012) and Kang (2015).

2.3.2. Street environments The SMG pedestrian survey reported the street environments and number of walkers for pedestrian-counted spots. The environments include sidewalk width, number of street lanes, existence or nonexistence of street furniture, nearby crosswalks, and street slope. Street furniture refers to streetlights, phone booths, trees, and road signs. Furthermore, the report identified whether each street was for use by walkers only, mixed use for walkers and cars, or mixed use for walkers and bicycles. While continuous variables, such as width and number of street lanes, were converted into natural logarithms, we coded the presence of street features as dummy variables.

pedestrian environments constitute the weight when calculating metrics for spatial accessibility and centrality. As shown in Table 1, the following variables were measured for each pedestrian environment: (1) Density: residential density (housing units per ha); (2) Diversity: entropy index of all land use (housing, retail, office, manufacturing and warehouse, and other) and non-residential land use excluding residential use; (3) Design: number of pedestrian accidents as safety and park areas amenities; (4) Destination accessibility: total built areas of main destinations (regional and neighborhood retail, banks, offices, schools and private institutes, hospitals, churches, and recreation facilities); and (5) Distance to transit: transit ridership of subways and bus stops. To measure destination accessibility and access to public transits, we referred to Korean GPS data on pedestrians' destinations from 2010, which indicated that homemakers in their 30s and 40s in Seoul frequently walked to metro stations and bus stops, restaurants, neighborhood retail stores, schools, parks, and banks in their daily lives (Choi, Seo, & Park, 2011). As mentioned, the UNA generates two accessibility metrics, Reach and Gravity Index, and three centrality metrics, Betweenness, Straightness, and Closeness, within the specific network radius. Table 2 shows this in mathematical form. Reach measures the total value of destination features. The Gravity Index indicates Reach combined with the distance beta coefficient, while considering distance friction. The coefficient is 0.00217 (Handy & Niemeier, 1997). The use of the beta coefficient in this study is supported by the following. First, we did not have an empirically tested local beta value, due to unavailable data or relevant studies. Second, this study is somewhat similar to that of Handy and Niemeier (1997), which measured the beta value of travel

2.3.3. Census tract features of land use balance and intersection density To identify the features of land use balance and intersection density of roads and streets, we measured the residential and nonresidential balance index and intersection densities within the Korean census tract units. Previous studies have highlighted these as necessary factors in explaining the spatial variation of pedestrians (Cervero & Duncan, 2003; Kang, 2015). To measure the residential and nonresidential balance index, we employed data from the Seoul Building Registry. For residential sectors (Res), the total built area of housing was included, while nonresidential sectors (Nonres) included retail, office, industrial, and other nonresidential sites. The index values range from 0 to 1, where higher values indicate an even proportion within the given census tract units. The mathematical formula for measuring the balance index (Balance) within the census tract i is as follows (Eq. (1)):

Balancei = 1‐

5

Res ‐Nonres i Res+ Nonres i

(1)

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

Table 3 Descriptive statistics 1. Variables

Description

Mean

Standard deviation

Min

Max

Pedestrian volume on Weekdays Pedestrian volume on Saturday Spatial access to pedestrian environments (Vector S) Reach metrics log(housing density) log(land use mix)

Average number of weekday walkers Average number of Saturday walkers

3061.1 2986.9

3749.0 3786.7

6 7

106,186.0 113,606.0

log(housing units per ha) log(entropy index of housing, retail, office, industrial, and other land use) log(entropy index of retail, office, industrial) log(built areas of main destinations) log(transit ridership of bus and metro) log(number of pedestrian accidents) log(park areas)

69,393.1 4.6

1,493,993.0 3.8

0 0

66,600,000.0 18.7

5.5 88,141.4 41,210.8 3.3 24,982.2

4.8 70,798.8 46,299.1 5.5 112,524.2

0 0 0 0 0

25.1 735,845.3 346,361.0 52.0 2,670,837.0

log(housing units per ha) log(entropy index of housing, retail, office, industrial, and other land use) log(entropy index of retail, office, industrial) log(built areas of main destinations) log(transit ridership of bus and metro) log(number of pedestrian accidents) log(park areas)

35,295.1 2.3

744,849.6 1.9

0 0

34,200,000.0 9.9

2.8 44,888.7 21,328.4 1.7 12,473.7

2.4 36,745.2 24,199.1 3.0 56,578.7

0 0 0 0 0

14.6 357,758.1 161,141.4 33.1 1,706,040.0

log(housing units per ha) log(entropy index of housing, retail, office, industrial, and other land use) log(entropy index of retail, office, industrial) log(built areas of main destinations) log(transit ridership of bus and metro) log(number of pedestrian accidents) log(park areas)

155,003.2 10.9

5,249,102.0 16.1

0 0

327,000,000.0 210.9

13.4 205,119.5 84,485.3 6.1 30,884.1

21.8 318,850.1 217,516.3 29.2 191,155.6

0 0 0 0 0

400.8 7,465,600.0 3,033,673.0 1150.0 8,319,160.0

log(housing units per ha) log(entropy index of housing, retail, office, industrial, and other land use) log(entropy index of retail, office, industrial) log(built areas of main destinations) log(transit ridership of bus and metro) log(number of pedestrian accidents) log(park areas)

