URBAN STREETSCAPE DESIGN AND CRASH SEVERITY …docs.trb.org/prp/15-2942.pdf1 URBAN STREETSCAPE DESIGN AND CRASH SEVERITY ... 19 which provide important roof-life enclosure for many

Harvey and Aultman-Hall 1

URBAN STREETSCAPE DESIGN AND CRASH SEVERITY 1 2

Chester Harvey 3 Graduate Research Assistant 4 Transportation Research Center* 5 Phone: 802-377-2760 6 [email protected] 7 8 Lisa Aultman-Hall 9 Professor, School of Engineering 10 Transportation Research Center* 11 Phone: 802-656-1245 12 [email protected] 13 14 *Transportation Research Center 15 University of Vermont 16 210 Colchester Avenue 17 Burlington, VT 05405 18

19 20 21 22 23

24 A Paper Submitted August 1, 2014 for Presentation at the 2015 Annual Meeting of the 25 Transportation Research Board and Publication in the Transportation Research Record 26

27 Edited and Resubmitted November 13, 2014 28

29 Total words = 5,997 + 250*6 (4 Figures + 2 Tables) = 7,497 30


ABSTRACT 1 2 Streetscape design is increasingly acknowledged as a tool for improving traffic safety and 3 livability in urban settings. While traditional highway safety engineering promotes widening and 4 removing obstacles from roadside “clear zones” to reduce collision potential, a contrasting 5 framework proposes that smaller, more enclosed streetscapes may incentivize slower, less risky 6 driving and therefore improve both livability and safety outcomes. Such a strategy may have 7 particular relevance along urban arterials, where large clear zones may be impractical due to 8 complex adjacent land uses and promotion of livable spaces is an increasing focus. 9

This study examined streetscape design and traffic safety in urban settings by assessing 10 relationships between crash severity and streetscape design variables in New York City. A GIS-11 based method was used to precisely capture streetscape design measurements at the locations of 12 more than 240,000 crashes. Logistic regression models indicated that crashes on smaller, more 13 enclosed streetscapes were less likely to result in injury or death compared with those on larger, 14 more open streetscapes. These results point to infill development and street tree planting as 15 safety countermeasures that are consistent with additional livability goals such as walkability, 16 high quality public realm design, and provision of natural amenities. 17

18 Key Words: streetscape design, traffic safety, urban arterials, livable streets, GIS 19


INTRODUCTION 1 2 A growing body of research agrees that the built environment design of urban streetscapes has an 3 important and traditionally misunderstood effect on traffic safety. Highway engineering in rural 4 environments traditionally stresses the importance of clear zones on either side of a roadway to 5 provide long sight lines and leeway for driver error and recovery (1). In contrast, recent studies 6 in urban settings suggest that “human scale” streetscapes, which are smaller and more narrowly 7 enclosed by buildings and trees, may reduce crash risk by narrowing drivers’ fields of view and 8 encouraging slower, less risky driving behavior (2–11). Enclosure is the collective effect of large 9 objects surrounding a street, chiefly buildings and trees, to define the spatial extents of a 10 streetscape and restrict long sight lines; it is what some urban designers say makes a street feel 11 like an “outdoor room” (12) (Figure 1). Small and enclosed streetscapes are also consistent with 12 a host of livability goals, including increased land use density, walkability, and urban forestry 13 (13, 14). Traffic safety benefits offer yet another rationale for more densely developed 14 streetscapes, especially along urban arterials where high level of service for vehicular traffic 15 must be accommodated alongside complex land uses and design of vital public spaces. 16

17 FIGURE 1 Examples of Open Versus Enclosed Streetscapes. 18 19

To objectively investigate the effects of streetscape design on livability performance 20 measures such as traffic safety we use a GIS-based method for block-by-block measurement of 21 fundamental design characteristics—e.g. width, height, continuity of edges—forming a 22 streetscape skeleton that is elemental to user perceptions of spatial scale and enclosure. The GIS-23 based approach offers greater collection efficiency and measurement consistency compared with 24 manual auditing methods traditionally used to measure streetscape design (13, 15, 16). It also 25 offers more direct and spatially discrete measurements than built environment design proxies 26 based on from zoning, land use districts, or roadway characteristics such as the number travel 27 lanes (7, 17). Measurement of streetscapes at the resolution of individual blocks is particularly 28 important in urban settings, where there can be substantial design heterogeneity within 29 neighborhoods or along corridors. Spatially aggregate built environment measurements may not 30 adequately reveal the implications of design anomalies, such as vacant lots, along individual 31


street segments. GIS-based measurement of streetscape skeletons strikes a practical balance 1 between resolution of measured features and the feasibility of assessing large samples of 2 streetscapes. 3

