Spatially-varying effects of built environment determinants on walking

Highlights

• This study explores the spatially-varying relationship between walking and built environment.

• Geographically Weighted Regression is utilized for a spatially-sensitive analysis.

• This approach enhanced the understanding about the influential pattern of variables at various geographic scales.

• The coefficients spatially vary depending on the site-specific characteristics.

Introduction

Walking is the primary means of human locomotion, providing various benefits to society. By promoting public transportation uses, walking reduces vehicle operation and gasoline consumption (Litman, 2003, McCann, 2000), and thereby reduces pollution from motorized travel (Marshall et al., 2009, Whitelegg and Williams, 2000). In addition, larger number of pedestrians make communities more viable both economically and culturally (Forkenbrock et al., 2001, Local Government Commission, 2012). According to recent studies, walking increases frequent and brisk exercise, and relieves stress (Mammen and Faulkner, 2013, Marselle et al., 2013, Robertson et al., 2012). Walking also helps reduce obesity and mitigating various health problems (Boarnet et al., 2008, Jackson and Kochtitzky, 2001, Litman, 2003, Ogden et al., 2014).

For these reasons, many cities and municipalities are interested in promoting walking and have embraced design principles for pedestrian-friendly urban environments (Council US Green Building, 2009, Minister of Land, 2013, UK Department for Transport, 2007). Promoting walking has become an important part of urban design improvement (Rahman et al., 2013), and policymakers have started to look at improving walkability as a cost-beneficial solution to achieve sustainable transport (Park et al., 2017).

In this trend, researchers have made significant efforts to find determinants that promote walking. The most frequently referred determinants are the ‘3Ds’ proposed by Cervero and Kockelman (1997), and the ‘5Ds’ proposed by Cervero et al. (2009). The ‘3Ds’ are the three built environment attributes–Density, Diversity, and Design; the ‘5Ds’ are an extended version of the ‘3Ds’ proposed by adding two accessibility attributes, ‘Distance to transit’ and ‘Destination Accessibility.’ In an effort to find out built environment determinant for walking, they are the most frequently tested variables for previous studies. One difficulty in relation to the measurement is that built environment attributes can be measured at various geographical scales. For example, measurements have been made at the location level, at the block level, and based on catchment areas.

Many studies have revealed land use patterns as a key determinant for choosing walking over motorized modes of transport. Several studies have found that exclusive land use, also known as Euclidean zoning increases pedestrian volume, in case of commercial use (Miranda-Moreno et al., 2010, Sung et al., 2015) or residential use (Christiansen et al., 2016). But, most studies have emphasized that a higher mix of land uses or a balance of residential and retail uses generates more walking activity (Cervero, 1996, Frank and Pivo, 1994, Christiansen et al., 2016, Sung et al., 2013, Kang, 2015, Kang, 2018, Kim et al., 2014). In contrast, Kang (2017) derived a negative relation between land use mix and walking.

Transit accessibility has also been considered an important component for promoting walking. To assess the level of transit accessibility, some studies have measured the distance to the nearest transit station (Christiansen et al., 2016, Kang, 2015, Kim et al., 2014, Sung et al., 2013, Sung et al., 2015, Vale and Pereira, 2016); others have counted the number of stations within a certain area (Kim et al., 2014, Larrañaga et al., 2016, Liu and Griswold, 2009, Miranda-Moreno et al., 2010, Sung et al., 2013, Kang, 2017).

In addition, street network topology and physical design attributes at the street-level have been highlighted as possible determinants of walking. The significance of well-connected networks has been tested using intersection density or road density (Cervero and Kockelman, 1997, Cervero et al., 2009, Ewing and Cervero, 2010, Jacobs, 1961, Kang, 2017, Kim et al., 2014, Larrañaga et al., 2016, Liu and Griswold, 2009, Sung et al., 2013, Song and Knaap, 2004). Some recent studies used the type and dimensions of neighborhood roads to uncover the effects of physical design on walking at the street-level (Cervero et al., 2009, Kim et al., 2014, Liu and Griswold, 2009, Park et al., 2015, Sung et al., 2013, Sung et al., 2015). Among socio-economic variables, higher population and employee density show significant positive associations with pedestrian volume (Kang, 2015, Larrañaga et al., 2016, Liu and Griswold, 2009, Miranda-Moreno et al., 2010). Dwelling (household) density has also been tested to see whether it has a relation with pedestrian volume (Liu and Griswold, 2009, Vale and Pereira, 2016).

To quantify the effects of the aforementioned factors on walking, early studies developed empirical models. Most studies have relied on typical regression methods, such as a linear regression model (Liu and Griswold, 2009, Miranda-Moreno et al., 2010, Ozbil et al., 2011, Sung et al., 2013, Sung et al., 2015) and multilevel regression model (Cervero et al., 2009, Kang, 2015, Kang, 2017, Kang, 2018). However, some previous studies have overlooked the geographical nature of walking, which results in ‘spatial autocorrelation’ and ‘spatial non-stationarity.’ ‘Spatial autocorrelation’ is a similarity of nearby observations and is a common phenomenon found in the data collected from large areas. Without proper specification, regression analysis may lead to underestimation of the standard error of coefficient estimates (Anselin, 1988). Meanwhile, ‘spatial non-stationarity’ refers to the fact that the relationship between independent and outcome variables varies across areas. Although both phenomena are frequently observed in walking data, most studies have used conventional regression models that only estimate a set of universal coefficients for all locations (Fotheringham et al., 2003). One piece of evidence indirectly supporting this argument is that values of coefficients estimated in previous studies were quite different for the same variables. In accurately reflecting different spatial conditions, conventional regression methods may be ineffective.

In this context, the present study is conducted with an objective of exploring the spatially-varying effects of urban built environments on walking. For a spatially-sensitive analysis, the geographically weighted regression (GWR) method is used to analyze pedestrian volume data measured at 10,000 locations in Seoul Metropolitan City, South Korea. This study intends to elucidate how urban built environments influence pedestrian volume and to see if the influence has a geographical pattern. The remainder of this paper is organized as follows. The first section describes the data and variables. The second section explains the analysis method used to discover the spatially-varying relation between built environments and walking. In the third section, the modeling outcomes and findings are interpreted. In the last section, the implications of the analysis results are discussed and suggestions are made for future policies.