57,328.6 3.5

1,326,750.0 3.0

0 0

61,800,000.0 31.3

4.2 67,973.6 33,293.7 2.7 19,878.3

3.7 55,312.3 40,504.0 4.5 92,131.1

0 0 0 0 0

24.0 616,826.9 1,518,993.0 46.6 2,371,569.0

log(housing units per ha) log(entropy index of housing, retail, office, industrial, and other land use) log(entropy index of retail, office, industrial) log(built areas of main destinations) log(transit ridership of bus and metro) log(number of pedestrian accidents) log(park areas)

0.0002 0.2

0.0056 3.9

0 0

0.4 199.9

0.1 0.000001 0.000001 0.000862 0.000002

1.0 0.000011 0.000004 0.002036 0.000023

0 0 0 0 0

87.9 0.0005 0.0002 0.0818 0.0012

log(nonresidential land use mix) log(destination accessibility) log(access to transit) log(access to pedestrian accidents) log(access to parks) Gravity Index metrics log(housing density) log(land use mix) log(nonresidential land use mix) log(destination accessibility) log(access to transit) log(access to pedestrian accidents) log(access to parks) Betweenness metrics log(housing density) log(land use mix) log(nonresidential land use mix) log(destination accessibility) log(access to transit) log(access to pedestrian accidents) log(access to parks) Straightness metrics log(housing density) log(land use mix) log(nonresidential land use mix) log(destination accessibility) log(access to transit) log(access to pedestrian accidents) log(access to parks) Closeness metrics log(housing density) log(land use mix) log(nonresidential land use mix) log(destination accessibility) log(access to transit) log(access to pedestrian accidents) log(access to parks)

Greenwald, & McMillan, 2008; Næss, 2005). Finally, earlier studies have verified that walking volume responds to the nearest land use (Kang, 2015). Thus, this study included access to each land use type in the empirical models.

We added the intersection density of roads and streets to the Korean census tract units, as utilized by previous studies (Boer, Zheng, Overton, Ridgeway, & Cohen, 2007; Peiravian et al., 2014). The density is equivalent to the total number of intersections of roads or streets divided by the net areas, excluding non-developable land area of each census tract unit i. The formula is defined as (Eq. (2)):

Intersection densityi =

Number of intersections i Net areas i

2.4. Multilevel regression models (2)

Multilevel regression models were employed to determine the effect of spatial access to pedestrian environments on walking behaviors, because the features of individual units (pedestrian-counted spots) are nested within neighborhoods (Korean census tracts). Specifically, while empirical models using data on neighborhood units face ecological fallacy, in failing to consider individual features, models employing data on individual behaviors lead to atomistic fallacy, which is the loss of neighborhood contexts (Alker, 1969; Robinson, 1950). Furthermore, empirical studies have verified the validity of multilevel regression models in isolating the effect of neighborhood and individual variables on individual behaviors (Shuttleworth & Gould, 2010). Previous studies have applied Intraclass Correlation (ICC) to measure the fitness of

2.3.4. Location attributes To capture the relative location of each pedestrian-counted spot, this study used three factors: control value of streets; distance to city center (the CBD) and sub-CBDs; and distance to nearest housing, retail area, office, and industrial properties. Among the diverse metrics of Space Syntax methods to measure street layout, we selected control, because it fits the models theoretically and empirically. Additionally, studies have confirmed that relative distances to the CBD and sub-CBD are key to explaining the spatial variation in household density, employment density, economic activities, and real estate values (Boarnet, 6

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

Table 4 Descriptive statistics 2. Variables Street conditions (Vector C) logsidewalk_w loglane_no Street_Furniture Side_walk1 Side_walk2 Crosswalk Slope Census tracts (Vector B) bal_resi_nonresi logroad_den logst_den Location attributes (Vector L) logcontrol logcbd_dis logscbd_dis logne_resi_dis logne_com_dis logne_off_dis logne_ind_dis

Description

Mean

Standard deviation

Min

Max

log(width of sidewalk) log(number of street lanes) Existence of street furniture (yes = 1, no = 0) Pedestrian/car mixed street (yes = 1, no = 0) Pedestrian/bicycle mixed street (yes = 1, no = 0) Existence of crosswalk nearby roads (yes = 1, no = 0) Existence of street slope (yes = 1, no = 0)

4.0 2.9 0.9 0.4 0.1 0.5

2.2 2.4 0.3 0.5 0.2 0.5

1 1 0 0 0 0

24.3 18 1 1 1 1

0.2

0.4

0

1

Balance between housing and non-housing use log(density of road intersections) log(density of street intersections)

0.5 2.4 0.002

0.3 4.4 0.001

0.00000022 0 0.00

1 46 0.088

log(control of space syntax) log(distance to CBD) log(distance to sub-CBD) log(distance to nearest residential building) log(distance to nearest commercial building) log(distance to nearest office building) log(distance to nearest industrial building)

1.0 8501.9 4331.0 47.0 38.3 170.6 214.1

0.003 3912.3 2450.6 65.3 46.1 174.2 174.3

1.0 21.7 7.0 0.2 0.1 1.2 1.5

1.1 17,647.3 11,935.9 807.4 807.4 2127.3 1609.9

neighborhood level and 74% by street level. Further, the ICC value of the Betweenness model suggests that 30% of the variation in walking volume is at the neighborhood level and 70% at the street level (Woltman, Feldstain, MacKay, & Rocchi, 2012). In addition, the AIC value of each model revealed that the Gravity Index models of all neighborhood scales are best fitted, because they have a lower AIC value than others. This study interpreted the coefficients of variables within a statistical significance of 5%.