This study investigates the safety implications of streetscape design at the sites of more 4 than 240,000 traffic crashes throughout New York City. Logistic regression models demonstrate 5 that crash severity was affected by streetscape size and enclosure variables based on the massing 6 of surrounding buildings and trees. Crashes were less likely to be severe—resulting in injury or 7 death—if they occurred in smaller or more enclosed streetscapes. Arterial classification was 8 associated with greater propensity for severe crashes, likely as a result of higher vehicle speed. 9 As such, streetscape design may provide a useful countermeasure for balancing the safety risks 10 posed by arterials. Infill development and tree planting to create smaller, more enclosed 11 streetscapes along urban arterials may improve traffic safety by encouraging more responsible 12 driving, complementary to traditional traffic calming techniques. 13 14 LITERATURE REVIEW 15 16 The traditional framework associating roadside design with traffic safety is rooted in rural and 17 suburban settings with ample space to either side of roadways and limited diversity of users. It 18 encourages consistently large shoulders and clear zones to allow ample space for recovery from 19 potential driver error or reaction to unforeseen circumstances, such as an animal darting into the 20 roadway (1) (Figure 2, A). More open space along roadsides provides an opportunity for drivers 21 to regain control of a wayward vehicle before crashing into a fixed object. 22

Unfortunately, such designs can also encourage higher speeds and riskier driving 23 behavior, negatively affecting traffic safety. This is particularly problematic along urban 24 arterials, which attempt to accommodate both heavy vehicular traffic and non-vehicular users 25 who are vulnerable to high-speed crashes. Urban arterials also have limited space for roadside 26 clear zones due to dense land uses and multimodal infrastructure such as sidewalks and bicycle 27 lanes. Traffic safety along urban arterials is, therefore, largely dependent on encouraging drivers 28 to maintain moderate speeds and avoid risky behavior. Rather than assume dense and complex 29 urban roadside environments are a hindrance to traffic safety, as the traditional framework 30 suggests, it is pertinent to investigate how enclosed streetscapes, which are often inherent to 31 urban settings, may instead improve safety by encouraging responsible driver behavior. 32

An alternative framework relating roadside design to traffic safety, potentially better 33 suited to urban settings, posits that drivers are more aware of potential hazards and engage in less 34 risky behavior when their environment is more constrained and offers less design consistency 35 (Figure 2, B) (3–7). It takes advantage of driver agency to respond to environmental conditions 36 with slower speeds, improving capacity to react to unpredictable events and reducing the severity 37 of crashes when they do take place. This framework may be particularly appropriate for urban 38 settings where complex traffic patterns and diversity of road users make speed a dominating 39 factor in crash severity (18–22). 40

More enclosed built environment designs with minimal building setbacks, smaller spaces 41 between buildings, and substantial street tree canopy, are also consistent with livability goals 42 promoting walkability, vibrant public spaces, and urban forestry (13). Determining how smaller 43 and more enclosed streetscapes are beneficial to traffic safety may eliminate the perceived 44 tradeoff between clear zones for traffic safety prioritization and “human scale” streetscapes for 45 livability. By using compact streetscape design to encourage responsible driving behavior it may 46


be possible for heavy vehicular traffic flows, bicyclists, pedestrians, and dense land uses to more 1 safely share arterial corridors as livable urban spaces. 2

3 FIGURE 2 Contrasting Frameworks for Safety Design. 4 5

Existing research has demonstrated the safety benefits of smaller, more enclosed 6 streetscapes in a generalized sense, but has inadequately investigated the effect of specific 7 streetscape design characteristics. Such studies have drawn on land use and roadway data as 8 indicators of built environment design rather than direct measurements of the streetscape. 9 Several papers by Dumbaugh have evaluated built environment-safety relationships using parcel-10 level land use data aggregated by census units (3, 5, 8, 11, 22). He and his coauthors consistently 11 found that streets lined by parcels with strip malls and big box stores were associated with 12 greater crash risk, while those lined by “pedestrian-scale” retail were less risky. The Dumbaugh 13 method assumes consistent site and architectural design among land uses, and does not capture 14 differences in building setback, spacing and height that have important effects on streetscape 15 enclosure. Other studies have used the number of traffic lanes or roadway width as an indicator 16 of streetscape size, though this captures little to nothing about the built environment surrounding 17 the roadway (4, 19). Neither of these methods sufficiently accounts for the role of street trees, 18 which provide important roof-life enclosure for many streetscapes. Increasing availability of high 19 resolution building and tree canopy GIS data in urban centers makes it more realistic to include 20 block-level streetscape design measurement in livability analyses. 21

Many studies have also evaluated relationships between urban form—largely discussed in 22 terms of street patterns and development density—and traffic safety, assuming that gridiron 23 street networks with smaller blocks and higher population density indicated smaller, more 24 enclosed streetscapes (3–5, 8, 10, 11, 22–25). These variables were consistently associated with 25 reduced crash risk, but potentially due to their effect on traffic flow and travel demand more than 26 streetscape design. Urban form logically impacts safety outcomes by affecting mode share—27 denser, more connected land uses require fewer vehicle trips (21)—and distribution of traffic 28 throughout the network. Sprawling land use patterns can feed traffic into a hierarchical structure, 29


redistributing crashes away from local streets and toward collector and arterial streets that are 1 also more likely to include car-focused retail and commercial land uses (11). Thus, there is 2 greater propensity for crashes where higher speed vehicular traffic, pedestrian flows and 3 commercial activity are concentrated. These conditions likely have a major effect on livability, 4 but are not necessarily the result of deliberate streetscape design. 5

