Citation
Child pedestrian and bicyclist safety : a proactive approach via safety perceptions

Material Information

Title:
Child pedestrian and bicyclist safety : a proactive approach via safety perceptions
Creator:
Ferenchak, Nicholas Nathan
Place of Publication:
Denver, CO
Publisher:
University of Colorado Denver
Publication Date:
Language:
English

Thesis/Dissertation Information

Degree:
Doctorate ( Doctor of philosophy)
Degree Grantor:
University of Colorado Denver
Degree Divisions:
College of Engineering and Applied Sciences, CU Denver
Degree Disciplines:
Engineering and applied science
Committee Chair:
Janson, Bruce
Committee Members:
Marshall, Wesley E.
McAndrews, Carolyn
Weible, Christopher M.
Thomas, Deborah S. K.

Notes

Abstract:
The purpose of this dissertation is to develop a more comprehensive manner of considering non-motorized transportation safety. I accomplish this through three papers. First, I perform a traditional crash-based analysis that – while identifying new urban areas that deserve attention in terms of pedestrian safety – sheds light on the shortcomings of reactive traffic safety approaches. Then, I build our theoretical framework for proactive safety analyses, develop a set of methodologies, and administer a survey to explore the relationship between safety perceptions and pedestrian and bicyclist travel behavior. Finally, I apply this proactive pedestrian and bicyclist safety approach and compare its results to reactive analyses’ results. If our goal is to get more people walking and biking safely – as opposed to simply reducing the number of crashes – then findings suggest that proactive analyses can contribute new and meaningful perspectives on pedestrian and bicyclist safety.

Record Information

Source Institution:
University of Colorado Denver
Holding Location:
Auraria Library
Rights Management:
Copyright Nicholas Nathan Ferenchak. Permission granted to University of Colorado Denver to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.

Downloads

This item has the following downloads:


Full Text
CHILD PEDESTRIAN AND BICYCLIST SAFETY:
A PROACTIVE APPROACH VIA SAFETY PERCEPTIONS
by
NICHOLAS NATHAN FERENCHAK B.A., Lafayette College, 2010 M.A., West Chester University, 2013
A dissertation submitted to Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Engineering and Applied Science Program
2018


©2018
NICHOLAS NATHAN FERENCHAK
ALL RIGHTS RESERVED


This thesis for the Doctor of Philosophy degree by Nicholas Nathan Ferenchak has been approved for the Engineering and Applied Science Program by
Bruce Janson, Chair Wesley E. Marshall, Advisor Carolyn McAndrews Christopher M. Weible
Deborah S. K. Thomas


Ferenchak, Nicholas Nathan (PhD, Engineering and Applied Science Program)
Child Pedestrian and Bicyclist Safety: A Proactive Approach via Safety Perceptions Thesis directed by Associate Professor Wesley E. Marshall
ABSTRACT
The purpose of this dissertation is to develop a more comprehensive manner of considering non-motorized transportation safety. I accomplish this through three papers. First, I perform a traditional crash-based analysis that — while identifying new urban areas that deserve attention in terms of pedestrian safety — sheds light on the shortcomings of reactive traffic safety approaches. Then, I build our theoretical framework for proactive safety analyses, develop a set of methodologies, and administer a survey to explore the relationship between safety perceptions and pedestrian and bicyclist travel behavior. Finally, I apply this proactive pedestrian and bicyclist safety approach and compare its results to reactive analyses’ results. If our goal is to get more people walking and biking safely — as opposed to simply reducing the number of crashes — then findings suggest that proactive analyses can contribute new and meaningful perspectives on pedestrian and bicyclist safety.
The form and content of this abstract are approved. I recommend its publication.
Approved: Wesley E. Marshall
IV


ACKNOWLEDGEMENTS
I would like to first thank my adviser, Wesley Marshall, whose tireless guidance and keen transportation insights have done much to aid the development of both my career and me as a person. I cannot thank you enough for taking the time to encourage and advise me through the last four years. I would also like to express my gratitude to everyone in the Civil Engineering Department at the University of Colorado Denver, especially Bruce Janson — whose help through numerous classes, personal correspondences, and as the chair of this committee has not only done much to strengthen my quantitative abilities but also shown me new and perceptive ways of looking at transportation. I would be remiss to not mention the guidance I have received from work being done by the Urban and Regional Planning Department as well as everyone on this committee, which helped provide direction for this work. I am also indebted to colleagues too numerous to mention here with whom I have collaborated and learned from over the last four years as I advanced through this program.
I would like to express my deep gratitude to Laura for being the most incredibly supporting and loving partner anyone could ask for. Thank you for always being willing to discuss ideas when I was stuck, helping me to see things from different perspectives, aiding whenever there was a deadline looming, and always being there for me in every aspect. These have been four of the most enjoyable and rewarding years of my life, and it was in a very large part thanks to you. Thank you for everything. To my parents, thank you for always being there for me throughout my life and for instilling in me the value of hard work and persistence. Those characteristics were certainly vital over the last four years. In addition to support from the rest of my family, I would also like to honor Jennifer Jacksits who provided motivation for my work and kept me going when I might otherwise have given up.
v


The COMIRB submission for protocol 17-1624 for the Safe Routes to School survey (Submission ID APP001-1) was awarded an exemption on September 21, 2017. This work was supported by the Mountain-Plains Consortium University Transportation Center (grant numbers MPC-515 and MPC-557). The Mountain-Plains Consortium was not involved in study design, analysis, or writing of the papers. Wesley Marshall was co-author for each paper.
vi


TABLE OF CONTENTS
CHAPTER
I. INTRODUCTION............................................................1
II. REDEFINING THE CHILD PEDESTRIAN SAFETY PARADIGM
Introduction............................................................4
Data....................................................................6
Crash Data...........................................................7
Exposure Data........................................................7
Child-Friendly Destinations Data.....................................8
Methods................................................................10
Phase I Study Methodology...........................................10
Phase II Study Methodology..........................................12
Results................................................................13
Conclusion.............................................................14
III. QUANTIFYING SUPPRESSED CHILD PEDESTRIAN AND BICYCLE TRIPS
Introduction...........................................................18
Theory.................................................................19
Data...................................................................22
Parental Perceptions Data...........................................22
Population and Built Environment Data...............................27
Methods................................................................27
Suppression Rates...................................................28
Network Analysis....................................................29
Suppressed Trips....................................................31
vii


Results
32
Trip Suppression Factors.....................................................32
Trip Suppression Rates.......................................................36
Number of Suppressed Trips...................................................43
Discussion......................................................................44
Conclusion......................................................................49
IV. SUPPRESSED CHILD PEDESTRIAN AND BICYCLE TRIPS AS AN INDICATOR OF SAFETY
Introduction....................................................................53
Theory..........................................................................55
Data............................................................................58
Reactive Analyses............................................................59
Existing Safety Reports..............................................59
Crash Cluster Analysis...............................................60
Proactive Analysis...........................................................61
Methods.........................................................................62
Reactive Analyses............................................................63
Existing Reports.....................................................63
Crash Cluster Analysis...............................................64
Proactive Analysis...........................................................66
Results.........................................................................72
Reactive Analyses............................................................72
Existing Reports.....................................................72
Crash Cluster Analysis...............................................79
viii


Proactive Analysis..............................................82
Comparison......................................................88
Equity Analysis.................................................92
Conclusion.........................................................93
V. CONCLUSION.........................................................95
REFERENCES...............................................................98
ix


CHAPTER I
INTRODUCTION
The goal of this dissertation is to develop a more comprehensive traffic safety analysis methodology for pedestrians and bicyclists through a proactive approach. I developed this proactive approach through the administration of a survey intending to better understand the relationship between demographics, roadway characteristics, traffic safety perceptions, and behavior. I then compare results from this proactive analysis to results from a reactive crash-based analysis to determine how the safety perspectives of the different approaches align and vary.
Traffic safety is a pressing public health issue. In 2016, the number of people killed on American roadways rose to 37,461, a level not seen since 2007. This translates to more than 102 people (on average) killed every day in the United States. In addition, there were an estimated 2,443,000 reported injuries resulting from motor vehicle crashes in 2015 (NHTSA 2016). These safety issues are especially serious for vulnerable populations such as children and non-motorized users. Traffic fatalities are the leading cause of death for Americans between the ages of 10 and 24 years, while there were 5,987 pedestrians and 840 bicyclists killed in 2016, numbers that overrepresent the share of trips made by these modes (CDC 2017; NHTSA 2017).
If we are to improve these traffic crashes, injuries, and fatalities, we must better understand them. Where are they occurring? What is causing them? Who is most affected? Analyzing crash data to identify patterns can help us to answer these questions. By understanding where, why, and when crashes are occurring, we can work towards avoiding them. Certainly, examining crashes has done much to improve traffic safety on our streets.
However, examining pedestrian and bicyclist crashes is a reactive approach, one that necessitates that we wait for a crash to occur and then try to understand why it happened. In other words, we must wait for these safety issues to manifest themselves before we can identify and fix
1


them. Another way to think about this reactive perspective is that the only pedestrians and bicyclists we are considering in a crash-based analysis are those who are on the roads in the first place — the people that have determined that the road is safe enough to try to walk or bike on. What about the people that — because of traffic safety concerns — decide not to walk or bike in the first place? These people are not getting hit and are therefore not considered in reactive analyses. Furthermore, the only locations that are considered when taking a reactive crash-based approach are those that people deem safe enough to walk and bike on in the first place. What about the areas where — because of traffic safety concerns — levels of walking and biking have been suppressed?
To identify these people and places, we need a proactive approach to identify safety issues before they manifest themselves, or even before any walking or biking occurs at all. Such a proactive pedestrian and bicyclist approach would be able to identify areas where safety concerns have suppressed walking and biking trips, thereby lowering levels of walking and biking activity, lowering the quantity of crashes, and hiding safety issues from the objective eye. Trips suppressed specifically because of traffic safety concerns becomes our new proactive safety metric with which we can then compare reactive and proactive safety approaches.
We accomplish these objectives through three papers. First, in Chapter II, we complete a reactive crash-based analysis of child pedestrian fatalities in six of the quickest growing American cities. Along with identifying parks as an urban area that deserves additional focus in terms of child pedestrian safety, this study reveals some of the drawbacks of such a reactive approach to safety. Then, in Chapter III, we develop a proactive analysis framework, create a methodology, and administer a survey to explore the relationship between traffic safety perceptions and child pedestrian and bicyclist trip suppression in Denver, Colorado. Findings show that sidewalks are the most important roadway factor for walking trip suppression, vehicle volumes and bike lanes are the most important factors for biking trip suppression, and trip suppression is concentrated near
2


network characteristics such as barrier connections and tributary configurations. These results are used in Chapter IV to implement a proactive child pedestrian and bicyclist safety analysis in Denver. We finally compare results from this proactive safety analysis with results from reactive analyses to understand what additional insights a proactive approach may be able to lend.
Findings suggest that proactive analyses add new and important perspectives on pedestrian and bicyclist safety. Specifically, such approaches identify areas where trips would be expected to occur if it were not for the presence of inadequate facilities. This is opposed to results from a reactive crash-based approach that focuses on areas with high levels of existing exposure. Proactive methods are also capable of providing fine-grained results within neighborhoods as opposed to reactive results focused primarily on major roads. If our goal is to reduce the number of pedestrian and bicyclist crashes, then a crash-based analysis makes sense. However, if our goal is to get more people walking and biking safely, then we also need to consider safety from a proactive perspective.
3