multilevel regression of the data and indicate the proportion of variation in pedestrian volume, as explained by street and neighborhood level. Application of the models is justified when the ICC value exceeds 0.05. We set the mathematical equation with the given data units as follows:

Pij = γ00 + β1 Sijk + β2 Cijk + β3 Bijk + β4 Lijk + μ0j + εij

(3)

where Pij= number of pedestrians at spot i in neighborhood j; βk= estimated parameters of each explanatory variable; γ00= constants; Sijk= spatial access to pedestrian environments at location i in tract j; Cijk= physical attributes of streets in location i in tract j; Bijk= land use balance and intersection density of neighborhood attributes in tract j; Lijk= location attributes in location i in tract j; μ0j, εij= the error terms of tract j and location i (Kang, 2015). We added the vectors S, C, B, and L in Eq. (3) to Tables 3 and 4 to link each vector with an individual variable. As there is no consensus on selecting functional forms after testing, this study selected the log-log functional form, which converts dependent and explanatory variables, excepting dummy variables, into natural logarithms. This form fits our empirical models theoretically and empirically, and interprets the parameters as mean elasticities. Finally, Akaike's information criterion (AIC) values for each model indicate their relative fitness. A smaller value indicates a better data fit. Tables 3 and 4 provide the descriptive statistics of all variables. Specifically, the variable “spatial access to pedestrian environments” lists the values for accessibility and centrality, weighted by the features of pedestrian environments. For instance, the log (housing density) below Reach metrics indicates the total value of residential density within the given network radius measured by Reach metrics.

3.1. Spatial access to pedestrian environments Figs. 4 to 6 summarize the marginal effects of spatial access to pedestrian environments on local variations in walking volume. After interpreting the overall effects of the 5Ds on walking, we compared the spatial accessibility and centrality to pedestrian environments over the three neighborhood scales ranging from 500 m to 2 km. For pedestrian environments, walking volume strongly responded to spatial access to the total built areas of main destinations and transit ridership at bus stops and metro stations within walkable neighborhoods. This confirms that walkers tend to be located near their work and daily destinations and are part of a transit-oriented community with higher transit demand, consistent with the GPS data on walking in Seoul and relevant studies (Choi et al., 2011). Additionally, the models comparing the effects of access to transit to metro stations and bus stops confirmed that bus stops with more transit demand correlate with more walking than metro stations with more ridership. Pedestrians enjoy access to widely-dispersed bus stops that connect with metro stations to reach their destinations, as confirmed in previous studies (Boarnet et al., 2008; Targa & Clifton, 2005). Notably, walking volume was highly associated with pedestrian accidents. Generally, walkers prefer safe places. However, the results revealed that a higher concentration of walkers increases the probability of pedestrian accidents. Thus, urban policy and design for traffic safety is required in areas with many walkers. Contrary to existing studies, this study confirmed that the impacts of land use diversity and residential density on walking volume were inconsistent (Frank et al., 2007; Rajamani et al., 2003). While nonresidential land use mix is positively linked with walking, mixed all land use generated no effects up to 1 km, and had adverse effects at 2 km. Residential density and access to parks were negatively associated with walking, except gravity access to density within the 2-km radius, which had weak positive effects, as verified in prior studies (Kitamura et al., 1997).

3. Results The multilevel regression models shown in Tables 5 to 7 apply to 9848 pedestrian-counted spots (Level 1) within 4264 census tract units (Level 2). To save space, this study provides the model results for the 500-m network radius. We separate the Reach, Gravity Index, Betweenness, Straightness, and Closeness models to avoid collinearity between metrics. Regarding the ICC value > 0.05 in the results, the models fit to identify the association of spatial access to pedestrian environments and walking volume. Furthermore, the ICC values of the Reach, Gravity Index, Straightness, and Closeness models indicate that 26% of the variation in pedestrian volume is explained by 7

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

Table 5 Models for predicting walking: Reach and Gravity index models (r = 500 m). Variables

Spatial access to pedestrian environments log(housing density) log(land use mix) log(nonresidential land use mix) log(destination accessibility) log(access to transit) log(access to pedestrian accidents) log(access to parks) Street conditions logsidewalk_w loglane_no Street_Furniture Side_walk1 Side_walk2 Crosswalk Slope Census tracts bal_resi_nonresi logroad_den logst_den Location and transportation logcontrol logcbd_dis logscbd_dis logne_resi_dis logne_com_dis logne_off_dis logne_ind_dis Constant ICC Number of spots Number of census tracts AIC

Reach models

Table 6 Models for predicting walking: Betweenness and Straightness models (r = 500 m).

Gravity index models

Variables

Betweenness models

Straightness models

Model 1 (Weekday)

Model 2 (Saturday)

Model 1 (Weekday)

Model 2 (Saturday)

Model 1 (Weekday)

Model 2 (Saturday)

Model 1 (Weekday)

Model 2 (Saturday)