As with all traffic safety research, those evaluating the effect of built environment design 6 have struggled to properly account for risk exposure. Standardizing crash rates by vehicle miles 7 traveled (VMT) has been the most common strategy to account for exposure; several studies 8 have used it as the sole exposure metric for all types of crashes, including those involving 9 pedestrians (3, 5–8, 24). Other researchers have constructed pedestrian exposure metrics using 10 land use proxies. Ukkusuri et al. (4) used population and the number of intersections as proxies 11 for pedestrian exposure. Ewing et al. (10) measured pedestrian exposure as the proportion of 12 work trips made by walking and public transit. Wang and Kockelman (25) used modeled 13 pedestrian traffic volume based on a host of land use variables. Each of these provided indirect 14 measures of exposure within areal units, making them particularly inadequate for disaggregate 15 analysis of crashes at specific locations. None accounted for nonlinearity between safety 16 performance and traffic volume by adjusting exposure according to validated safety performance 17 functions (26). Following Hanson et al. (19) and Moudon et al. (20), this study avoids the 18 necessity of exposure control by modeling the effect of streetscape design on the severity of 19 crashes rather than their occurrence. It investigates how specific streetscape design 20 characteristics may be treated as countermeasures to improve livability by reducing crash 21 severity. 22 23 METHODS 24 25 Study Area 26 27 New York City was an opportune study area for examining relationships between urban 28 streetscape design and crash severity because it offered a large sample of traffic crashes reported 29 with a consistent format and occurring in highly variable streetscape environments. With a 30 population of over eight million, more than 45,000 kilometers of public roadways 31 (approximately 8% classified as arterials), and an estimated 108 million kilometers of vehicle 32 travel in 2014, there are more than 100,000 crashes each year on New York City roadways (27, 33 28). The sheer number of crashes reported by the New York Police Department (NYPD) 34 provided a sample size and consistency that would have been difficult to achieve in other U.S. 35 cities. Moreover, the City of New York provided public access to high quality built environment 36 data, including building footprints, building heights, and tree canopy coverage, that allowed us to 37 precisely measure streetscape design at the location of each crash. 38 39 Data 40 41 All crash and streetscape design data used for this study were available from the New York City 42 OpenData web portal (27). The study assessed 244,684 NYPD-reported crash records dating 43 from August, 2011 through September, 2013. As demonstrated by Figure 3, crash sites were 44 distributed throughout the city. Each crash record included geographic coordinates and the 45 number of resulting injuries or deaths among drivers, passengers, bicyclists, and pedestrians. 46 Information about the types of vehicles involved, causes, and other factors were inconsistently 47


coded, so they were not used for analysis. Severity was calculated for each record based on 1 whether the crash resulted in an injury death: those with no injuries or deaths (73% of records) 2 were coded as 0 and labeled property damage; those resulting in at least one injury or death 3 (27% of records) were coded as 1 and labeled injury or death. This coding potentially 4 oversimplified the relationship between injury and death, but provided reasonably balanced 5 categories for analysis. Less than 1% of crashes in the sample resulted death. Among injury or 6 death crashes, 40% included the injury or death of at least one driver, 34% of at least one 7 passenger, 34% of at least one pedestrian, and 12% of at least one bicyclist. The dataset 8 unquestionably underrepresents pedestrian involvement because it counts them only if they were 9 injured or killed; no property damage crashes involved pedestrians. The dataset only reported 10 crashes involving at least one motor vehicle. Bicycle-pedestrian or bicycle-bicycle crashes were 11 not included in the database. 12

13 FIGURE 3 Crash Sites Distributed Throughout New York City. 14

15 Crash records were linked to streetscape design variables measured along street segments 16

at the approximate location of each crash. Because approximately 96% of crashes occurred at 17 intersections of two or more streets, variables were averaged among all segments within twenty 18


meters of each crash location. The twenty-meter radius was large enough to aggregate 1 measurements from streetscapes on opposite sides of large intersections while being small 2 enough to avoid inappropriate aggregation of adjacent, parallel blocks. While it would have 3 improved analytical precision to account for streetscape design only for segments along which 4 participants in each crash were traveling, pre-crash trajectories were not included in the crash 5 data. Moreover, because the scale and proportions of intersections were influenced by all 6 streetscapes leading into them, we considered it important to account for all potentially 7 influential streetscapes. 8