CHAPTER II
REDEFINING THE CHILD PEDESTRIAN SAFETY PARADIGM: IDENTIFYING HIGH FATALITY CONCENTRATIONS IN URBAN AREAS1
Introduction
Child pedestrians are some of the most vulnerable users of our transportation systems, and they deserve particular attention when we consider traffic safety. The objective of this work is to identify urban locations in which child pedestrians are at particular risk for fatal collisions with vehicles. This paper examines thirty years of crash data for six American cities in order to locate areas with high child pedestrian fatality concentrations. Phase I of the study, which examines Denver, CO, reveals higher concentrations of child pedestrian fatalities around parks as compared to other areas that children have been shown to frequent. In Phase II of the study, we specifically examine fatality concentrations near parks as compared to schools. Statistical analyses suggest that, once exposure is controlled for, child pedestrian fatalities concentrate around parks in densities 1.04 to 2.23 times higher than around schools. Also, the concentration of child pedestrian fatalities around parks is 1.16 to 1.81 times higher than the respective citywide concentration. Traffic risks for children around parks deserve further examination as we pursue the goals of Vision Zero and child safety on our streets.
Walking for transportation during childhood has important health and social benefits as it encourages physical activity and independence (Larsen, Buliung, & Faulkner 2013; Loukaitou-Sideris & Sideris 2009). Yet, children are often not able or allowed to safely and comfortably walk to their destinations. Traffic safety is one of the primary barriers to such active transportation in children (Centers for Disease Control and Prevention 1999). Motor vehicle collisions are the leading cause
1 Portions of this chapter were previously published in Injury Prevention (Issue 23-6, 2017) and are included with the permission of the copyright holder.
4


of death for individuals from the age of 4 through the age of 24 in the United States, with pedestrians being the second most at-risk user type (Centers for Disease Control and Prevention 2014). Every hour, an average of 40 children die on roadways around the world, most of whom are vulnerable road users such as pedestrians (Toroyan & Peden 2007). Despite the unfortunate road safety statistics of child pedestrians and the known health benefits of childhood walking, our transportation networks remain alarmingly dangerous for the few children that still walk independently. The question addressed through this work is: are there other land uses where we should be focusing our resources — beyond our traditional focus on schools — to alleviate large concentrations of child pedestrian fatalities?
Many researchers and practitioners have exerted considerable effort exploring child pedestrian safety around schools. These researchers and practitioners have found success when the necessary resources are allotted to combat the problem near school grounds. For example, reduced speed limits in school zones have been shown to lower vehicle speeds while projects funded by the Safe Routes to School program have reduced child pedestrian injury rates (Graham & Sparkes 2010; Abdul-Hanan, King, & Lewis 2011; Dumbaugh & Frank 2007; DiMaggio & Li 2013; Orenstein, Gutierrez, Rice, et al. 2007). Flowever, other locations within our cities that are frequented by children remain relatively unexplored (Kattan, Tay, & Acharjee 2011; Tay 2009). The scant literature on the subject suggests that the areas around trails have relatively few child pedestrian crashes, while other research found that areas with few child pedestrian injuries contained a prevalence of parks and play areas, and similarly, that areas at high risk for traffic crashes involving pedestrians under the age of 15 were characterized by an absence of parks (Stutts & Flunter 1999; Kraus, Flooten, Brown, et al. 1996; Joly, Foggin & Pless 1991). Furthermore, an analysis of child injuries associated with playground visits in the United States found that pedestrian injuries were so uncommon that a statistical analysis was not possible (Phelan, Khoury, Kalkwarf, et al. 2001). This current work will
5


fill the gap in the literature by further exploring concentrations of child pedestrian fatalities throughout our urban areas.
In Phase I of the study, we use spatial and statistical analyses to compare child pedestrian fatality concentrations around schools to concentrations around other areas that children may frequent — such as recreation centers, parks, and trails — in Denver, CO using 31 years of crash data (Lee, Booth, Reese-Smith, et al. 2005; Loukaitou-Sideris 2003). In Phase II, based on findings from the first analysis, we then examine parks in more depth relative to schools in six different cities. The goal of these research activities is to identify the location of high concentrations of child pedestrian fatalities, thereby starting the critical conversation of how to best protect children in our cities’ transportation systems.
Data
The study cities were selected because focusing on rapidly growing cities would allow for the examination of current development patterns. While early U.S. cities were designed with pedestrians and streetcars in mind, those developed over the last century were primarily designed to cater to the automobile. Studying these modern auto-centric cities will allow the results to inform current building practices. By having a clearer understanding of the implications of our current community designs, we can build safer places for even the most vulnerable road users. According to Census data, the South was the quickest growing region between 2000 and 2013, while the West was close behind (Cohen, Hatchard, & Wilson 2015). Therefore, cities from these two regions became the focus of this study.
Of the 25 most populous places across the United States, Austin had the largest percentage increase in population from 2000 to 2013, Charlotte had the 2nd largest increase, Denver had the 3rd largest increase, and Dallas had the 10th largest increase (Cohen, Hatchard, & Wilson 2015).
6


Houston had the 2nd largest growth in total population while Los Angeles had the 4th largest total growth (Cohen, Hatchard, & Wilson 2015). These cities with substantial population growth are important to study because they are installing new infrastructure in new and unique land use configurations. The safety outcomes of these new land use configurations are what we hope to explore through this work. Other cities with high growth rates were not used because comprehensive schools, parks, trails, and crash data were not available due to a lack of data collection.
Crash Data
We acquired child pedestrian fatality locations from the National Highway Traffic Safety Administration’s Fatality Analysis Reporting System (FARS) for the years 1982 through 2012. This data was available from 1982 to 2000 in address format, and from 2001 until 2012 with latitude and longitude coordinates. Crashes from 1982 to 2000 were geocoded on either the address-level or, if the data did not contain enough detail, to the street-level. Children were defined as persons under the age of 18. City boundaries were defined by the ‘Places’ shapefile provided by the U.S. Census Bureau through their Topologically Integrated Geographic Encoding and Referencing (TIGER) products.
Exposure Data
Due to consistency issues, finding reliable child pedestrian exposure data in geographically-broad studies has historically been difficult (Wier, Weintraub, Humphreys, et al. 2009). The best option for this particular study, when numerous exposure approaches were assessed, was to use a population-based exposure metric such as that used by DiMaggio and Li (2013) in their safety examination of the Safe Routes to School program. In their study, DiMaggio and Li (2013) used the
7


number of pedestrian crashes in selected census tracts and the number of persons living in those same census tracts to create a rate of crashes per 10,000 population for each of the tracts. While past studies modeled pedestrian exposure using proxy factors such as road network characteristics, land use, and socio-economics, population-based exposure metrics are also common and have proven useful for preliminary and/or geographically-broad work (Wier, Weintraub, Humphreys, et al. 2009; Jacobsen 2003). Given that our study fits both of these conditions, a population-based metric facilitated a consistent child pedestrian exposure metric to study road safety across six U.S. cities. Although there were a number of limitations associated with this population-based exposure metric (which will be detailed in the Conclusion section), analysis on the block group level did allow for finer-grained contexts to be considered.
The exposure variable for the analysis was the number of children living within the analysis zones. This variable was created by pulling the child populations for each block group from the 2010 Census and creating a random point for each child resident. This served as an indicator of the total number of children and a proxy for the relative level of child pedestrian traffic exposure in the study areas. This population-based exposure approach allows for a conservative analysis in terms of parks due to the fact that exposure around schools is typically higher than around parks. Almost all children attend school while not all children use parks. Also, schools get usage from weekday school trips and recreational trips for playgrounds and sports fields on the grounds while parks only experience usage for recreational purposes. Using the same population-based exposure approach for all areas ensured a thoroughly conservative analysis of the risk around parks.
Child-Friendly Destinations Data
We chose child-friendly destinations because past research identified them as public places that children frequent as both recreational and physical activity resources (Lee, Booth, Reese-Smith,
8


et al. 2005; Loukaitou-Sideris 2003). We obtained locational data for the child-friendly destinations from the publicly-available 2015 open data catalogs for the respective cities. The number of schools and parks within the study cities ranged widely (Table 1). Buffers were then created based on the location of the buildings for schools and recreation centers and based on the parcel boundary for parks. This facilitated a more accurate representation of access than, for instance, if parks were based on a single point. The buffers for the trails were drawn adjacent the entire trail; however, this may not be representative of the actual access points. Study areas were designated by quarter-mile buffers around the facilities. This quarter-mile buffer size was chosen because it has been shown to be an appropriate access threshold for children, or the longest distance that children are typically allowed or able to independently walk to their destinations (Wolch, Wilson, & Fehrenbach 2005). Also, the shortest service area with which parks and recreation areas are typically designed is one-quarter of a mile (Cohen, Ashwood, Scott, et al. 2006). For instance, regional parks are normally designed to serve entire cities, while pocket parks may be designed to serve just the surrounding blocks. Because every park has at least a quarter-mile service area, this is an effective buffer size to use.
An ‘Erase’ command was run on the park buffers so that the actual parks were not included in the buffer area. Since there were no fatalities within the parks, erasing the park area did not impact the number of fatalities, but ensured that the exposure variable was not inflated. Thus, the park buffer consisted of only the land one quarter-mile outside of each park.
9


Table 1. Descriptive Statistics for Study Cities
Population Schools Park Area (Hectare) Rec. Centers (count) Trails (km)
Austin 931,840 226 6,742.6 - -
Charlotte 827,121 263 8,016.2 - -
Dallas 1,300,082 221 7,618.6 - -
Denver 682,545 227 7,583.2 30 142
Houston 2,298,628 1,180 10,236.5 - -
Los Angeles 3,971,896 3,689 25,868.4 - -
Methods
In Phase I, we examined child pedestrian fatality concentrations at four destinations that children frequent (i.e. schools, recreation centers, trails, and parks) in Denver, CO. The other study cities were omitted in Phase I because of data limitations. In Phase II, we investigated schools and parks in more detail across six study cities.
Phase I Study Methodology
Upon completion of the data collection and formatting, we initiated spatial analysis by defining the study area buffers and calculating the number of child pedestrian fatalities in those study areas. This was completed through spatial joins in ESRI’s ArcMap (Figure 1).
10