Coefficients

Coefficients

Coefficients

Coefficients

Coefficients

Coefficients

Coefficients

Coefficients

−0.0112⁎

−0.0129⁎

−0.00313

−0.00587

0.00544 0.0166⁎⁎

0.00553 0.0174⁎⁎

−0.00581 0.0183⁎

−0.00553 0.0194⁎

0.0362⁎⁎⁎

0.0365⁎⁎⁎

0.115⁎⁎⁎

0.112⁎⁎⁎

0.0227⁎⁎⁎

0.0224⁎⁎⁎

0.0965⁎⁎⁎

0.0954⁎⁎⁎

0.0934⁎⁎⁎

0.0967⁎⁎⁎

0.0977⁎⁎⁎

0.106⁎⁎⁎

−0.00609⁎⁎

−0.00597⁎⁎

−0.0041

−0.00366

0.282⁎⁎⁎ 0.109⁎⁎⁎ 0.0982⁎⁎⁎ −0.219⁎⁎⁎ −0.108⁎⁎ 0.147⁎⁎⁎ −0.112⁎⁎⁎

0.283⁎⁎⁎ 0.116⁎⁎⁎ 0.101⁎⁎⁎ −0.214⁎⁎⁎ −0.106⁎⁎ 0.147⁎⁎⁎ −0.120⁎⁎⁎

0.274⁎⁎⁎ 0.111⁎⁎⁎ 0.102⁎⁎⁎ −0.254⁎⁎⁎ −0.104⁎⁎ 0.151⁎⁎⁎ −0.0932⁎⁎⁎

0.275⁎⁎⁎ 0.118⁎⁎⁎ 0.104⁎⁎⁎ −0.250⁎⁎⁎ −0.101⁎⁎ 0.150⁎⁎⁎ −0.101⁎⁎⁎

−0.0738⁎ 0.00237 −0.0239⁎⁎

−0.0621 0.00265 −0.0242⁎⁎

−0.0835⁎ 0.00117 −0.0272⁎⁎⁎

−0.0716⁎ 0.00147 −0.0274⁎⁎⁎

9.269⁎⁎ −0.103⁎⁎⁎ −0.0416⁎ 0.121⁎⁎⁎ −0.200⁎⁎⁎ −0.135⁎⁎⁎ 0.0584⁎⁎⁎ 8.310⁎⁎⁎ 0.30 9848 4264

8.839⁎ −0.102⁎⁎⁎ −0.0351⁎ 0.121⁎⁎⁎ −0.201⁎⁎⁎ −0.124⁎⁎⁎ 0.0579⁎⁎⁎ 8.168⁎⁎⁎ 0.30

6.957⁎ −0.123⁎⁎⁎ −0.0149 0.151⁎⁎⁎ −0.175⁎⁎⁎ −0.0697⁎⁎⁎ 0.0697⁎⁎⁎ 6.090⁎⁎⁎ 0.26

6.409 −0.121⁎⁎⁎ −0.00914 0.150⁎⁎⁎ −0.176⁎⁎⁎ −0.0593⁎⁎⁎ 0.0693⁎⁎⁎ 5.992⁎⁎⁎ 0.26

22,983.73

22,904.13

22,704.35

22,632.47

−0.00533

−0.00795

−0.00585

−0.00852

−0.00504 0.0178⁎

−0.00473 0.0191⁎

−0.00788 0.0177⁎

−0.00771 0.0189⁎

0.113⁎⁎⁎

0.110⁎⁎⁎

0.128⁎⁎⁎

0.125⁎⁎⁎

0.0929⁎⁎⁎

0.0918⁎⁎⁎

0.110⁎⁎⁎

0.109⁎⁎⁎

0.0894⁎⁎⁎

0.0974⁎⁎⁎

0.101⁎⁎⁎

0.112⁎⁎⁎

−0.00417

−0.00372

−0.00549

Spatial access to pedestrian environments log(housing density) log(land use mix) log (nonresidential land use mix) log(destination accessibility) log(access to transit) log(access to pedestrian accidents) log(access to parks) Street conditions logsidewalk_w loglane_no Street_Furniture Side_walk1 Side_walk2 Crosswalk Slope Census tracts bal_resi_nonresi logroad_den logst_den Location and transportation logcontrol logcbd_dis logscbd_dis logne_resi_dis logne_com_dis logne_off_dis logne_ind_dis Constant ICC Number of spots Number of census tracts AIC

−0.005

0.276⁎⁎⁎ 0.119⁎⁎⁎ 0.101⁎⁎⁎ −0.254⁎⁎⁎ −0.104⁎⁎ 0.153⁎⁎⁎ −0.0930⁎⁎⁎

0.278⁎⁎⁎ 0.127⁎⁎⁎ 0.103⁎⁎⁎ −0.249⁎⁎⁎ −0.102⁎⁎ 0.152⁎⁎⁎ −0.101⁎⁎⁎

0.272⁎⁎⁎ 0.112⁎⁎⁎ 0.100⁎⁎⁎ −0.258⁎⁎⁎ −0.101⁎⁎ 0.149⁎⁎⁎ −0.0920⁎⁎⁎

0.273⁎⁎⁎ 0.120⁎⁎⁎ 0.102⁎⁎⁎ −0.254⁎⁎⁎ −0.0986⁎⁎ 0.149⁎⁎⁎ −0.100⁎⁎⁎

−0.0780⁎ 0.00123 −0.0270⁎⁎⁎

−0.066 0.00153 −0.0272⁎⁎⁎

−0.0815⁎ 0.00165 −0.0280⁎⁎⁎

−0.0694⁎ 0.00198 −0.0282⁎⁎⁎

7.222⁎ −0.122⁎⁎⁎ −0.0145 0.154⁎⁎⁎ −0.177⁎⁎⁎ −0.0691⁎⁎⁎ 0.0699⁎⁎⁎ 6.070⁎⁎⁎ 0.26 9848 4264