A GIS-based streetscape measurement method was used to measure each of the 73,764 9 street segments within twenty meters of the crash sites (29). Input GIS data for the method 10 included street centerline, building footprint and height, and tree canopy layers, all acquired from 11 the NYC OpenData web portal. The street centerline dataset represented approximate roadway 12 centerlines. Building footprint geometry and heights were originally derived from 13 orthophotography and stereoscopic height measurement. Tree canopy areas were originally 14 derived by the University of Vermont Spatial Analysis Lab using orthophotography and aerial 15 LiDAR data to precisely detect tree crowns in urban settings at one square meter resolution. 16 Intersections were defined where multiple centerlines intersected at a single point. Street 17 segments representing city blocks, the units of analysis, were defined as the centerline distance 18 between adjacent intersections. Segments less than twenty meters long, which largely 19 represented turning lanes within large intersections, were excluded from analysis, as were alleys, 20 ramps, bridges, tunnels, and expressways. 21

To “search” for streetscape edges the method drew progressively larger buffers at one-22 meter intervals from each side of each block-length centerline and evaluated the proportion of 23 each buffer overlapping with building footprints. A streetscape edge, where façades consistently 24 aligned, was identified at the buffer distance where the proportion of building overlap most 25 rapidly increased relative to the previous buffer. The method searched for façade alignment 26 within forty meters of each side of each centerline. Segments with no buildings within forty 27 meters were assumed to have no well-defined street wall. We considered this a reasonable search 28 distance because, to our knowledge, the widest New York City street with a well-defined street 29 wall was Eastern Parkway in Brooklyn, with a building-to-building width of approximately 30 eighty meters along two consecutive blocks. The space between streetscape edges, identified by 31 the buffering procedure, defined the horizontal extent of each streetscape. Within these extents, 32 each streetscape was measured for seven variables by gathering attribute and aerial statistics 33 from building footprint and tree canopy layers (Table 1). 34

Width was the distance between opposing streetscape edges (Figure 4, A). Importantly, 35 this represented the distance between opposing building facades, the field of view for a user. It 36 should be noted that this is distinctly different from the curb-to-curb roadway width typically 37 used for geometric analyses. Length was the centerline length of the one-block street segment 38 (Figure 4, B). Height was the average height of buildings along the single edge, of the two edges 39 along each segment, with the taller average height (Figure 4, C). Cross-sectional proportion, the 40 quotient of height divided by width, described the interaction of these dimensions (Figure 4, D). 41 Narrow streets lined by tall buildings had large cross-sectional proportions, creating highly-42 enclosed upright streetscapes, while wide streets lined by short buildings had small cross-43 sectional proportions, creating shallow streetscapes with less enclosure. Street wall continuity 44 was the proportion of an edge that intersected a façade and thus formed a street wall (Figure 4, 45 E). For each segment, street wall continuity was reported only for the more continuous of the two 46 sides. Buildings per length was the count of buildings along both sides of a segment standardized 47


by centerline length (Figure 4, F). Tree canopy coverage was the proportion of area between 1 edges that was covered by tree canopy (Figure 4, G). Streetscape design variables were measured 2 as constants along entire block-length segments, no matter their lengths. Dividing segments into 3 smaller, equal-length pieces might improve measurement precision but would substantially 4 increase processing time, result in short “remainder” segments at block ends, and might not have 5 revealed additional variability given the internal consistency in urban design along many blocks. 6 The one-block unit has also been commonly used by other studies surveying urban design 7 (13,15,16). 8

To control for potentially greater crash hazard presented by arterial streets designed to 9 prioritize speed and traffic volume, each crash was also coded with a dummy variable denoting 10 whether any intersecting street was classified as arterial according to US Census Feature Class 11 Codes included in 2012 ESRI StreetMap centerline data. Crashes on or near arterials comprised 12 8.8% of the sample. 13

14 TABLE 1 Streetscape Design Variable Descriptive Statistics 15 16 Streetscape

Design Variable Spatial Definition Mean Std. Dev. Min. Max.