City
Park
C COflAX KV9
— V> TJ O
g *, < O §
a. j. E 11th Ave-
tn
7th
PI
-A)
0 5 Mi es
E
X
Figure 1. Child pedestrian fatalities relative to park buffers in Denver.
After the number of child pedestrian fatalities was derived through spatial queries for each of the zones, the same procedure was run again to find the total number of children living within those zones. This formed the study’s exposure variable, allowing for a rate of fatalities per 10,000 children to be operationalized.
There were no child pedestrian fatalities around recreation centers in Denver (Table 2). This suggests that recreation centers are not a primary problem for child pedestrian safety. Trails had rates similar to schools and parks. However, it is not clear if children use trails in the same manner that they use parks and schools. Access to trails is typically constrained, and the location of the child pedestrian fatalities near trails did not appear to necessarily correlate with trail access points. Trails
11


typically have limited access points while child-friendly destinations such as parks have more permeable access along their borders (Cutts, Darby, Boone, et al. 2009; Krizek, El-Geneidy, & Thompson 2007; Krizek, Barnes, & Thompson 2009; Price, Reed, & Muthukrishnan 2012). Parks were of interest due to the fact that they had the highest fatality rates. We therefore examined parks in Phase II by comparing their fatality rates to the fatality rates around schools, which have been the traditional focus.
Table 2. Child Pedestrian Fatality Rates per 10,000 Children Near Child-Friendly Locations
Schools Rec Centers Trails Parks
Fatalities near child- 3.51 per 0.00 per 3.58 per 3.64 per
friendly locations 10,000 children 10,000 children 10,000 children 10,000 children
Phase II Study Methodology
Based on findings from the preliminary study, a second analysis of child pedestrian safety around parks was warranted. Parks and schools were therefore examined in more detail for six cities: Austin, TX; Charlotte, NC; Dallas, TX; Denver, CO; Flouston, TX; and Los Angeles, CA.
Using the same procedure from the previous analysis, the child populations and the number of child pedestrian fatalities were derived for analysis (Table 3). These variables were considered for areas near schools, areas near parks, areas near schools or parks, and areas near neither schools nor parks. The level of risk was derived for each city, location type, and year within the study. Confidence intervals were then computed.
12


Table 3. Child Pedestrian Fatality Statistics & Child Resident Statistics Near Destinations
Child Pedestrian Fatalities Total Schools (%) Parks (%) Child Population (thousands) Total Schools (%) Parks (%)
Austin 32 5 (15.6%) 15 (46.9%) 173 33 (18.9%) 70 (40.4%)
Charlotte 62 12 (19.4%) 42 (67.7%) 177 23 (12.7%) 66 (37.4%)
Dallas 108 13 (12.0%) 62 (57.4%) 325 57 (17.6%) 122 (37.6%)
Denver 37 17 (45.9%) 29 (78.4%) 129 48 (37.6%) 80 (61.8%)
Flouston 172 39 (22.7%) 45 (26.2%) 564 128 (22.8%) 121 (21.5%)
Los Angeles 417 246 (59.0%) 167 (40.0%) 872 486 (55.8%) 208 (23.9%)
Results
The results suggest that, for all of the study cities, child pedestrian fatality rates are significantly higher in areas near a school or a park than in areas near neither a school nor a park (Table 4). Fatality rates in areas that are near a park or a school are significantly higher than the average citywide rates for five of the six study cities and not significantly different for one of the study cities.
Table 4. Child Pedestrian Fatality Rates per 10,000 Children Living Around Schools or Parks or Neither Schools nor Parks with 95% Confidence Intervals
Citywide Schools or Parks Neither Schools nor Parks % Difference*
Austin 1.85 (1.71, 1.99) 2.14 (1.91,2.37) 1.57 (1.40, 1.74) 36.3%
Charlotte 3.51 (3.28, 3.74) 5.77 (5.36, 6.18) 1.72 (1.56, 1.88) 235.5%
Dallas 3.32 (3.17, 3.47) 4.36 (4.10, 4.62) 2.39 (2.22, 2.56) 82.4%
Denver 2.87 (2.73, 3.01) 3.34 (3.15, 3.53) 1.52 (1.30, 1.74) 119.7%
Houston 3.05 (2.95, 3.15) 3.60 (3.43, 3.77) 2.69 (2.57, 2.81) 33.8%
Los Angeles 4.78 (4.58, 4.98) 5.34 (5.11, 5.57) 3.73 (3.53, 3.93) 43.2%
*Statistically Significant Percent Differences from Schools or Parks to Neither Schools nor Parks are Bold
13


Risk was found to be higher around parks than around schools for all of the study cities (Table 5). Dallas has the largest difference between schools and parks in terms of risk, with child pedestrians being over twice as likely to experience a fatality within close proximity to a park than within close proximity to a school. All of the fatality rates around parks are significantly higher than the rates around schools except for in Denver. Rates around parks are also higher than the average rates citywide for all study cities and are significantly higher for each city except for Austin. Rates around schools are higher than the average citywide rates for just three of the six study cities, and only two of these are significantly higher.
Table 5. Child Pedestrian Fatality Rates per 10,000 Children Living Around Schools and Parks with 95% Confidence Intervals
Schools Parks % Difference*
Austin 1.53 (1.31, 1.75) 2.14 (1.90, 2.38) 40.5%
Charlotte 5.33 (4.79, 5.87) 6.35 (5.87, 6.83) 19.1%
Dallas 2.27 (2.02, 2.52) 5.07 (4.74, 5.40) 123.3%
Denver 3.51 (3.20, 3.82) 3.64 (3.42, 3.86) 3.7%
Houston 3.04 (2.87, 3.21) 3.71 (3.49, 3.93) 22.0%
Los Angeles 5.06 (4.82, 5.30) 8.01 (7.51,8.51) 58.3%
*Statistically Significant Percent Differences from Schools to Parks Are Bold
Conclusion
While past efforts to ensure child pedestrian safety have focused primarily around schools, findings from this work suggest that parks may be an important location to focus on as well. In all of the six study cities, risk for child pedestrian fatalities is higher around parks than around schools, although not all of these differences were statistically significant. The risk around parks has, prior to this research, been largely overlooked. Reasons for higher rates around parks may include unsafe
14


streets along with a general lack of awareness, focus, education, and engagement in terms of the transportation safety issues present.
There are two perspectives through which we may interpret solutions to this problem: transportation and urban design. Taking a transportation approach to the problem would have us lowering vehicle speeds and making drivers aware of child pedestrians through street design changes such as traffic calming, road diets, or pedestrian crossing treatments. A broader urban design approach would focus on the siting of our parks. If we site a park next to a 6-lane roadway with a high design speed, few transportation treatments would be able to help. Within the study cities, it was not uncommon to have a park separated from the community that it serves by roadways with four or six lanes. Some of these roadways have been documented with vehicle speeds greater than 70mph next to the adjacent park (Marshall 2015). Siting parks on slow and narrow local roads within neighborhoods may help alleviate safety issues and thereby induce higher levels of independent walking. The most effective solution to the problem may very well lie in a combination of both of these approaches. We will need to ensure that parks are sited safely within neighborhoods and pedestrian infrastructure is included in a cohesive network to ensure safe access. In addition to these built environment improvements, other approaches — such as child education, driver education, and enforcement methods — may prove effective.
There were several limitations present in this study. Many of the limitations were related to the measurement of child pedestrian exposure. A consistent exposure metric was necessary, which led to a population-based exposure metric. We considered conducting a survey in order to measure exposure, but survey data have been found to significantly underrepresent child pedestrian exposure, and low response rates may introduce self-selection issues (Routledge, Repetto-Wright, & Howarth 1974; Roberts, Keall, & Frith 1994). We also considered observational data, but observational data fails to properly consider potential endogeneity issues between perceived risk and exposure; in other
15


words, a road perceived to be dangerous could be the cause of the low exposure and result in a seemingly good safety record. This would violate the independence assumption of most statistical models (Cho, Rodriguez, & Khattak 2009). Moreover, observational data is difficult to acquire across multiple cities in large enough numbers to ensure sample sizes that reach statistical significance and are representative of actual conditions (Stevenson 1991). For these reasons, a population-based exposure metric was utilized. The exposure metric assumes that individual children will be exposed to traffic dangers at similar rates across the study cities. While this assumption is not necessarily ideal, most children walking to a child-friendly destination such as a school or park would likely live within a quarter-mile of that school or park (Wolch, Wilson, & Fehrenbach 2005). Examining finer geographic levels and exploring different methods of operationalizing child pedestrian exposure will be necessary in order to obtain a better understanding of the issue.
The fact that children of all ages are assumed to act similarly and experience similar risk is another limitation of the exposure metric. In other words, the risk to a 5-year old pedestrian walking independently to a park is most likely higher than the risk to a 13-year old walking independently to a park. However, the 5-year old pedestrian is more likely to be accompanied by a parent, typically alleviating some of the risk. This relationship between age and risk is complex and deserves more attention. Also, examining risk for child pedestrian injuries around parks would provide larger sample sizes and more robust statistical analysis than child pedestrian fatalities. Focusing on finer geographic levels may allow for an injury-specific analysis.
A further limitation was the lack of knowledge pertaining to installation dates of schools and parks. It should also be noted that results may be exclusive to the generally warm climates of the study cities, and generalizability of the findings should not be assumed for other contexts. Other factors that may prove to be of importance include social factors such as population density,
16


poverty, and crime, and built environment factors such as travel lanes, vehicle speeds, and cartway width.
Child pedestrians, being highly vulnerable users of our transportation systems, find themselves at substantial risk as they move about our cities. Ensuring their safety is of the utmost importance. However, in order to ensure that safety, one must understand where safety risks are located. This study has shown that, opposed to traditional beliefs, there are higher concentrations of child pedestrian fatalities around parks than around schools. A shift in the child traffic safety paradigm is now needed to focus treatment efforts around our parks.
17


CHAPTER III
QUANTIFYING SUPPRESSED CHILD PEDESTRIAN AND BICYCLE TRIPS
Introduction
Traditional pedestrian and bicycle safety analyses take a reactive approach to traffic safety by investigating crashes, injuries, or fatalities after they occur. Also examining trips that have been suppressed because of road safety concerns allows for a more proactive safety approach; however, a methodology must first be developed to estimate the number of pedestrian and bicycle trips that are suppressed specifically due to road safety concerns. To accomplish this, we examine child pedestrian and bicycle trips to and from schools in Denver, Colorado. By combining suppression rates derived from a survey examining parental perceptions of safety and the upper limit of trip frequencies derived from a GIS network analysis, we explore how grade level, gender, and adult supervision are related to childhood travel allowance in terms of street-level design characteristics such as posted speed limits, vehicle volumes, presence of sidewalks and bike lanes, and the number of vehicle lanes. We then investigate how widespread these suppressed trips are by quantifying the number of children that are impacted and how their routes would be altered. We finally detect built environment characteristics — such as street-level designs, network configurations, barriers, and destination siting — linked with high levels of suppressed trips. By incorporating this tool into traditional traffic safety analyses, we hope to not only make the places where children are currently walking and bicycling safer, but to improve safety for all places where children want or need to walk and bike.
When traffic safety researchers examine children’s pedestrian and bicycle trips to and from school, they typically analyze crashes, injuries, or fatalities while accounting for the number of child pedestrians and bicyclists that are on the street — also known as exposure. However, this approach to traffic safety is a reactive one, only looking at pedestrians and bicyclists that have deemed the
18