6.682 −0.120⁎⁎⁎ −0.00871 0.153⁎⁎⁎ −0.178⁎⁎⁎ −0.0586⁎⁎⁎ 0.0695⁎⁎⁎ 5.974⁎⁎⁎ 0.26

7.269⁎ −0.122⁎⁎⁎ −0.0148 0.152⁎⁎⁎ −0.169⁎⁎⁎ −0.0566⁎⁎⁎ 0.0678⁎⁎⁎ 5.892⁎⁎⁎ 0.26

6.696⁎ −0.120⁎⁎⁎ −0.00923 0.152⁎⁎⁎ −0.171⁎⁎⁎ −0.0461⁎⁎⁎ 0.0674⁎⁎⁎ 5.795⁎⁎⁎ 0.26

22,738.34

22,666.75

22,597.37

22,523.82

⁎

p < 0.05. ⁎⁎ p < 0.01. ⁎⁎⁎ p < 0.001.

⁎ ⁎⁎

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

⁎⁎⁎

The results indicated the combined effects of street configuration and pedestrian environments by measuring the metrics for spatial accessibility and centrality. The association of spatial access to pedestrian environments with walking volume varied over the three neighborhood scales. The marginal effects of destination accessibility and access to transit increased as the network radii expanded from 500 m to 2 km. Within walkable neighborhoods, higher access to highly dense working and daily destinations as well as transit nodes with a higher transit demand were consistently favorable urban settings for walkers. Of the metrics on destination accessibility and distance to transit, the Gravity Index was most strongly associated with walking, followed by Reach, Straightness, Closeness, and Betweenness. Walkers prefer places with higher access and directly routed streets to reach bus stops and metro stations up to the 1 km radius. Within 2 km, only the Gravity Index and Straightness to transit significantly affected walking volume. Spatial access to public transit also varied with neighborhood scales. Thus, urban design to create transit-oriented neighborhoods should consider a

locally fitted street layout to enhance the use of transit. Furthermore, higher spatial accessibility and centrality to pedestrian accidents was associated with more walkers for all neighborhood scales. However, the marginal effects were strongest for the 500-m radius and weak at 2 km. This implies that pedestrian accidents occurred locally in neighborhoods with a higher volume of walkers, consistent with previous studies (Rothman et al., 2014). Interestingly, while higher nonresidential land use mix increased walking volume from 500 m to 2 km, the land use mix of all uses had no statistical significance up to 1 km and a negative association at 2 km. This suggests that co-location of nonresidential use, such as commercial and office space, increases walking volume, while mixing residential land use is associated with less walking in urban areas, consistent with previous research (Koster & Rouwendal, 2012). The variation between the spatial accessibility and centrality metrics was similar, implying that spatial access to land use mix responds to the degree of land use mix, rather than street layout. For access to 8

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

500 m and 1 km radii, but had no effects at 2 km. Briefly, more people walk within neighborhoods with more built areas of destinations, easier access to transit with higher ridership, more pedestrian accidents, higher mix of nonresidential uses along a highly dense street network. Furthermore, pedestrian presence was positively associated with a direct route to daily destinations, public transits with more ridership, and spots with higher risk of pedestrian accidents.

Table 7 Models for predicting walking: Closeness models (r = 500 m). Variables

Spatial access to pedestrian environments log(housing density) log(land use mix) log(nonresidential land use mix) log(destination accessibility) log(access to transit) log(access to pedestrian accidents) log(access to parks) Street conditions logsidewalk_w loglane_no Street_Furniture Side_walk1 Side_walk2 Crosswalk Slope Census tracts bal_resi_nonresi logroad_den logst_den Location attributes logcontrol logcbd_dis logscbd_dis logne_resi_dis logne_com_dis logne_off_dis logne_ind_dis Constant ICC Number of spots Number of census tracts AIC

Closeness models Model 1(Weekday)

Model 2(Saturday)

Coefficients

Coefficients

0.00629 0.00472 −0.014 −0.0669⁎⁎⁎ −0.0489⁎⁎⁎ −0.0230⁎⁎⁎ 0.00144

0.00927 0.00492 −0.0150⁎ −0.0657⁎⁎⁎ −0.0486⁎⁎⁎ −0.0237⁎⁎⁎ 0.00116

0.285⁎⁎⁎ 0.124⁎⁎⁎ 0.105⁎⁎⁎ −0.247⁎⁎⁎ −0.113⁎⁎ 0.158⁎⁎⁎ −0.0993⁎⁎⁎

0.287⁎⁎⁎ 0.132⁎⁎⁎ 0.108⁎⁎⁎ −0.242⁎⁎⁎ −0.111⁎⁎ 0.158⁎⁎⁎ −0.107⁎⁎⁎

−0.0698⁎ 0.00138 −0.0255⁎⁎⁎

−0.0574 0.00168 −0.0257⁎⁎⁎

8.357⁎ −0.122⁎⁎⁎ −0.0159 0.149⁎⁎⁎ −0.195⁎⁎⁎ −0.105⁎⁎⁎ 0.0690⁎⁎⁎ 6.563⁎⁎⁎ 0.28 9848 4264 23,058.20

7.894⁎ −0.120⁎⁎⁎ −0.00932 0.148⁎⁎⁎ −0.196⁎⁎⁎ −0.0945⁎⁎⁎ 0.0683⁎⁎⁎ 6.464⁎⁎⁎ 0.28

3.2. Physical street conditions Tables 5 to 7 show that the physical features of a street affect walking alternatives. This study found that walkers prefer wider streets with more lanes, street furniture, and crosswalks of nearby roads, consistent with previous studies (Cerin, Macfarlane, Ko, & Chan, 2007; Kang, 2015). However, there are fewer walkers in streets mixed with cars and bicycles and those that are more sloped, similar to findings by Giles-Corti et al. (2005).