1 Width (meters) Distance between edges (building-to-building across the street)

35.5 12.7 5.7 80.0

2 Length (meters) Centerline distance between intersections 116.5 37.4 20.1 716.1

3 Height (meters) Average building height, highest side 14.8 16.7 0.0 199.3

4 Cross-sectional proportion Width (building-to-building)/ Height on the highest side

0.658 0.870 0.000 16.793

5 Street wall continuity Proportion of edge intersecting buildings on the more continuous side

0.613 0.173 0.000 1.000

6 Buildings per length Count of buildings on both sides/Length 0.088 0.050 0.000 0.312

7 Tree canopy coverage Proportion of street area covered by tree canopy

0.161 0.124 0.000 0.942

17


1

2 FIGURE 4 Streetscape Design Variables. 3 4


Modeling 1 2 Binary logistic regression with IBM SPSS Statistics 21 was used to examine whether the 3 likelihood of a crash resulting in injury or death was affected by streetscape design variables and 4 arterial classification. The model was appropriate for predicting the likelihood of binary crash 5 outcomes with readily interpretable results. Our implementation estimated a constant and 6 coefficients for each streetscape variable to maximize the likelihood of correctly predicting crash 7 severity. Because logistic regression assumes predictor independence, and several of the 8 streetscape design variables were calculated as a function of one another—cross-sectional 9 proportion was, for instance, the quotient of height and width variables—two separate models 10 were constructed to adhere to this assumption. Model A examined the effects of variables 11 measuring streetscape size: width, length, and height. Model B examined the effects of variables 12 accounting for streetscape proportions, particularly those related to enclosure: cross-sectional 13 proportion, street wall continuity, buildings per length, and tree canopy coverage. Both models 14 included the arterial street dummy variable. Model terms were judged for significance according 15 to Wald Chi-Square tests with a P = 0.05 threshold for assumption of null hypothesis validity. 16 Overall model fit was judged according to the Nagelkerke pseudo R2. 17 18 RESULTS AND DISCUSSION 19 20 The results from both models (Table 2) support the premise that crashes on more enclosed, 21 “human scale” streetscapes are less likely to be severe, while those on or at the intersection of 22 arterial streets are more likely severe. Model A indicates small but significant effects among 23 streetscape size variables. The odds of a given crash resulting in injury or death increase 0.5% 24 for each additional meter of streetscape width. It is important to note that this width includes not 25 just paved roadway, but shoulders and building setbacks that contribute to a streetscape’s visual 26 field. Thus, a narrow roadway with buildings set back a substantial distance to either side would 27 constitute a wide streetscape and would be more likely to foster severe crashes than a similar 28 roadway with smaller building setbacks. Severe crashes are also less likely where blocks are 29 shorter, likely as a result of reduced distance for vehicles to pick up speed between intersections. 30 Each additional meter of block length increases the odds of a severe crash by 0.1%. Streetscape 31 height, in contrast, shows a negative relationship. Each additional meter in the average height of 32 buildings along the taller side of a streetscape lowers the odds of severe crashes by 0.2%. Thus, 33 severe crashes are least likely on streetscapes that are tall and narrow with short block lengths. 34 This finding is consistent with the premise that more constrained fields of view may improve 35 safety outcomes by encouraging slower, less risky driving. It should be noted that field of view, 36 in this case, refers to sight lines in all directions. As such, a streetscape might provide ample 37 sight distance along the length of a roadway, allowing adequate detection of oncoming traffic 38 hazards, while promoting caution through short sight lines that provide a sense of enclosure to 39 either side. 40

Model B similarly demonstrates that proportions contributing to streetscape enclosure 41 substantially affect crash severity. Each unit increase in cross-sectional proportion, the ratio of 42 streetscape height to width, reduces the odds of severe crashes by nearly 9%. While the majority 43 of streetscapes have cross-sectional proportions less than one, those in downtown areas may have 44 proportions where the height of surrounding buildings exceeds the width by a factor of ten or 45 more, nearly eliminating the likelihood of a severe crash according to the model. High street wall 46 continuity similarly reduces the odds of severe crashes. Odds of severe crashes in streetscapes 47


with highly continuous street walls are nearly 9% lower than in streetscapes with large spaces 1 between buildings. This highlights the multifaceted capacity of infill development—filling holes 2 in the urban fabric presented by vacant lots or relatively undeveloped spaces such as parking lots 3 and gas stations—to improve traffic safety while simultaneously providing a more appealing 4 public realm for pedestrians. Interestingly, while the number of buildings within a streetscape is 5 cited as an important criteria for pedestrian appeal (13)—a greater number of buildings provides 6 more opportunities for interesting stylistic variation—it has no significant effect on crash 7 severity. Traffic safety appears to be affected only by basic size and proportions of streetscape 8 design rather than more nuanced aspects of their design. 9

Enclosure due to tree canopy appears to have the largest implications for crash severity, 10 indicating that natural resource management may have important and overlooked role in traffic 11 safety and livability. Crashes on streetscapes fully covered by tree canopy are 51% less likely to 12 result in injury or death than those on streetscapes without trees. While street trees require time 13 and financial investment to plant, grow, and maintain in maturity, where they likely offer the 14 most benefit, trees may offer an efficient option for improving streetscape enclosure along streets 15 with few existing buildings or large setbacks. Infill building development is a complex and 16 piecemeal process demanding cooperation among private landowners, financers, and planning 17 agencies. In contrast, tree planting and maintenance within public rights-of-way may be 18 spearheaded by a comparatively small set of stakeholders and supported by urban forestry 19 initiatives that already exist in many cities. It can take decades for trees to grow to maturity, but 20 infrastructure for trees may be planned and installed along an entire street in a relatively short 21 timeframe. 22

Both models indicate that crashes along or at intersections with arterial streets are 23 approximately 20% more likely to result in injury or death compared with those away from 24 arterials. This result is consistent with the assumption that greater vehicle speeds are associated 25 with more severe crashes. Interestingly, the magnitude of effect from arterial classification, 26 which is similar in both models, is not substantially greater than the counterbalancing effects 27 offered by several streetscape design variables, chief among them tree canopy. If even 40% an 28 arterial streetscape is covered by tree canopy, the result of just a few mature trees along a typical 29 city block, this may offset the hazard presented by arterial classification. Increased propensity for 30 severe crashes along urban arterials further substantiates the value of streetscape design to 31 moderate crash risk in these places. 32