traffic environment safe enough to use, and only looking at the streets where those pedestrians and bicyclists are currently walking or biking. Therefore, these approaches neglect both the pedestrians and bicyclists who want to walk or bike but do not feel safe enough to do so and the places where such trips are being suppressed.
To create a more proactive traffic safety analysis, we also need to account for the pedestrian and bicycle trips that never occurred in the first place because of road safety concerns. How would we measure such suppressed trips? Which personal and built environment characteristics would be associated with road safety-related trip suppression? How many children would be impacted by trip suppression, and how would their routes be altered? While traditional mode choice models output the expected share of different modes, we create a model that instead predicts the percentage of trips that are suppressed due to road safety concerns in order to answer these research questions.
To create this safety perception-based mode choice model, we used results from a survey that we administered to parents of elementary and middle school students in Denver, Colorado, along with linear and logistic regressions to explore how grade level, gender, adult supervision, and street-level design characteristics (e.g. posted speed limits, sidewalks, bike lanes, number of lanes, vehicle volumes) are related to trip suppression rates. We then derived the total number of trips expected under ideal conditions based on a GIS network analysis. Finally, we combined trip suppression rates with the upper limit of trip frequencies to determine the total number of trips being suppressed specifically due to road safety concerns.
Theory
Crashes, injuries, and fatalities — normalized to levels of user exposure — are typically employed to analyze transportation safety of both motorized and non-motorized users (TRB 2001; Waldheim, Wempe, & Fish 2015; FHWA 2006; Zegeer et al. 2010). However, this reactive
19


approach only accounts for individuals who are using the facility and neglects those individuals who have deemed the roadway too unsafe to use in the first place. Accounting for suppressed trips is especially prescient for pedestrian and bicycle safety analyses, where many possible users could be expected to be dissuaded because of road safety concerns (Schneider, Ryznar, & Khattak 2003). Proactive safety approaches account for such suppressed trips, with high levels of suppressed trips signaling road safety issues, regardless of the presence of objective outcomes (Schneider, Ryznar, & Khattak 2003; Nevelsteen et al. 2012). However, past proactive safety analyses have only examined specific areas — an approach that is not generalizable to other roadways — while accounting for a limited number of roadway characteristics. A more holistic method of measuring pedestrian and bicycle trips that have been suppressed because of traffic safety concerns that can be applied more widely is therefore necessary.
Few researchers have ventured to estimate the number of pedestrian and bicycle trips that are suppressed because of traffic safety concerns. Schneider et al. (2003) formulated an early approach by developing a survey to identify areas on the campus of the University of North Carolina at Chapel Hill that are perceived as unsafe — in terms of traffic safety — by pedestrians. By asking individuals to identify the three locations on campus that felt the least safe for pedestrians, the researchers were able to identify areas with poor road safety perceptions and theoretically high trip suppression. However, these results are not transferable because the perceptions were not associated with specific built environment characteristics (i.e. street design, network connectivity, land use, etc.). Also, this approach did not unitize the results (i.e. we may know which site is ranked as the least safe, but we do not know the number of actual trips that are being suppressed); therefore, it is difficult to compare levels of suppression between different sites. Bellemans et al. (2009) similarly used a travel diary that asked respondents to record trips that they had planned but never executed. While Bellemans et al. (2009) did associate the suppressed trips with built
20


environment characteristics, they looked primarily at household and personal schedule factors, rather than the specific impact of road safety. Furthermore, this methodology does not capture suppressed trips that never reached the planning stage.
Cho et al. (2009) related trip suppression rates to built environment factors but only examined macro-level environmental characteristics (land uses and road network density) to estimate suppressed pedestrian and bicyclist trips. While these large-scale factors are certainly important, we wish to examine the impact of individual roadway characteristics because, while it is not easy to change established land use or road network configurations, transportation planners and engineers can more feasibly alter street-level design characteristics.
Nevelsteen et al. (2012) related suppressed pedestrian and bicycle school trips to individual roadway characteristics; however, the researchers only examined two factors (speed limits and the presence of non-motorized facilities). Furthermore, the study took place in the Flemish Region of Belgium, which, with its high levels of active mobility, presents a radically different context than that of the typical American city. All of these past studies used suppressed trip estimates to take a more proactive look at road safety. We will build upon this past work by using parental perceptions of roadway characteristics to determine trip suppression and then apply those results in a citywide analysis.
If we were to estimate suppressed trips based on street-level design characteristics, which characteristics would be important to consider? Past mode choice models found that pedestrian and bicycle facilities, crosswalks and crossing treatments, traffic volumes and speeds, traffic calming features, and crossing guards are important roadway characteristics that predict child mode choice to school (Larsen, Buliung, & Faulkner 2013). Evers et al. (2014) found that, for walking trips to school, parents perceive a lack of sidewalks and the presence of large streets as particularly influential in their decision to allow their child to walk or not. While crossing guards are not
21


included in our analysis because their location may change, and traffic calming features are not included in our analysis because they are captured through the inclusion of traffic volumes and speeds, all other pertinent factors are included in our analysis. It is important to account for as many influential factors as possible so that we may understand how these factors react with each other in terms of suppressed trips (i.e. maybe vehicle speed is an important factor, but it becomes less important if sidewalks are present).
Data
Through this work, we endeavor to design a trip-suppression model based on parental perceptions of roadway characteristics. To do so, data regarding both road safety perceptions and trips are necessary. We garner perceptions through a survey and derive trips from a closest facility GIS analysis using child populations (origins) and school locations (destinations).
We utilized the City and County of Denver to build our safety perception-based model. Denver is the heart of Colorado’s Front Range region with a 2015 population of 649,654 residents (118,886 under 15 years of age) spread out over the city’s 155 square miles. The dense downtown is surrounded by medium-density neighborhoods laid out in predominantly gridded street networks. According to Denver Public Schools (DPS), there are 92,331 children enrolled in DPS’s 207 schools throughout the city.
Parental Perceptions Data
We targeted a survey at parents of children in grades pre-kindergarten through 8th grade to garner parental perceptions of traffic safety. The survey excluded parents of high schoolers because high school students have more independence than elementary and middle school students and would also be more likely to drive themselves or carpool with a friend. The survey was offered
22


exclusively online and was marketed through newsletters, fliers, and social media by Denver Public Schools, parent-teacher organizations, the City and County of Denver, and local advocacy groups. The survey was open for one month from October 5th, 2017 until November 5th, 2017.
Since 36.8% of DPS students identify as Spanish speakers and 55.5% are Hispanic, we provided the survey and promotional materials in both languages. Thus, respondents first answered whether they would like to take the survey in English or Spanish. We next asked parents how many children they would like to complete the survey for. Parents could complete the survey for up to four children simultaneously. Respondents then provided the grade level and gender of each child that was included in the survey response.
The Leuven Travel Behavior of Children to Primary School Survey (Nevelsteen et al. 2012) served as a prototype for the travel behavior questions on the survey. Parents answered whether they would allow their child to either walk or bike along ten different picture-based roadway scenarios on the child’s trip to school (five scenarios for pedestrian questions and five for bicycle questions). While the survey from Nevelsteen et al. (2012) included posted speed limits and presence of active transportation facilities as explanatory factors, Larsen et al. (2013) determined that crossings and vehicle volumes are also important explanatory roadway characteristics. Because crossings at intersections can be complex and difficult to represent in a picture (e.g. varying phasing, signalization, markings, signage, turning movements, etc.), we chose to utilize a combination of variables — including the number of lanes, posted speed limits, and vehicle volumes — as a proxy for crossing risk. While not perfect, these variables are related to the amount of time spent exposed to traffic risk and the degree of that risk. Accordingly, each roadway scenario in our survey had four different characteristics that were identified for the parent: the speed limit of the roadway, the number of lanes, the presence of active transportation facilities (i.e. a sidewalk for walking questions or a bike lane for bicycling questions), and the approximate vehicular volume of the roadway.
23


Two of the factors had three levels (speed limit: 25 mph, 35 mph, or 45 mph; and number of lanes: 2 lanes, 3 lanes, or 4 lanes) while two of the factors had two levels (presence of facility: yes or no; and volume of the roadway: low or high). This resulted in a total of 36 different scenarios for walking and 36 different scenarios for bicycling. We removed non-existent or rare scenarios from each pool (e.g. 45 mph roadways with two lanes and low volumes), resulting in a total of 20 scenarios for each of the walking and bicycling question pools.
Each parent respondent answered five random walking scenarios and five random bicycling scenarios. Each scenario had a picture of a roadway and asked if the parent would let their child/children walk or bike to school along the roadway (Figure 1). The available responses were “No”, “Yes, with trusted adult supervision”, and “Yes, without adult supervision”. Respondents also had the ability to leave open-ended comments after the scenarios were presented. We then asked respondents about the amount of physical activity their child/children get on a weekly basis and gave them the choice to enter an email address for the chance to win one of ten $50 gift cards that were offered as a survey incentive.
Of the 1,298 survey respondents, 924 provided complete responses. These 924 complete parent responses accounted for 1,331 children. There was an appropriate distribution of responses across grade levels and gender, while the majority of surveys were completed for one or two children (Table 1 and Figure 2).
24


8. Would you allow your child to use this roadway on foot to get to school?
25 mph Speed Limit 3 Lanes Sidewalks Low Vehicle Volume
34. Would you allow your child to use this roadway on bike to get to school?
35 mph Speed Limit 2 Lanes Bike Lanes Low Vehicle Volume
Figure 1. Picture-based roadway scenarios from the survey we administered to parents.
25


Children
Table 1. Survey Response Descriptive Statistics
Gender
Male 667
Female 658
Other 4
Number of Children for each Survey
1 431
2 330
3 54
4 10
Minutes you would like your child to be physically active during school
20 9
30 96
40 117
50 45
60 322
60+ 244
Days your child got 60 minutes of physical activity in the last week
0 7
1 19
2-3 148
4-5 314
6+ 351
180
160
140
120
100
80
60
40
20
0
â–  Yes, without adult supervision â–  Yes, with trusted adult supervision â–  No
Figure 2. Number of parental allowance responses from the survey for all roadway scenarios grade level.