3.3. Census tract features of land use balance and intersection density The empirical models also tested how the attributes of land use balance and road/street affect walking volume, with reference to the extant literature. The balance of residential and nonresidential land use negatively correlated with pedestrian volume, because walkers tend to be near commercial and office land use areas. Regarding the intersection density of roads and streets, a higher street density decreases walking volume, while road density had no statistical significance. We interpret these patterns as indicating that roads for automobiles are not a favorable setting for walkers, while only dense streets do not promote pedestrian presence (Kang, 2015).

3.4. Location attributes

22,991.11

⁎

The Space Syntax approach was employed to measure street layout. Among the metrics, higher control over the street network was associated with more pedestrians, in alignment with previous studies (Baran et al., 2008). While access to the CBD affects walking volume, access to sub-CBDs inconsistently changed the volume, due to variation in local contexts surrounding sub-CBDs. Regarding access to different land use areas, walking volume increased closer to the nearest commercial and office properties. However, the models revealed the opposite results, indicating that walking decreased around residential and industrial properties.

p < 0.05. p < 0.01. ⁎⁎⁎ p < 0.001. ⁎⁎

residential density and parks, only Betweenness and the Gravity Index for residential density and Betweenness for parks were statistically significant. Neighborhoods with higher residential density with more detour volume were linked with less walking across neighborhood scales, while higher access to density generated more walking volume at the 2-km radius. Finally, access to parks was weakly negative in the

Fig. 4. Marginal effects of spatial access to pedestrian environments on pedestrian volume (500 m).

9

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

Fig. 5. Marginal effects of spatial access to pedestrian environments on pedestrian volume (1 km).

Fig. 6. Marginal effects of spatial access to pedestrian environments on pedestrian volume (2 km).

4. Discussion

investment and urban design must prioritize easier access to commuting and daily travel destinations and a transit-oriented community. Second, higher access to bus-stops with more transit demand was more associated with pedestrian volume than metro stations. Thus, urban planning should focus on the street layout connecting each land use to nearby bus-stops, as well as a transit-walking connected package that encourages walking and transit use. Third, pedestrian volume was highly associated with a higher probability of pedestrian accidents. Generally, walkers prefer safe places. However, this result verified that pedestrian accidents were concentrated in spots with more walking volume due to the specific condition of built environments in which walkers are greatly exposed to traffic (Rothman et al., 2014). Thus, urban planning and urban design should encourage traffic calming (roundabouts and speed humps) and create public spaces and recreation areas near spots with higher pedestrian presence. Fourth, density and diversity of overall land use are not decisive factors in attracting more walkers. These pedestrian environments are necessary conditions, but not sufficient. Thus, density and diversity within walkable neighborhoods require attractive place-making to harmonize with other pedestrian environments along well-organized street settings. Fifth, comprehensive and flexible land use strategies should be developed, focusing on the walkable district units in which street and land use are closely integrated. This planning style requires the creation of walkable communities based on local contexts and the spatial scope of effects.

Consistent with the core hypotheses, walking volume was positively associated with spatial access to the total built areas of main destinations and bus stops and metro stations with higher ridership. Furthermore, higher pedestrian presence was highly associated with more pedestrian accidents. Conversely, diversity of all land use, residential density, and access to parks were negative or statistically insignificant factors. Regarding spatial accessibility and centrality metrics, more people walk within neighborhoods with more built areas of destinations, easier access to transit with higher ridership, and more pedestrian accidents along a highly dense street network. Furthermore, pedestrian presence was positively associated with a direct route to destinations, public transits with more ridership, and spots with higher risk of pedestrian accidents. Finally, the association of spatial access to pedestrian environments with walking volume varied over the three neighborhood scales. The marginal effects of destination accessibility and access to transit increased as the network radii expanded from 500 m to 2 km. In contrast, the effects of pedestrian accidents weakened over the neighborhood scales. The key findings suggest insightful implications for academic and professional communities. First, destination accessibility and access to public transits are important in local variations in walking volume. Thus, to create pedestrian-friendly cities and neighborhoods, public

10

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

Indeed, relevant projects should tailor pedestrian environments to walking behavior, land use patterns, spatial distribution of residents and workers, and micro-level street layout. Sixth, periodic monitoring of the relations between street configuration, pedestrian environments, and walking volume as well as collaboration among relevant authorities provides critical information and an efficient process for creating more walkable neighborhoods and cities. Specifically, quantitative and qualitative analyses of street-based pedestrian environments provide a cornerstone for a customized street and land use design, suitable for local contexts. Finally, if the framework of this study is robustly tested and approved, it could be applied to understanding the relationship between street configuration and relevant urban issues, such as land use change, property price variation, the spatial pattern of households and firms, and traffic flow. Furthermore, future studies could evaluate the effects of pedestrian-friendly policy and projects for creating sustainable and livable cities using the approach of this study. This research framework could also serve as a reference to identify the similarities and differences between local and global cases.