While the models indicate safety benefits can be gleaned from streetscape design, it is 33 important to recognize that neither model predicts crash outcomes with any certainty. Both have 34 Nagelkerke pseudo R2 values less than 0.01 and classify 100% of crashes as property damage, 35 resulting in approximately 27% classification error. Their lack of predictive power, even while 36 model terms have highly significant effects, indicates the complexity of factors affecting crash 37 outcomes. Factors that are most influential to crash severity, such as seat belt use, are often 38 related to driver behavior and equipment rather than environmental conditions (30). Nonetheless, 39 the models indicate that livable streetscape design contributes to traffic safety improvements, and 40 should be considered alongside more traditional traffic safety countermeasures. At the very least, 41 smaller and more enclosed streetscapes, which may be valuable to street livability in terms of 42 walkability and economic development, should not be considered a traffic safety hazard. 43 Combined land use and transportation planning in urban settings should challenge conventional 44 wisdom about the necessity for wide, unobstructed roadside clear zones to promote traffic safety, 45 making way for a more context-sensitive approach to designing livable arterial streetscapes. 46


1 TABLE 2 Binary Logistic Regression Modeling Results 2 3 MODEL A – Streetscape Size

Response: Crash Severity (1 = injury or death, 0 = property damage)

N = 244,689

Predictors β Odds Ratio

Width (meters) 0.005* 1.005

Length (meters) 0.001* 1.001

Height (meters) -0.002* 0.998

Arterial (1 = arterial, 0 = non-arterial) 0.187* 1.205

Constant -1.261* 0.283

Model Fit

-2 log likelihood: 285567.073

Nagelkerke R2: 0.003

MODEL B – Streetscape Proportions/Enclosure

Response: Crash Severity (1 = injury or death, 0 = property damage)

N = 244,689

Predictors β Odds Ratio

Cross-sectional proportion (ratio) -0.093* 0.911

Street wall continuity (ratio) -0.088* 0.916

Buildings per length (ratio) -0.002† 0.988

Tree canopy coverage (ratio) -0.711* 0.491

Arterial (1 = arterial, 0 = non-arterial) 0.167* 1.181

Constant -0.777* 0.460

Model Fit

-2 log likelihood: 285333.383

Nagelkerke R2: 0.004

* P < 0.01 † P > 0.9

4 Results from this study should be interpreted with consideration for limitations inherent 5 to the input data and modeling technique. Chiefly, the study only evaluates crashes and 6 streetscapes within New York City because both types of data were readily accessible for this 7 geography. While New York City includes impressive urban design heterogeneity, it is 8 nonetheless one of the densest urban environments in the western hemisphere, with built 9 environment, vehicle traffic, bicycling, and pedestrian characteristics that are unrepresentative of 10 other cities, particularly newer, sprawling cities in the southern and southwestern United States. 11


Further research should gather appropriate data and extend analysis into cities with diverse urban 1 form and traffic characteristics. Generalized measurement of urban design characteristics at crash 2 sites also limits the precision of our results. The GIS-based streetscape measurement method 3 provided novel efficiency and consistency, but could be improved to offer higher resolution 4 measurements at more precise locations. Moreover, if crash data were more precisely located and 5 included information about direction of travel, the analysis could account for urban design 6 influences at more discrete locations rather than generalizing within a constant radius of each 7 crash site. 8 Crash, building, and tree data used for analysis were subject to error in coding and spatial 9 position. Crash data based on police reports may have incompletely described injuries from 10 crashes involving many people. Spatial position of crashes may have also been imprecise or 11 inconsistent due to various strategies—cross streets, street addresses, GPS coordinates—12 potentially used to report location. Building footprints and tree canopy areas were both drawn 13 using automated and manual process, generalizing their shapes and introducing spatial 14 inaccuracies through machine misinterpretation and human error. Streetscape design 15 measurements were affected by these inaccuracies, though their overall impact on our analysis 16 was likely small. 17 Logistic regression modeling of binary crash severity provided readily interpretable 18 results but with limited precision. This modeling technique was prompted, in part, by the 19 simplistic coding of crash data, which reported only injuries and deaths without nuanced severity 20 information. With more specific injury data, future studies might discriminate between minor 21 and severe injuries using ordinal models. They might also predict crash counts by location, or the 22 count of injuries or deaths as an indicator of crash magnitude. These methods, however, would 23 require accounting for exposure among vehicles, bicycles, and pedestrians, which is difficult to 24 determine accurately for specific locations along a large sample of streets. 25 26 CONCLUSION 27 28 While traditional traffic safety suggests safety benefits from clear zones which are difficult to 29 achieve in urban settings, our analysis suggests that crashes in urban contexts are less likely to be 30 severe when they take place in smaller, more enclosed streetscapes. This finding is consistent 31 with other studies proposing that speed is a primary factor in crash severity, especially in urban 32 settings where there are complex traffic patterns and diverse road users, and that drivers operate 33 at slower speeds when their field of vision is more constrained by smaller or more enclosed 34 spaces (31–33). Identifying the safety benefits of smaller and enclosed streetscapes recasts 35 streetscape design as an opportunity for constructive countermeasures rather than a hindrance to 36 traditional safety engineering standards. Streetscape skeleton variables might even be considered 37 as predictor variables for crash modification models such as those contained in the Highway 38 Safety Manual (34). 39 Safety benefits offered by specific characteristics of streetscape design can only be 40 adequately identified with precise measurement of a large number of streetscapes where crashes 41 have occurred. This study uses an efficient, GIS-based method to capture streetscape design 42 measurements along tens of thousands of city blocks, allowing expedient assessment of a suite of 43 design variables at the locations of tens of thousands of crashes. Compared with previous studies 44 drawing on land use or roadway characteristics as generalized indicators of “human scale” or 45 enclosed streetscapes, this strategy unpacks the contribution of specific design characteristics to 46 safety outcomes. Understanding how variables such as streetscape width, street wall continuity, 47