Population and Built Environment Data
We collected age-based population data on the block group level from the 2015 American Community Survey via the National Historical Geographic Information System (NHGIS) (Manson et al. 2017). The Denver Open Data Catalog provided the sidewalk network, roadway network, and off-road trail network in polyline shapefile format. The Denver Regional Council of Governments’ (DRCOG) Regional Data Catalog provided traffic volumes and school locations in point shapefile format. We created the bike lane network in polyline shapefile format based on the location of bike lanes per Google Maps, satellite imagery, and Google Street View.
Methods
The goal of this work was to create a trip-suppression model based on parental perceptions of roadway characteristics to determine which personal and street-level design characteristics impact trip suppression. We then integrated trip-suppression rates derived from the survey with the upper limit of trip frequencies — the number of expected trips under ideal conditions as derived from a GIS network analysis — to determine the number of active transport trips to school that are suppressed because of road safety concerns, how routes are altered, and which built environment characteristics suppressed trips are associated with.
To accomplish these goals, we identified the pertinent roadway characteristics for each roadway segment and used survey results to determine the percentage of trips that we would expect to be suppressed on each segment due to traffic safety concerns. We then defined children’s homes as origins and their closest school as the corresponding destination, deriving optimal trip routes through a closest facility network analysis. Then, we used these optimal routes to derive the upper limit of trips that would theoretically utilize each roadway segment. After accounting for the impact of network connectivity on walking levels, we finally combined the trip suppression rates (from the
27


survey) with the upper limit of trips (from the network analysis) for each roadway segment and used this to answer our research questions.
Suppression Rates
With logistic regressions, we created the trip-suppression model derived from the parental perceptions survey to determine which factors are associated with trip suppression. We then examined grade level, gender, adult supervision and street-level design characteristics (e.g. posted speed limit, sidewalks, bike lanes, number of lanes, and vehicle volumes) to establish their relationship with trip suppression rates. For differences in suppression between gender and grade level, we created an independent logistic regression model for each student population subset (i.e. 5th grade males, 7th grade females, etc.). When exploring the impact of adult supervision, we accounted for all students in a single logistic regression model for walking and a single logistic regression for bicycling.
We next determined what percentage of trips would be suppressed due to road safety concerns for different roadway scenarios. We coded all forty roadway scenarios featured in the survey based on their four predictor variables while designating the outcome variable as the percentage of parents that would not allow their children to use the roadway. Since the diversity of actual Denver roads exceeded what we were able to reasonably include in the parental survey, we created a linear regression using the four roadway characteristics as predictor variables and the percentage of disallowance as the outcome variable (Table 2). Using these linear regressions, we then took the street-level design characteristics for each scenario that was not featured in the survey and derived the corresponding suppression rate.
28


Table 2. Linear Regression Coefficients for Trip Disallowance Derived from Survey Results
Walk (R2 = 0.9654) Bike (R2 = 0.9227)
Intercept 0.037 -0.076
Speed (mph) 0.015*** 0.010***
Lanes 0.046* 0.105***
Facilities -0.248*** -0.086***
Volume 0.131*** 0.230***
*p<0.10 ** p<0.05 *** p<0.01
Network Analysis
We then deduced the maximum number of trips that would occur under ideal conditions through a network analysis in GIS. We utilized all public elementary and middle schools in Denver for analysis. DPS does not provide busing for elementary students that live less than a mile from their school. To capture these populations that would be more apt to pursue active modes of transportation, we created a Euclidian distance one-mile buffer (i.e. an as-the-crow-flies buffer instead of a network buffer) around each of the elementary and middle schools and designated this as the study area. The study area included the majority of Denver, except for the far northeast portion of the city comprised of the airport.
After clipping the roadway centerlines to the study area, we removed any limited access highways and merged off-road trails into the layer. All divided roadways were represented by one line instead of two. We avoided edge issues by including roadways in neighboring municipalities that fell within one mile of a DPS school and ensuring that all stray road segments were connected to the larger network. Finally, we cleaned access points around the schools so that students in the model could approach their school from the same side they would in reality.
To account for explanatory roadway characteristics, we utilized speed limits and the number of lanes that were provided from the City and County of Denver’s roadway layer, after checking for
29


and amending any errors. We utilized vehicle volumes provided by DRCOG, with any roadway having more than 1,000 vehicles per day being designated as high volume (Cornell Local Roads Program 2014). We used the City and County of Denver’s sidewalk layer to manually identify roadways with or without a sidewalk. Google Maps, satellite imagery, and Google Street View were utilized to identify roadways with bike lanes. Roadways with no sidewalks/bike lanes were given a value of zero, roadways with a sidewalk/bike lane on one side were given a value of one, and roadways with a sidewalk/bike lane on both sides were given a value of two.
Once the roadway network was complete, we accounted for origins and destinations. Destinations were defined as public DPS elementary and middle schools within the City and County of Denver. Origins were based on child populations from the 2015 American Community Survey. For each Census block group located within the study buffer, we created one random point for each child living in that Census block group. Because the average block group used in the analysis had an area of 248 acres, we created random points only in residential zones so as to realistically represent home origins. While it would be ideal to know exactly how many attending children live within one mile of each school and the home address of those children, that information was not available due to privacy concerns.
There were 112,648 children in Denver and 23,490 in neighboring municipalities included in the analysis. The number of children included in the analysis is higher than the number of children attending DPS schools because the analysis also included children living in Denver that attend private schools or are home schooled. Also, some children living in Denver that were included in the analysis may be attending schools in neighboring municipalities or may be too young to attend school. The largest of the 587 block groups had 3,326 children while 23 block groups had no children. Census block groups with no children consisted of either undeveloped land or land uses other than residential.
30


We then ran a closest facility network analysis using GIS. This takes the origin (child) and finds the shortest route to their respective destination (school). While pedestrians and bicyclists often do not use the shortest available path because of road safety and comfort concerns, we wanted to start with a baseline of how many trips could occur under ideal conditions and then derive the number of suppressed trips based on that value (Krizek, El-Geneidy, & Thompson 2007). Alternatively, if we derived the number of trips being suppressed because of safety concerns from a trip count weighted on traffic safety concerns, this would result in multicollinearity issues.
Children’s routes began at their closest street. Denver is a predominantly flat city located on the western edge of the Great Plains. Therefore, no elevations were factored into the network analysis. One-way streets were not accounted for in the pedestrian analysis but were accounted for in the bicycling analysis. We integrated the impact of roadway network connectivity on active transportation levels into the analysis by accounting for intersection density (Schlossberg et al. 2006). We did not account for crime because of mixed findings in terms of the relationship between objective crime and walking levels, mainly due to more walkable environments attracting different types of crimes (Foster et al. 2014).
The network analysis resulted in 136,138 routes, as all origin/destination pairs were successfully connected. We then derived the number of routes that utilized each roadway segment within the City and County of Denver. This resulted in the total number of children that could be expected to walk or bike on each segment under ideal conditions, assuming that children would be walking or biking within a mile to their closest school.
Suppressed Trips
Now that each roadway segment had a set of roadway characteristics, a corresponding percentage of parents that would not allow their child to walk or bike, and the total number of
31


possible trips under ideal conditions, we multiplied the number of ideal trips by the percentage of disallowance to derive the number of trips that would theoretically be suppressed because of traffic safety concerns. This was done for each roadway segment in the study area. Doing so allowed us to perform spatial analysis to understand where there are high numbers of suppressed trips and which built environment characteristics are associated with those high numbers of suppressed trips.
Results
We first utilized logistic regressions to examine the impact of demographic and roadway characteristics on parental allowance for children’s walking and biking school trips. Then, we employed linear regressions to explore the percentage of trips that could be expected to be suppressed due to traffic safety concerns for a variety of different scenarios. Finally, we integrated these trip suppression rates with the upper limit of modal frequencies derived through GIS network analyses to discover areas of Denver that have high levels of trip suppression and found associated built environment characteristics.
Trip Suppression Factors
Typically, the roadway characteristics utilized in our models (i.e. posted speed limits, number of lanes, presence of active transport facilities, and vehicular volumes) would be collinear. However, because a wide range of roadway scenarios were chosen for the survey instrument, multicollinearity was avoided. A variance inflation factor (VIF) threshold of 1.46 for the variables signaled low multicollinearity. A VIF of 5.0 or above is typically indicative of multicollinearity issues (Vatcheva et al. 2016).
When simply looking at parental allowance in a binary fashion (i.e. allowance both with and without adult supervision are categorized together) for all children captured by the survey, gender
32


was not significantly related to parental allowance for either walking or bicycling, but the variable did strengthen the model based on a lower AIC (Table 3). Other than the number of siblings and the days of physical activity in the bicycling model, all other explanatory variables were significant. The presence of sidewalks was by far the strongest predictor for the walking model, with a 14.079 odds ratio being interpreted as parents being approximately 14 times more likely to allow their child to walk on a roadway with sidewalks present than a roadway without sidewalks, with all other factors held constant. This coincides with past findings examining parental perceptions of children’s walking trips to school (Evers et al. 2014). Vehicle volumes were the strongest predictor for the bicycling model, followed closely by the presence of bicycle facilities (bicycle lanes). Parents are not as concerned with vehicle speed or volumes when considering walking, as long as children have a sidewalk outside of the traffic lanes. We utilized the grade variable by individual grade level, so a 1.078 odds ratio means that a 5th grade student would be 1.078 times more likely to be allowed to walk than a 4th grade student, with all other factors held constant.
Table 3. Parental Allowance Logistic Regression Odds Ratios
Walk (R2 = 0.338) Bike (R2 = 0.223)
Grade 1.078*** 1 178***
Gender (male—0) 0.940 0.896*
Speed (10mph increments) 0.402*** 0.581***
Lanes 0.751*** 0.576***
Facilities 14.079*** 2.502***
Volume (low/high) 0.445*** 0.297***
Number of Siblings 0.808*** 1.045
Days of Physical 1 143*** 1.052
Activity
* p<0.10
** p<0.05
*** p<0.01
We then took a more thorough look at these results in terms of grade level and gender. Results suggest that sidewalks remain the strongest predictor for walking suppression across all
33


grade levels and genders (Figure 3). Posted speeds are typically the second strongest predictor for walking, followed by vehicle volumes. In terms of parental allowance for bicycling, we found vehicle volumes to be the strongest predictor for younger children. However, for some higher-grade levels, bike lanes become the most important street-level design characteristic. For bicycling allowance, posted speed limits and the number of lanes are the least important factors. There are no significant differences in terms of gender, as the results above suggest. There are notable odds ratio increases in 3rd grade, 4th grade, and 5th grade, possibly signaling this as a time when parents feel that children may walk and bike.
34


Odds Ratio Odds Ratio
35
Pre-K K 1st 2nd 3rd 4th 5 th 6 th 7 th 8 th
â–  Speed â–  Lanes â–  Sidewalk â–  Volume
6
Pre-K K 1st 2nd 3rd 4th 5 th 6 th 7tli 8tli
â–  Speed â– Lanes â–  Bike Lanes â–  Volume
Figure 3. Odds ratios of street-level design factors in relation to suppression by grade level and gender for walking (above) and bicycling (below) (hollow bars are not significant at 90%).
35