from the San Francisco Bay Area. American Journal of Public Health, 93(9), 1478–1483. Cervero, R., & Kockelman, K. (1997). Travel demand and the 3Ds: Density, diversity, and design. Transportation Research Part D: Transport and Environment, 2(3), 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 Institution of Civil Engineers-Urban Design and Planning, 166(3), 176–199. Choi, Y.-M., Seo, H.-L., & Park, S.-H. (2011). Walking destinations and boundaries of everyday lives in residential areas. Journal of the Architectural Institute of Korea Planning & Design, 27(8), 91–102. 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. Dong, X., Ben-Akiva, M., Bowman, J., & Walker, J. (2006). Moving from trip-based to activity based measures of accessibility. Transportation Research Part A, 40(2), 163–180. El-Geneidy, A. M., & Levinson, D. (2011). Place rank: Valuing spatial interactions. Networks and Spatial Economics. 11, 643–659. Ewing, R., & Cervero, R. (2010). Travel and the built environment: A meta-analysis. Journal of the American Planning Association, 76(3), 265–294. Frank, L. D., Kavage, S., Greenwald, M., Chapman, J., & Bradley, M. (2009). I-PLACE3S health & climate enhancements and their application in King County. Seattle, WA: King County Health Scape. Frank, L. D., Kerr, J., Chapman, J., & Sallis, J. (2007). Urban form relationships with walk trip frequency and distance among youth. American Journal of Health Promotion, 21(4), 305–311. Franklin, J. P., & Waddell, P. (2003). A hedonic regression of home prices in King County, Washington, using activity-specific accessibility measures. Paper presented at the Proceedings of the Transportation Research Board 82nd Annual Meeting, Washington, DC. Geurs, K. T., & Van Wee, B. (2004). Accessibility evaluation of land-use and transport strategies: Review and research directions. Journal of Transport Geography, 12(2), 127–140. Giles-Corti, B., Broomhall, M., Knuiman, M., Collins, C., Douglas, K., Ng, K., ... Donovan, R. (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. L., Cao, X., & Mokhtarian, P. L. (2006). Self-selection in the relationship between the built environment and walking—Empirical evidence from Northern California. Journal of the American Planning Association, 72(1), 55–74. Handy, S. L., & Niemeier, D. A. (1997). Measuring accessibility: An exploration of issues and alternatives. Environment and Planning A, 29(7), 1175–1194. Hansen, W. G. (1959). How accessibility shapes land use. Journal of the American Institute of Planners, 25(2), 73–76. Hillier, B. (1996). Space is the machine. Space syntax. Jacobs, J. (1961). The death and life of great American cities: Vintage. Kang, C. (2015). The effects of spatial accessibility and centrality to land use on walking in Seoul, Korea. Cities, 46, 94–103. Kang, C. (2016). Spatial access to pedestrians and retail sales in Seoul, Korea. Habitat International, 57, 110–120. Kang, C. (2017). Measuring the effects of street network configurations on walking in Seoul, Korea. Cities, 71, 30–40. Kang, C. D., & Cervero, R. (2009). From elevated freeway to urban greenway: Land value impacts of the CGC project in Seoul, Korea. Urban Studies, 46(13), 2771–2794. Kitamura, R., Mokhtarian, P. L., & Laidet, L. (1997). A micro-analysis of land use and travel in five neighborhoods in the San Francisco Bay Area. Transportation, 24(2), 125–158. Koh, P. P., & Wong, Y. D. (2013). Comparing pedestrians' needs and behaviours in different land use environments. Journal of Transport Geography, 26, 43–50. Koster, H. R., & Rouwendal, J. (2012). The impact of mixed land use on residential property values. Journal of Regional Science, 52(5), 733–761. 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(3), 204–215. Lee, S. (2016). Walking City, Seoul Seoul Seoul Institute. Loo, B. P., & Chow, S. (2006). Sustainable urban transportation: Concepts, policies, and methodologies. Journal of Urban Planning and Development, 132(2), 76–79. Manaugh, K., & Kreider, T. (2013). What is mixed use? Presenting an interaction method for measuring land use mix. Journal of Transport and Land Use, 6(1), 63–72. Miller, H. J. (2005). Place-based versus people-based accessibility. In D. M. Levinson, & K. J. Krizek (Eds.). Access to destinations (pp. 63–89). Oxford: Elsevier. Moudon, A. V., & Lee, C. (2003). Walking and bicycling: An evaluation of environmental audit instruments. American Journal of Health Promotion, 18(1), 21–37. Mueller, B. A., Rivara, F. P., Lii, S. M., & Weiss, N. S. (1990). Environmental factors and the risk for childhood pedestrian-motor vehicle collision occurrence. American Journal of Epidemiology, 132(3), 550–560. Næss, P. (2005). Residential location affects travel behavior—But how and why? The case of Copenhagen metropolitan area. Progress in Planning, 63(2), 167–257. Osland, L., & Thorsen, I. (2008). Effects on housing prices of urban attraction and labormarket accessibility. Environment and Planning A, 40(10), 2490–2509. Ozbil, A., Peponis, J., & Stone, B. (2011). Understanding the link between street connectivity, land use and pedestrian flows. Urban Design International, 16(2), 125–141. Park, S., Choi, Y., Seo, H., Moudon, A. V., Bae, C.-H. C., & Baek, S.-R. (2016). Physical