and tree canopy coverage can reduce propensity for severe crashes may help planners make 1 recommendations that are readily translated into design practice. 2 There is a need to improve livability along urban arterials that are challenged to provide 3 high levels of service for vehicular traffic while maintaining a safe and attractive environment 4 for non-vehicular users and commercial activity. To reduce crash severity, planners may promote 5 infill development that creates smaller, more enclosed streetscapes with greater street wall 6 continuity, taller buildings, and narrower setbacks. Dense built environments, however, often 7 take decades to develop. In the short term, street tree planting and maintenance may be the most 8 efficient approach to creating more enclosed streetscapes. Even small trees, planted close 9 together, may provide a sense of enclosure along a roadway. Trees can be installed relatively 10 efficiently, often on public land, under the direction of municipal agencies. Moreover, infill 11 development and street trees are consistent with livability goals such as place making, 12 walkability, destination accessibility, and provision of natural amenities. By demonstrating that 13 small and enclosed urban streetscapes are also a traffic safety asset, this study opens the door to 14 countermeasure design that aligns with other aspects of livability planning and can help urban 15 arterials comfortably accommodate their diverse roles. 16


REFERENCES 1 2

1. American Association of State Highway and Transportation Officials. A policy on 3 geometric design of highways and streets, 2011. American Association of State Highway 4 and Transportation Officials, Washington, D.C., 6th Ed. 2011. 5

2. Gårder, P. E. The impact of speed and other variables on pedestrian safety in Maine. 6 Accid. Anal. Prev., Vol. 36, No. 4, 2004, pp. 533–42. 7

3. Dumbaugh, E., and W. Li. Designing for the Safety of Pedestrians, Cyclists, and Motorists 8 in Urban Environments. J. Am. Plan. Assoc., Vol. 77, No. 1, 2010, pp. 69–88. 9

4. Ukkusuri, S., L. F. Miranda-Moreno, G. Ramadurai, and J. Isa-Tavarez. The role of built 10 environment on pedestrian crash frequency. Saf. Sci., Vol. 50, No. 4, 2012, pp. 1141–11 1151. 12

5. Dumbaugh, E., and Y. Zhang. The Relationship between Community Design and Crashes 13 Involving Older Drivers and Pedestrians. J. Plan. Educ. Res., Vol. 33, No. 1, 2013, pp. 14 83–95. 15

6. Dumbaugh, E., and J. L. Gattis. Safe Streets, Livable Streets. J. Am. Plan. Assoc., Vol. 71, 16 No. 3, 2005, pp. 283–300. 17

7. Dumbaugh, E. Design of Safe Urban Roadsides: An Empirical Analysis. Transp. Res. 18 Rec., Vol. 1961, No. 1, 2006, pp. 74–82. 19

8. Dumbaugh, E., R. Rae, and D. Wunneberger. Using GIS to develop a performance-based 20 framework for evaluating urban design and crash incidence. URBAN Des. Int., Vol. 16, 21 No. 1, 2011, pp. 63–74. 22

9. Jones, D., and M. K. Jha. Recent Advances in Computer Engineering and Applications-23 Proc. of the 4th WSEAS Int. Conf. on Computer Engineering and Applications, CEA., 24 2010. 25

10. Ewing, R., R. A. Schieber, and C. V. Zegeer. Urban sprawl as a risk factor in motor 26 vehicle occupant and pedestrian fatalities. Am. J. Public Health, Vol. 93, No. 9, 2003, pp. 27 1541–5. 28

11. Dumbaugh, E., and R. Rae. Safe Urban Form: Revisiting the Relationship Between 29 Community Design and Traffic Safety. J. Am. Plan. Assoc., Vol. 75, No. 3, 2009, pp. 30 309–329. 31

12. Cullen, G. The Concise Townscape. Van Nostrand Reinhold, New York, 1971. 32 13. Ewing, R., and S. Handy. Measuring the Unmeasurable: Urban Design Qualities Related 33

to Walkability. J. Urban Des., Vol. 14, No. 1, 2009, pp. 65–84. 34 14. Locke, D. H. et al. Prioritizing Preferable Locations for Increasing Urban Tree Canopy in 35