When we further parse the results by examining how factors are correlated in terms of whether parents reported trusted adult supervision as required, findings suggest that both types of allowance follow similar patterns (Table 4). Namely, the presence of sidewalks remains the most important factor in terms of walking allowance, followed by vehicle volumes and child grade level. In terms of biking allowance, vehicle volume remains the most important factor, followed by the grade level of the child. These results suggest that those factors identified as important in the models not accounting for supervision are also important (although less so) when supervision is accounted for, with grade level becoming a stronger predictor, as we would expect.
Table 4. Parental Allowance Logistic Regression Odds Ratios
Walk (R2 = 0.452) Bike (R2 = 0.374)
Grade 1.814*** 1 697***
Gender (male—O) 0.823** 0.880
Speed (10mph increments) 0.642*** 0.629***
Lanes 0.863** 0.795***
Facilities 4 1 "77* ** *** * * 1.055
Volume (low/high) 0.533*** 0.447***
*p<0.10
** p<0.05
*** p<0.01
Trip Suppression Rates
Once we understand which demographic and street-level design factors are related to trip suppression, we can begin to explore how many children in Denver are impacted by trip suppression and how that trip suppression impacts route choice. Since we did not detect any non-linear relationships, linear regression was used in this analysis. We created a linear regression (Table 2) with the results from the survey to derive a mode choice model that outputs the rate of trip suppression due to traffic safety concerns for a variety of different roadway scenarios (Table 5). For instance, while parents reported that only 11.7% of walking trips would be suppressed on a low-
36


volume, 25 mph, 3-lane road with sidewalks, parents reported that the same road without sidewalks would see 59.4% of walking trips suppressed (Figure 4). Parents reported that for a high-volume, 25 mph, 3-lane road with sidewalks, 16.4% of walking trips would not be allowed. Here we can see that the change in facilities had a much larger impact on allowance than the change in vehicle volume. Rate suppression for bicycling was generally higher than for walking, as parents were less willing to let their children bicycle to school. Presence of facilities can be seen to have a large impact for both walking and bicycling. There is a substantial increase in trip suppression rates when going from 25 mph to 35 mph for bicycling, while there is a similarly substantial increase when going from 35 mph to 45 mph for walking. Also, vehicle volumes have a greater impact on trip suppression for bicycling than for walking.
Table 5. Percentage of Trips Suppressed Based on Survey and Linear Regression (Variables Held at 3 Lanes, Presence of Facilities, and Low Volume; Survey Values in Bold)
Walk 25 mph Bike Walk 35 mph Bike Walk 45 mph Bike
Lanes
2 5.1% 23.8% 14.5% 33.1% 31.6% 41.2%
3 11.7% 28.7% 25.6% 42.0% 36.2% 51.7%
4 10.5% 42.2% 20.1% 52.2% 43.3% 67.9%
Facilities
No 59.4% 52.3% 65.8% 63.1% 85.7% 68.9%
Yes 11.7% 28.7% 25.6% 42.0% 36.2% 51.7%
Volume
Low 11.7% 28.7% 25.6% 42.0% 36.2% 51.7%
High 16.4% 49.8% 32.4% 64.0% 49.2% 74.7%
37


Figure 4
• 25 mph
• 3 Lanes
• Sidewalks
• Low Volume
11.7% of child walking trips will not be allowed according to parent responses.
• 25 mph
• 3 Lanes
• No Sidewalks
• Low Volume
59.4% of child walking trips will not be allowed according to parent responses.
• 25 mph
• 3 Lanes
• Sidewalks
• High Volume
16.4% of child walking trips will not be allowed according to parent responses.
38


• 25 mph
• 3 Lanes
• Bike Lanes
• Low Volume
28.8% of child biking trips will not be allowed according to parent responses.
• 25 mph
• 3 Lanes
• No Bike Lanes
• Low Volume
52.3% of child biking trips will not be allowed according to parent responses.
• 25 mph
• 3 Lanes
• Bike Lanes
• High Volume
49.8% of child biking trips will not be allowed according to parent responses.
Figure 4. Examples of parent-reported trip disallowance for different walking and biking roadway scenarios.
39


We next identified to what extent children are encountering these roads with high levels of disallowance. More children in Denver encounter roads that parents have deemed unsafe for bicycling than encounter roads that parents have deemed unsafe for walking. Approximately 2.3% of children in Denver would encounter a road with 75% disallowance or greater for walking (a road that is perceived as particularly unsafe), assuming that they take the shortest route to school (Table 6). However, 31.8% of children in Denver would encounter a road with 75% disallowance or greater for bicycling. For roads with 50% disallowance (still perceived as relatively unsafe), more than half of children in Denver would encounter those roads for a bicycling trip and 12% would encounter them for walking trips. This converts to tens of thousands of trips each day, showing that the issue is widespread throughout Denver.
Table 6. Percentage of Children Encountering Roads with Varying Disallowance Rates
25% Disallowance 50% Disallowance 75% Disallowance Walk 40.5% 12.2% 2.3%
Bike 64.9% 61.4% 31.8%
Children that encounter roads that are perceived as unsafe for walking — specifically those roads which more than half of parents would not allow their child to walk on — are primarily found in two different areas: an area with sidewalk gaps near the border of the Montclair and East Colfax neighborhoods of Denver (the top-right concentration) and an area near the border of the Mar Lee and Ruby Hill neighborhoods (the bottom-left concentration) (Figure 5). This first neighborhood has high numbers of children that encounter roads perceived as unsafe because of gaps in the existing sidewalk network while the second neighborhood has high concentrations of children near Federal Boulevard and Florida Avenue, roads that are perceived as unsafe. Children that encounter roads that are perceived as unsafe for bicycling are similarly found near Federal Boulevard, but also
40


in the Montbello neighborhood, which contains high concentrations of children, a curvilinear roadway network, and a lack of bicycle facilities. It is apparent that, in general, there are more children that encounter roads that parents would not allow them to bicycle on than children that encounter roads that parents would not allow them to walk on. The neighborhoods that had the highest number of children who encounter roads perceived as unsafe have median household incomes that are 6.2%, 15.1%, and 46.7% below average for Denver.
41


Figure 5. Densities of children with negatively impacted routes for walking (top) and bicycling (below) (red represents higher densities of impacted routes).
However, if a grid network is in place that presents pedestrians and bicyclists with different route options, children may be able to simply avoid these roads that are perceived as unsafe by using parallel streets. How much further would children’s trips be if these roads perceived as unsafe were
42


not utilized? To answer this inquiry, we reran the network analysis after the roads that are perceived as unsafe — particularly those with greater than 50% disallowance — were weighted so that they would be discouraged according to parental traffic safety concerns. We then compared the trip lengths under ideal conditions to those when roads perceived as unsafe were accordingly discouraged.
For walking trips, the average trip length across the city increased from 2,728 feet under optimal conditions to 2,937 feet once roads perceived as unsafe were discouraged. For bicycling trips, the average trip length increased from 2,728 feet under optimal conditions to 3,763 feet once roads perceived as unsafe were discouraged. Citywide, about 4,274 children were pushed out of a '% mile walkshed when road safety perceptions were accounted for, while 23,429 children were pushed out of a V2 mile bikeshed. The greatest increases in distance were concentrated near Interstate 25, the South Platte River, and Sheridan Boulevard. These neighborhoods have curvilinear tributary roadway networks or limited route options because of barriers, resulting in large increases in trip distance (upwards of an additional 5,228 feet to avoid roads perceived as unsafe in the curvilinear tributary neighborhood). Areas with grid networks that saw large percentages of roads perceived as unsafe did not have similarly large increases in trip distance because of the ability for pedestrians and bicyclists to select alternate routes with little additional distance.
Number of Suppressed Trips
We then integrated the results from the mode choice model with the number of total possible trips to derive the number of trips that are suppressed due to road safety concerns for roadways in Denver and identified which roadways have the most suppressed trips (Table 7). We focused on roads with greater than 25% disallowance because these roads are perceived as unsafe and are most likely in need of amendment. While some roads with less than 25% disallowance had
43


high numbers of suppressed trips, it was only because of high levels of ideal exposure, not because of a lack of perceived safety. Therefore, these roads were not considered. The high number of low-speed local roads in Denver’s grid networks resulted in a relatively low mean of suppressed trips per road segment. We utilize spatial analysis techniques in the Discussion to explore where these road segments with high numbers of suppressed trips are located.
Table 7. Number of Suppressed Trips per Road Segment
Min Max Mean SD
Walk 0 272 1.60 5.90
Bike 0 528 7.11 20.59
Discussion
This section presents a spatially-oriented, rather than statistical, analysis of where roadways are located with high numbers of suppressed trips. Walking trip suppression and bicycling trip suppression displayed similar spatial patterns within the study area. We found high numbers of suppressed trips primarily either near a connection through a barrier in the roadway network or near a school. The barrier connections were at impediments that have limited pathways over or under them (e.g. limited access highways or bodies of water), for which high rates of trips would optimally funnel through the few available connections. Because these barrier connections primarily serve vehicles, they are usually wide, high-speed roadways. Therefore, while such connections are vital to both motorists and non-motorists, the connections are often built to accommodate vehicles and present non-motorists with an option that is many times perceived as unsafe. Examples of these barrier connections include Lincoln Street where it goes under 1-70 (25 mph, two sidewalks, no bike lanes, high volume, two lanes; 14.3% walking disallowance & 61.4% biking disallowance; 14 suppressed walking trips & 41 suppressed biking trips) and Dunkirk Street as it goes across First
44


Creek (30 mph, two sidewalks, no bike lanes, high volume, four lanes; 31.1% walking disallowance & 87.4% biking disallowance; 99 suppressed walking trips & 286 suppressed biking trips) (Figure 6) It is also important to note that these numbers of suppressed trips are based on each individual to/from trip to school. If we multiply the number of suppressed trips by 180 annual school days and by two for the departing and returning trips, 14 suppressed trips is equivalent to 5,040 annual suppressed trips. We found high numbers of suppressed trips at barrier connections to be rare because children often have schools within their own neighborhoods that are closer than those on the other side of the barrier.
Figure 6. Examples of barrier connections with high trip suppression (Central Park Boulevard over Prairie Meadows Drive on the left; Dunkirk Street over First Creek on the right).
45


There were also high rates of trip suppression on roadways near schools. These roadways can be broken into two categories: those in tributary curvilinear loop networks and those in grid networks. While the majority of Denver consists of gridded networks, three of the top six areas for suppressed walking trips and two of the top six areas for suppressed biking trips near schools are found in the Montbello neighborhood of northeast Denver. Montbello has a predominately tributary curvilinear loop roadway network, as the bulk of its development occurred in the late 1960s and early 1970s. These networks consist of curved local streets that channel onto collector or arterial streets. While pedestrians and bicyclists in grid networks have the option to use a variety of roadways to get to their destination, non-motorized users in these curvilinear loop networks typically must channel to a main road that has been prioritized for vehicles.
The four most extreme cases of suppressed trips in the tributary curvilinear network in Montbello are Andrews Drive (46.3% walking disallowance & 82.4% biking disallowance; 51 suppressed walking trips & 374 suppressed biking trips), Maxwell Place (54.8% walking disallowance & 87.4% biking disallowance; 135 suppressed walking trips & 215 suppressed biking trips), Gateway Avenue (46.3% walking disallowance & 82.4% biking disallowance; 75 suppressed walking trips & 133 suppressed biking trips), and 46th Avenue (46.3% walking disallowance & 82.4% biking disallowance; 272 suppressed walking trips & 484 suppressed biking trips) (Figure 7). While the roads in these neighborhoods may not be perceived to be as dangerous as some in central Denver (the roadways have two sidewalks, do not have bike lanes, are signed at 25 mph or 30 mph, have four lanes, and are high volume), the high trip suppression rates in these northeast neighborhoods are being driven by the fact that trips are concentrated on these main roads because of the street network configuration.
46


Figure 7. Examples of high trip suppression around a school in a curvilinear loop network (Blue dots are schools; image is of Andrews Drive).
Relatively few areas with high trip suppression were found in the grid network, which is predominant across Denver. High trip suppression roadways that were found in the grid network
47


typically occurred when a school was placed on or near a major road. This scenario occurred outside of the grid network as well. Examples include S. Sheridan Boulevard (71.8% walking disallowance & 97.4% biking disallowance; 127 suppressed walking trips & 421 suppressed biking trips) and S. Monaco Parkway (54.8% walking disallowance & 87.4% biking disallowance; 80 suppressed walking trips & 314 suppressed biking trips) (Figure 8). While pedestrians and bicyclists in a grid network typically have options in regard to which roads they utilize, siting a school directly on an arterial can force them to use an unsafe road, thereby dissuading walking or biking trips.
Figure 8. Examples of high trip suppression around schools near major roads (Sheridan on the left; Monaco on the right).
48