5. Conclusion This study has identified the association between spatial access to pedestrian environments along the street network and walking volume in Seoul, Korea. We measured spatial accessibility and centrality metrics, including both street configuration and features of destinations, such as density, diversity, amenities measured through pedestrian accidents and park areas, the total built areas of main neighborhood destinations, and public transit with ridership. The metrics expanded perspectives on the link between walkable neighborhoods represented by the 5Ds and pedestrian behaviors. Thus, we suggest the “S + 5Ds” model as a more reliable measure of pedestrian environments. The key findings highlight the significance of spatial access to pedestrian environments on walking volume in urban planning and design. Thus, policies and projects to create pedestrian-friendly neighborhoods should consider the multidimensional relationship between spatial access to pedestrian environments and walking behaviors. The empirical models in this study require further research. First, we require greater understanding of how temporal changes in spatial access to pedestrian environments affect walking volume. If panel data were available, this could be tested. Second, further studies should compare the effects in traditional urban structures and suburban areas. References 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(5), 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. Alker, H. R. (1969). A typology of ecological fallacies. Quantitative ecological analysis in the social sciences (pp. 69–86). Cambridge, MA: MIT Press Cambridge, MA ed. Baran, P. K., Rodríguez, D. A., & Khattak, A. J. (2008). Space syntax and walking in a new urbanist and suburban neighbourhoods. Journal of Urban Design, 13(1), 5–28. Bento, A. M., Cropper, M., Mobarak, A. M., & Vinha, K. (2003). The impact of urban spatial structure on travel demand in the United States. Policy Research Working Paper. Washington: World Bank. Boarnet, M. G., Greenwald, M., & McMillan, T. E. (2008). Walking, urban design, and health: Toward a cost-benefit analysis framework. Journal of Planning Education and Research, 27(3), 341–358. Boer, R., Zheng, Y., Overton, A., Ridgeway, G. K., & Cohen, D. A. (2007). Neighborhood design and walking trips in ten US metropolitan areas. American Journal of Preventive Medicine, 32(4), 298–304. Caschili, S., & De Montis, A. (2013). Accessibility and complex network analysis of the US commuting system. Cities, 30, 4–17. 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. (2005). Accessible cities and regions: A framework for sustainable transport and urbanism in the 21st century. UC Berkeley Center for Future Urban Transport: A Volvo Center of Excellence. Cervero, R. (2006). Alternative approaches to modeling the travel-demand impacts of smart growth. Journal of the American Planning Association, 72(3), 285–295. Cervero, R., & Duncan, M. (2003). Walking, bicycling, and urban landscapes: Evidence

11

Cities xxx (xxxx) xxx–xxx

C.-D. Kang

businesses in dense urban environments. Journal of Planning Education and Research, 34(4), 374–393. Sevtsuk, A., & Mekonnen, M. (2012). Urban network analysis. Revue internationale de géomatique–n, 2, 287–305. Shuttleworth, I., & Gould, M. (2010). Distance between home and work: A multilevel analysis of individual workers, neighbourhoods, and employment sites in Northern Ireland. Environment and Planning A, 42(5), 1221–1238. 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(4), 279–288. Talen, E., & Koschinsky, J. (2013). The walkable neighborhood: A literature review. International Journal of Sustainable Land Use and Urban Planning, 1(1), 42–63. Targa, F., & Clifton, K. (2005). The built environment and trip generation for non-motorized travel. Journal of Transportation and Statistics, 8(3), 55–70. 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 & Design, 34(3), 539–555. Vale, D. S., Saraiva, M., & Pereira, M. (2015). Active accessibility: A review of operational measures of walking and cycling accessibility. Journal of Transport and Land Use, 9(1). 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. Woltman, H., Feldstain, A., MacKay, J. C., & Rocchi, M. (2012). An introduction to hierarchical linear modeling. Tutorial in Quantitative Methods for Psychology, 8(1), 52–69. 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(1), 112–131. Zhang, M. (2004). The role of land use in travel mode choice: Evidence from Boston and Hong Kong. Journal of the American Planning Association, 70(3), 344–361.

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., Hadjinikolaou, E., Livieratos, C., & Fatouros, D. (1989). The spatial core of urban culture. Ekistics, 43–55. Peponis, J., & Wineman, J. (2002). Spatial structure of environment and behavior. Handbook of Environmental Psychology (pp. 271–291). . Porta, S., Crucitti, P., & Latora, V. (2006). The network analysis of urban streets: A primal approach. Environment and Planning. B, Planning & Design, 33(5), 705–725. Rajamani, J., Bhat, C., Handy, S., Knaap, G., & Song, Y. (2003). Assessing impact of urban form measures on nonwork trip mode choice after controlling for demographic and level-of-service effects. Transportation Research Record: Journal of the Transportation Research Board, 1831, 158–165. Reggiani, A., Bucci, P., & Russo, G. (2011). Accessibility and impedance forms: Empirical applications to the German commuting networks. Robinson, W. S. (1950). Ecological correlations and the behavior of individuals. American Sociological Review, 15(3), 351–357. Rothman, L., Buliung, R., Macarthur, C., To, T, & Howard, A. (2014). Walking and child pedestrian injury: A systematic review of built environment correlates of safe walking. Injury Prevention, 20(1), 41–49. 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. Seo, H.-L. (2006). The characteristics of walking environments in Buckchon Residential Districts, Seoul (Master)Seoul, Korea: Seoul National University. Seoul Institute (2017). Seoul research data services. Seoul Metropolitan Government (2010). Seoul pedestrian volume report. Sevtsuk, A. (2014). Location and agglomeration: The distribution of retail and food

12