The severity of pedestrian crashes: an analysis using Google Street View imagery

Highlights

• Google Street View is used to obtain pedestrian and road infrastructure features.

• Logit models of the severity of pedestrian casualties are estimated with this data.

• The probability of pedestrian crashes cannot be analyzed with this data.

• Sidewalks and buffers reduce severity, high speed roads increase severity.

• Crashes during darkness increase the severity of pedestrian casualties.

Abstract

Data derived from visual inspection of Google Street View imagery associated with a variety of pedestrian and road infrastructure features are analyzed with a database of pedestrian casualties. These features include the presence of sidewalks, buffers between the road and the sidewalk, street lighting, number of travel lanes and the presence of medians, traffic controls at intersections, and posted speed limits. The analysis focuses on how these features affect the severity of a pedestrian casualty once it has occurred. Other controls used in the analysis include the age of the victim, ambient lighting conditions, and whether the vehicle driver was intoxicated. Results suggest that severity of pedestrian casualties is associated with the lack of sidewalks and buffers, high-speed roads, roads with six or more lanes and a median, and lack of traffic lighting when it is dark. Speed is a critical factor in determining the severity of crash outcomes, and most road characteristics affect crash outcomes to the extent that they moderate or facilitate high speeds. Casualties are more severe when it is dark than when it is daylight. Older pedestrians tend to have more severe casualties. A key contribution of this work is the use of Google Street View imagery; however, a limitation is that the analysis cannot examine the probability of a pedestrian casualty. Implications for road safety are consistent with national efforts to make streets safer via Complete Streets policies.

Introduction

Pedestrian safety has typically been dealt with by controlling pedestrian movements, educating the pedestrian, and enforcing laws aimed at both driver and pedestrian behavior. The control of pedestrian movements has relied largely on traffic signals, designated pedestrian crossing points, and barriers to prevent pedestrians from crossing high-speed roads. Historically, the conflict between vehicle movement and pedestrian access has favored the vehicle (Norton, 2008). The provision of sidewalks, medians, crossing points, and barrier fences in London in the early 1900s was predicated on the desire to maintain traffic flow by keeping pedestrians out of the street (Ishaque and Noland, 2006).

A recent review of the built environment and traffic safety discusses these issues from the perspective of urban planning (Ewing and Dumbaugh, 2009). Urban sprawl tends to be associated with more vehicle crashes and can result in unsafe pedestrian environments. One issue is that highway designs meant for high-speed roads are often applied within urbanized areas that have pedestrian activity (Dumbaugh and Rae, 2009), leading to dangerous conditions for pedestrians and even reducing the level of pedestrian activity (Ewing and Cervero, 2010).

Much of the research that has examined various urban design factors on road safety has used spatial analysis techniques. A common technique is to use geographic information systems to construct various measures that represent the built environment and the road network. These measures include road-network density (i.e., miles of road per unit area), intersection density (nodes or junctions per unit area), land uses (e.g., commercial, industrial, and residential, amongst others), development intensity (population and employment density), and various spatial demographic indicators (income, population, car ownership, etc.). Some studies examine the association of these variables with vehicle crashes (Aguero-Valverde and Jovanis, 2006, Dumbaugh and Rae, 2009, Noland and Quddus, 2004, Noland and Quddus, 2005), while others explore pedestrian crashes (Graham and Glaister, 2003, Kim et al., 2006, LaScala et al., 2000, Loukaitou-Sideris et al., 2007, Miranda-Moreno et al., 2011, Ukkusuri et al., 2011, Wier et al., 2009) and child pedestrian crashes (Graham et al., 2005, Petch and Henson, 2000). Much of this research has also focused on the role that area-based deprivation can have on casualties (Graham et al., 2005, Noland and Quddus, 2004).

While this research has been instrumental in showing the associations between various spatial measures and both vehicle and pedestrian crashes, it misses details about how the specific design features of roads can influence the safety of the pedestrian environment. These design details include the presence or absence of features that can benefit pedestrians, including sidewalks, medians, street buffers (planting strips), and crosswalks. Some studies have examined details such as lane width and the number of lanes, and found that increases in both are associated with more pedestrian casualties (Ukkusuri et al., 2011). This result has also been found for total road casualties (Noland, 2003). Pedestrian-friendly design policies emphasize smaller roads, adequate sidewalks, pedestrian medians, and buffers between the road and the sidewalk. In theory, these would likely increase pedestrian safety, but not much is known about the specific effects of these design treatments. Many of these may moderate vehicle speeds and thus lead to less severe crash outcomes and fewer crashes.

Recent guidelines provide engineering criteria for planning and designing of pedestrian facilities consistent with these goals (AASHTO, 2004). Information is provided on when and how to place sidewalks along streets, including their width, continuity, medians and refuges, and recommended landscaping and tree coverage. Traffic calming as a way to improve safety is also discussed. Given that these guidelines are relatively new, most existing infrastructure has not been upgraded to match recommendations.¹

The research presented here examines these issues using detailed design data recorded primarily from Google Street View imagery for New Jersey (and other internet mapping resources, if needed) and using this in statistical models associating these features with the severity of pedestrian-involved crashes that have occurred. Other research has recently evaluated the reliability of using Google Street View imagery for characterizing the built environment by comparing the information collected with field audits. All have concluded that Google Street View images are a reliable source of this type of data and useful for research (Clarke et al., 2010, Rundle et al., 2011, Wilson et al., 2012). One study evaluated the ability to survey features of the built environment associated with pedestrian safety and found good concordance with field audit data (Rundle et al., 2011).

These data are used in statistical models of pedestrian crash severity. One restriction of this analysis is that we must model only those crashes that have occurred, as we do not have Google Street View imagery for every location in New Jersey where pedestrian crashes did not occur. Thus, our models estimate the association of various infrastructure features with the probability of pedestrian fatal versus injury-only outcomes. To our knowledge, this is the first analysis that examines the associations of pedestrian and road infrastructure features with crash severity at this level of detail.