Conclusion
By combining trip suppression rates derived from a perception survey with the upper limit of trip frequencies from a GIS network analysis, our tool allows us to identify areas where trips are suppressed because of road safety concerns. Although identified areas are not technically objectively unsafe, only perceived as unsafe, identifying such areas will hopefully aid in the identification of road safety issues before they occur.
We find that, for all children both with and without adult supervision, sidewalks are the strongest predictor in terms of walking allowance and vehicle volumes are the strongest predictor in terms of biking allowance. Parents may be more concerned with the presence of vehicles for biking rather than walking because their child would likely be biking in the street. Gender of the child has a weak relationship with allowance. When we parse the results by age, sidewalks consistently remain the most important factor for walking allowance, but bike lanes are found to become more important for higher grade levels in terms of biking allowance. When we look at the role adult supervision plays, we see sidewalks (for walking) and vehicle volumes (for biking) remaining the most important factors (although losing importance), while grade level becomes a stronger predictor.
When looking at the pervasiveness of these issues, we find that over 61% of children encounter a road perceived as unsafe (50% or greater disallowance) for biking and over 12% encounter a similar road for walking. This indicates that the problem is prevalent across Denver. While many children’s routes were not substantially altered because of the trip suppression, some neighborhoods experienced large increases in the distance that must be travelled, with children in some neighborhoods needing to add an additional mile onto their route to avoid such roads.
Areas with high numbers of suppressed trips were heavily concentrated around schools in parts of the city with curvilinear loop and tributary networks. Grid networks seem to help to
49


alleviate high numbers of suppressed trips, provided that the school is not sited on or very near a major road. There are also high numbers of suppressed trips present at barrier connections. The segments of high trip suppression are typically not great in length, meaning that it may only take one or two blocks of conditions perceived as unsafe to dissuade a pedestrian or bicycle trip from occurring.
Primary limitations of the work are focused on the origin and destination location of the school trips employed in the model. Because of privacy issues, we could not account for the actual trip of each student, and instead assumed that children would be most likely to attend their closest school. This is an assumption we know to be imperfect because of Colorado’s open enrollment policy. While this may have impacted the implementation of the tool, the theoretical methodology developed is still sound. Future work could improve upon the current approach by examining areas with clearer origins and destinations. Furthermore, the number of parent respondents that took the survey in Spanish was low (3.6%) relative to the number of reported Spanish speaking students in DPS (36.8%). While we believe that some Spanish speakers took the survey in English, specifically concentrating on these populations in future efforts may result in more representative outcomes.
The decision to allow a child to walk or bike is typically influenced by a combination of street design variables. Future work may explore different explanatory variables utilized for the model. While we used more explanatory roadway-characteristic variables than past studies (Cho, Rodriguez, & Khattak 2009; Schneider, Ryznar, & Khattak 2003; Nevelsteen et al. 2012), roadways are complex, and more variables may lend further strength to our models. For the sake of this work, we used a combination of variables — including the number of lanes, posted speed limits, and vehicle volumes — as an indicator of the exposure to risk a child would encounter when crossing a road. Fiowever, in terms of crossings, there are other factors (e.g. signalization, phasing, crosswalks, medians, etc.) that would also be important to account for. Sidewalk conditions also vary, and these
50


varying conditions would be important to include in future models. Specifically for Denver, many neighborhoods have two-foot wide sidewalks for which parents may have varying perceptions of road safety relative to current five-foot wide sidewalk standards. Furthermore, examining the impact of actual vehicle speeds instead of simply assuming signed speed limits are reflective of actual conditions may improve future models. The vehicle volume factor could account for whether peak hour crests occur during times when children would be expected to be walking or biking on each road. Finally, future work may also account for school zones, which could improve traffic safety perceptions near schools. School zones in Denver are typically signed at 20 mph, but major roads use speed limits of up to 40 mph in school zones, highlighting the bias towards motor vehicles and against non-motorized trips.
In terms of macro-scale perspectives of the work, future analysis could account for varying land uses. While we were only concerned with trips to and from school, and it was therefore appropriate to only account for this one specific land use (Ewing, Schroeer, & Greene 2004), more holistic future examinations would be wise to account for the presence of other land uses and destinations. The impact of crime on levels of walking and biking could also be accounted for but would necessitate a thorough examination of the types of crime occurring relative to the land uses throughout the study area. For example, while violent crimes in residential neighborhoods may dissuade walking and biking, high rates of shoplifting may signal a strong commercial area that may have high levels of walking and biking (Foster et al. 2014). Such a thorough analysis of crime in Denver was outside the purview of this work. This aspect of the work also hints at equity issues, namely that lower-income or minority neighborhoods may have to more frequently deal with both crime and traffic safety issues, more so than their more affluent counterparts. The neighborhoods that had the highest number of children who encounter roads perceived as unsafe were found to have median household incomes that are below average for Denver.
51


Future work could see integration of this trip suppression tool into a proactive safety analysis by similarly identifying areas with high trip suppression. Results from such a proactive analysis could be compared to a traditional reactive safety analysis. Outcomes from the two analyses would be expected to vary. Hopefully, we would be able to identify areas with high rates of trip suppression but low objective crashes or injuries. These would be areas with road safety issues that dissuade non-motorized users enough to preclude objective outcomes, and therefore have thus far been neglected. It is recommended that planners and engineers utilize such analysis approaches and also deploy recommendations from this work, namely employing grid networks, siting schools — and other locations that children may be expected to frequent — on more minor roads, and ensuring that there are pedestrian and bicycle facilities present where there are vital connections across barriers.
Pedestrian and bicycle trips that have been suppressed because of traffic safety concerns can be an important indicator of road safety in our transportation systems. The tool developed in this paper allows for the identification of roadways with high levels of suppressed trips in terms of street-level design characteristics. This approach allows for the methodology to be applied widely, enabling utilization by academics and practitioners alike. Through the application of this tool in Denver, we identified important personal and design characteristics that act as predictors of trip suppression, as well as the importance of grid networks, barrier connections, and destination siting. By identifying these areas with high numbers of suppressed trips, and by enabling others to do the same, we have facilitated the proactive identification of traffic safety issues on our roadways.
52


CHAPTER IV
SUPPRESSED CHILD PEDESTRIAN AND BICYCLE TRIPS AS AN INDICATOR OF SAFETY: ADOPTING A PROACTIVE SAFETY APPROACH IN DENVER
Introduction
Traditional pedestrian and bicyclist safety analyses typically examine crashes, injuries, or fatalities. However, this reactive approach only accounts for the places where people are currently walking or biking and those who are doing so. Would a proactive approach — examining areas where pedestrian and bicyclist activity is being suppressed because of safety concerns — illuminate other previously neglected safety issues?
The goal of this work is to compare results from reactive and proactive pedestrian and bicyclist safety analyses in Denver, Colorado. To accomplish this, we focus on child pedestrians and bicyclists because of the structured characteristics of their trips to school. We utilize conventional reactive analyses from Denver Public Works and the Denver Regional Council of Governments as well as our own cluster analysis of crashes. We then complete a proactive safety approach based on the number of trips that are suppressed due to traffic safety concerns. A parental perception survey forms the basis of the mode choice model we created to perform the proactive safety analysis.
Findings suggest that reactive approaches identify downtown Denver and major corridors as unsafe, while the proactive analysis identifies neighborhoods in west, east, and northeast Denver. Due to an absence of crashes, the majority of these areas would not normally be considered unsafe for pedestrians and bicyclists based on the conventional approach. However, the fact that they are perceived as unsafe may be limiting usage and thereby limiting the number of crashes. In order to improve safety both where children are walking and bicycling — as well as where they want or need to walk or bike — traditional analyses would benefit from augmentation by such a proactive safety approach.
53


Researchers traditionally analyze crashes, injuries, or fatalities when examining traffic safety of walking and bicycling trips. However, the only people that are accounted for in this reactive approach to safety are those who are already walking or biking — the people who have decided that those activities are safe enough to pursue. What about the people who — because of traffic safety concerns — have decided to not walk or bike in the first place? Furthermore, the only locations that are identified in such reactive approaches are the locations that have pedestrians and bicyclists present. What about the places where people have decided not to walk or bike? How would we proactively identify these places as unsafe before a crash occurs or even before any walking or biking occurs?
Past research has investigated how perceptions of traffic safety impact the choice to walk or bike. We propose building upon this theory by quantifying the number of walking and bicycling trips that are being suppressed due to traffic safety concerns and using this as an indicator of traffic safety risk. If high levels of pedestrian and bicycle trips are being suppressed because of traffic safety concerns, this suggests that there are traffic safety issues present, regardless of whether crashes are occurring. We then compare results from such a proactive safety analysis to results from traditional reactive analyses.
To accomplish these objectives, we reactively and proactively analyze the safety of walking and biking trips of children to and from school in Denver, Colorado. Children are some of the most vulnerable road users, who depend heavily on walking and biking. Their trips to school are highly structured, making this an ideal group to study. For the reactive approach, we use methodologies and results from existing safety reports focused on pedestrians and bicyclists of all ages to create our own crash cluster analysis specific to child pedestrians and bicyclists. To derive the number of child trips to school being suppressed because of safety concerns for the proactive approach, we combine trip suppression rates from a parental perceptions survey with shortest path
54


distances to derive safety perception-weighted distances. Finally, we compare results from this proactive analysis of trips suppressed due to safety concerns to results from the traditional reactive analyses of crashes. If our goals include promoting walking and biking activity instead of just reducing crashes, such a proactive safety analysis may provide an important perspective on pedestrian and bicyclist safety.
Theory
Traditional safety analyses rely on crashes, injuries, or fatalities to identify unsafe roadways (Transportation Research Board 2001; Waldheim, Wemple, and Fish 2015). This approach is utilized for safety studies of vehicles as well as pedestrians and bicyclists (Federal Highway Administration 2006; Zegeer, Nabors, Gelinne, Lefler, and Bushell 2010). Ideally, these crash-based safety analyses account for exposure —a pedestrian or bicyclist’s proximity to potentially harmful situations involving motor vehicles — in the form of distance or time traveled, user counts, times crossing a street, or the product of pedestrian or bicyclist and vehicle volumes (Molino et al. 2012). However, the lack of reliable pedestrian and bicyclist exposure data makes such comparisons difficult (Turner et al. 2017). In the absence of appropriate exposure data, cluster analyses of crashes are a common approach used to identify areas of safety concern (Blackburn et al. 2017).
The four most recent pedestrian and bicyclist safety analyses for Denver, Colorado that were completed by local and regional transportation agencies consisted of traditional crash-based analyses that did not account for exposure (Denver Public Works 2016; Denver Public Works 2017;
DRCOG 2012; DRCOG 2017). While such a focus on crashes can allow for the successful identification and reduction of those crashes, it is by nature a reactive approach that requires that a crash occur before any safety issues can be identified. Because a pedestrian or bicyclist crash cannot
55