New York City. Cities Environ., Vol. 3, No. 1, 2010, pp. 1–18. 36 15. Clifton, K. J., A. D. Livi Smith, and D. Rodriguez. The development and testing of an 37

audit for the pedestrian environment. Landsc. Urban Plan., Vol. 80, No. 1-2, 2007, pp. 38 95–110. 39

16. Day, K., M. Boarnet, M. Alfonzo, and A. Forsyth. The Irvine-Minnesota inventory to 40 measure built environments: Development. Am. J. Prev. Med., Vol. 30, No. 2, 2006, pp. 41 144–52. 42

17. Ewing, R., R. Schieber, and C. V. Zegeer. Urban sprawl as a risk factor in motor vehicle 43 occupant and pedestrian fatalities. Am. J. Public Health, Vol. 93, No. 9, 2003, pp. 1541–5. 44

18. Noland, R. B. Traffic fatalities and injuries: the effect of changes in infrastructure and 45 other trends. Accid. Anal. Prev., Vol. 35, No. 4, 2003, pp. 599–611. 46


19. Hanson, C. S., R. B. Noland, and C. Brown. The severity of pedestrian crashes: an 1 analysis using Google Street View imagery. J. Transp. Geogr., Vol. 33, 2013, pp. 42–53. 2

20. Moudon, A. V., L. Lin, J. Jiao, P. Hurvitz, and P. Reeves. The risk of pedestrian injury 3 and fatality in collisions with motor vehicles, a social ecological study of state routes and 4 city streets in King County, Washington. Accid. Anal. Prev., Vol. 43, No. 1, 2011, pp. 11–5 24. 6

21. Ewing, R., and E. Dumbaugh. The Built Environment and Traffic Safety: A Review of 7 Empirical Evidence. J. Plan. Lit., Vol. 23, No. 4, 2009, pp. 347–367. 8

22. Dumbaugh, E., W. Li, and K. Joh. The built environment and the incidence of pedestrian 9 and cyclist crashes. URBAN Des. Int., Vol. 18, No. 3, 2013, pp. 217–228. 10

23. Sun, J., and G. Lovegrove. Comparing the road safety of neighbourhood development 11 patterns  : traditional versus sustainable communities. Can. J. Civ. Eng., Vol. 45, No. 12 October 2012, 2013, pp. 35–45. 13

24. Gladhill, K., and C. M. Monsere. Exploring Traffic Safety and Urban Form in Portland, 14 Oregon. Transp. Res. Rec. J. Transp. Res. Board, Vol. 2318, 2012, pp. 63–74. 15

25. Wang, Y., and K. M. Kockelman. A Poisson-lognormal conditional-autoregressive model 16 for multivariate spatial analysis of pedestrian crash counts across neighborhoods. Accid. 17 Anal. Prev., Vol. 60, 2013, pp. 71–84. 18

26. Ivan, J. New Approach for Including Traffic Volumes in Crash Rate Analysis and 19 Forecasting. Transp. Res. Rec., Vol. 1897, No. 1, 2004, pp. 134–141. 20

27. NYC Digital. NYC OpenData. 2014, Available at: https://nycopendata.socrata.com/. 21 28. Miller, S. TSTC: NYC’s Regional Planners Underestimate the Shift Away From Driving. 22

StreetsBlog NYC, 2013, Available at: http://www.streetsblog.org/2013/09/03/bending-to-23 reality-projections-say-ny-area-driving-to-grow-more-slowly/ [Accessed April 26, 2014]. 24

29. Harvey, C. Measuring Streetscape Design for Livability Using Spatial Data and Methods. 25 University of Vermont, 2014. 26

30. Vianoa, D. C., and C. S. Parenteaua. Severe-to-Fatal Injury Risks in Crashes With Two 27 Front-Seat Occupants by Seat Belt Use. Traffic Inj. Prev., Vol. 11, No. 3, 2010, pp. 294–28 299. 29

31. Naderi, J. R., B. S. Kweon, and P. Maghelal. The Street Tree Effect and Driver Safety. 30 ITE J., Vol. 78, No. February, 2008, pp. 69–73. 31

32. Naderi, J. R. Landscape Design in Clear Zone: Effect of Landscape Variables on 32 Pedestrian Health and Driver Safety. Transp. Res. Rec., Vol. 1851, No. 03-3937, 2003, pp. 33 119–130. 34

33. Gatiss, J. L. Urban Street Cross Section and Speed Issues. TRB Circ., Vol. E-C019: Ur, 35 2000. 36

34. Highway Safety Manual. AASHTO, Washington, D.C., 2010. 37

Documents

URBAN STREETSCAPE DESIGN AND CRASH SEVERITY …docs.trb.org/prp/15-2942.pdf1 URBAN STREETSCAPE DESIGN AND CRASH SEVERITY ... 19 which provide important roof-life enclosure for many