take place unless there is a pedestrian or bicyclist present, this conventional approach only accounts for people who are currently walking or biking and the places where they are doing so.
A proactive approach to safety is one that identifies areas that are likely to experience crashes before those crashes occur. It could also be one that identifies areas where safety concerns have caused low levels of pedestrian and bicyclist activity, and therefore effectively hides safety issues from the objective eye. For instance, proactive approaches can be effective when exposure levels — and therefore crash levels — are so low that traditional indicators do not provide an accurate representation of risk (Cho, Rodriguez, and Khattak 2009; Schneider, Ryznar, and Khattak 2003). Perceptions of safety have been shown to correlate with exposure levels and may therefore proactively indicate safety issues (Noland 1995; Pucher, Dill, and Handy 2010). In other words, while a road perceived as unsafe may suppress walking and biking trips and therefore reduce or preclude pedestrian and bicycle crashes, these same negative safety perceptions can be used as an indicator that low levels of exposure have hidden safety issues.
Past researchers have taken a number of approaches when using perceptions of safety to proactively identify pedestrian and bicyclist safety issues. An early attempt surveyed pedestrians and drivers on the campus of the University of North Carolina at Chapel Hill, asking them to identify locations that posed safety issues to pedestrians (Schneider, Ryznar, and Khattak 2003). When the researchers compared these subjective perceptions to objective outcomes, it became apparent that there were areas perceived as unsafe that had no crashes occurring. Researchers determined that, while these areas were otherwise desirable to pedestrians, the perception of these areas as ‘accidents waiting to happen’ was reducing levels of exposure. While these results provide a theoretical foundation for our work, the method of identifying unique hotspots is not scalable. In other words, because the perceptions were not tied to specific characteristics of the built environment, the survey
56


would need to be re-administered for every new area that may be studied in the future. We seek a methodology that can be generalized and applied to other areas.
Cho et al. (2009) improved upon this scalability issue by creating a risk estimate based on built environment characteristics such as land use mix and street connectivity. They found that increased perceptions of risk for pedestrians and bicyclists reduced crash rates because of decreased usage. However, the methods did not account for street-level risk factors such as roadway width, vehicle speeds, and vehicle volumes, which could have an impact on behavior. This approach and its results have limited implications for built environment improvements as road network connectivity and land use mixes are not as easily addressed as vehicle volumes, vehicle speeds, or sidewalk gaps.
Nevelsteen et al. (2012) built upon these trip suppression studies by examining the relationship between perceptions of street-level risk factors and pedestrian and bicyclist trip allowance for children traveling to school. However, the researchers only examined two factors: the presence of pedestrian and bicycle facilities and vehicle speeds. Furthermore, the study was performed in a Belgian context that reports 40% of 11 and 12-year olds cycling to school — a transportation culture that is vastly different from that of much of the world.
Importantly, all of the studies proactively examining safety perceptions showed that perceptions of safety and trip suppression differ from objective safety outcomes such as crashes. Specifically, Cho et al. (2009) and Schneider et al. (2003) found that perceptions of unsafe conditions lowered use and therefore exposure, which improved objective safety outcomes, thereby hiding safety issues. However, these past proactive analyses were either focused on small areas, did not include all applicable roadway variables, or were in unfamiliar contexts. We therefore create a proactive model that quantifies trip suppression for an entire American city based on all applicable roadway risk factors and then compare results to those from more conventional reactive crash-based
57


analyses. We hypothesize that the reactive and proactive analyses will illuminate different safety concerns and complement one another.
In order to accomplish this, we look specifically at children’s trips to school. While children are an important focus because they are some of the most vulnerable road users and they rely on walking and biking to travel independently, they also have relatively structured trips that allow for more accurate modeling. This provides us with a foundation for our approach and allows us to avoid needing to model every pedestrian and bicycle trip throughout a city for every age and trip purpose.
Data
The goal of this paper is to compare reactive crash-based pedestrian and bicyclist safety analyses (using existing reports for pedestrians and bicyclists of all ages as a guide to form a crash cluster analysis specific to children) to a proactive analysis (based on where child trips are being suppressed because of safety concerns) that we develop for Denver, Colorado. Data for the reactive and proactive safety analyses are detailed below.
We utilized the City and County of Denver for our analysis. Denver has a population of 663,303 according to the 2016 American Community Survey. Because Denver has a variety of pedestrian and bicycle-specific safety reports along with detailed crash data available and high enough population, pedestrian, and bicyclist levels to get significant samples, it was an ideal location for of our study.
While there are suitable levels of walking and biking for this study, there are also pedestrian and bicyclist safety issues present. According to Denver Public Works (DPW) reports and DRCOG data, there were 1,508 pedestrians and 1,083 bicyclists hit by motor vehicles in Denver between 2011 and 2014 (Denver Public Works 2017). These collisions resulted in injuries for 918 pedestrians and
58


791 bicyclists, with 72 pedestrians and seven bicyclists killed (Denver Public Works 2016; Denver Public Works 2017). These are pressing safety issues that should be addressed.
Reactive Analyses
We utilized existing pedestrian and bicyclist safety reports along with our own crash cluster analysis for the reactive crash-based portion of the study. The existing safety reports focus broadly on pedestrians and bicyclists of all ages, while the focus of this research was on child pedestrians and bicyclists. Therefore, we supplemented these existing safety reports with results from our own child pedestrian and bicyclist crash cluster analysis.
Existing Safety Reports
There were two pedestrian and bicyclist safety reports published by Denver Public Works and two pedestrian and bicyclist safety reports published by the Denver Regional Council of Governments (DRCOG) — the regional Metropolitan Planning Organization — that analyzed pedestrian and bicycle safety and were considered in this study (Denver Public Works 2016; Denver Public Works 2017; DRCOG 2012; DRCOG 2017). All four of the pedestrian and bicyclist safety reports utilized conventional reactive crash-based analysis methodologies. None of these agency-produced reports accounted for levels of exposure.
The 2016 Denver Public Works report considered bicyclist crashes that occurred between 2008 and 2012, while the 2017 Denver Public Works report considered pedestrian crashes that occurred between 2011 and 2015. Crashes were defined as motor vehicle collisions that involved a bicyclist or pedestrian and were obtained from the Denver Police Department. Crashes included only those that were reported by the police. The reports noted that there were likely unreported pedestrian and bicycle crashes — due to the involved parties or the police not understanding the need
59


to report or not wanting to report their crashes — that were not considered in the analysis (Denver Public Works 2016; Denver Public Works 2017). Due to a lack of pedestrian and bicyclist count data, exposure was also not accounted for in these studies.
The DRCOG reports similarly examined pedestrian and bicyclist crashes for all ages without accounting for exposure (DRCOG 2012). The DRCOG report identified pedestrian and bicyclist crash hotspots for the entire Denver region for the years between 2003 and 2007. A pedestrian crash was defined as a collision between a motor vehicle and a person on foot (including persons in wheelchairs, skateboards, rollerblades, or other small personal conveyance devices), while a bicyclist crash was defined as a collision between a motor vehicle and a non-motorized pedal cyclist (DRCOG 2012). Crashes were pulled from the Colorado Department of Transportation’s crash database. DRCOG also completed an updated pedestrian and bicyclist safety report in 2017 that similarly utilized a reactive crash-based approach, but because it did not identify hotspots, it was not used in this analysis (DRCOG 2017).
Crash Cluster Analysis
To supplement the reactive crash-based analyses from Denver Public Works and DRCOG, we performed our own crash cluster analysis of pedestrian and bicyclist crashes in Denver that is specific to children. Crashes were pulled from DRCOG’s Regional Data Catalog in GIS point format for the years 2010-2014. We included any crash involving a pedestrian or bicyclist under the age of 14 in our analysis in order to match the proactive analysis survey. We did not account for exposure, as reliable counts of child pedestrians and bicyclists were not available.
60


Proactive Analysis
The goal of the proactive safety analysis was to quantify the number of child pedestrian and bicyclist trips that were suppressed because of traffic safety concerns and utilize this as an indicator of safety. There were two aspects of the proactive safety analysis: 1) suppression rates based on survey-derived parental perceptions of roadway characteristics; and 2) shortest path distances to school based on a GIS closest facility network analysis.
For the suppression rates, we administered a survey to parents of children enrolled in prekindergarten through 8th grade in Denver, Colorado, asking them which roadway characteristics they would allow their child to walk or bike on. Roadway characteristics included number of lanes, posted vehicle speed limits, vehicle volumes, and the presence of sidewalks and bike lanes. The survey excluded parents of high school students because high school students have more independence than younger students and would be more likely to drive themselves or carpool with a friend. The survey was offered exclusively online and was marketed through newsletters, fliers, and social media by Denver Public Schools, parent-teacher organizations, the City and County of Denver, and local advocacy groups. The survey was open for one month in 2017 and resulted in responses for 1,331 children.
To derive the shortest path distances to schools, we utilized a closest facility GIS network analysis with child home locations as origins and schools as destinations. We approximated child origin locations on the block group level by creating a random point for each child with population numbers coming from the 2015 American Community Survey on the National Historical Geographic Information System (NHGIS) (Manson et al. 2017). These origin points were clustered according to residential building footprints obtained from the City and County of Denver’s Open Data Catalog in polygon shapefile format. School destinations were in point shapefile format from
61


DRCOG’s Regional Data Catalog. Only public elementary and middle schools were considered in the analysis.
Data for Denver’s street network came from the Denver Open Data Catalog in line shapefile format. This layer included posted speed limits and the number of lanes for each road segment. We utilized vehicle volumes provided by DRCOG, with any roadway having more than 1,000 vehicles per day being designated as high volume (Cornell Local Roads Program 2014). We noted the presence of sidewalks based on the DRCOG Regional Data Catalog’s sidewalk layer provided in line shapefile format. We accounted for bike lanes based on their location per Google Maps, satellite imagery, and Google Street View and accounted for the off-road path network based on a layer provided by the Denver Open Data Catalog.
Methods
The goal of this paper is to compare a proactive pedestrian and bicyclist safety analysis — based on trips suppressed because of traffic safety concerns — to a reactive safety analysis. In order to develop the proactive analysis, we needed to look at children’s trips to school because such trips are structured and more readily modeled than adult walking and bicycling trips. However, existing crash-based analyses examined pedestrians and bicyclists of all ages. Therefore, we identified locations with high rates of child pedestrian and bicyclist crashes, just as the Denver Public Works and DRCOG analyses had done for all age groups. We show that reactive analyses for both children and all ages provide similar results, and then compare both to our proactive analysis.
62


Reactive Analyses
For the reactive analyses, we first utilized existing crash-based pedestrian and bicyclist safety reports. Since these studies were for all ages, we complemented their results with our own crash cluster analysis of child pedestrian and bicyclist crashes.
Existing Reports
The first reports forming the basis of our reactive analysis were produced by Denver Public Works. These pedestrian and bicycle safety reports identified locations that experienced the most pedestrian and bicyclist crashes and then make recommendations for engineering, education, enforcement, and evaluation treatments. The reports included maps and tables identifying hotspots and corridors where the highest number of pedestrian and bicyclist crashes occurred. Hotspots were exclusively at intersections. The corridors depend on roadway length as a form of exposure to generate a rate of crashes per mile. The reports examined pedestrian crashes and bicyclist crashes separately, but for all ages.
The second report forming the basis of our reactive analysis was created by DRCOG. It similarly located all-age pedestrian and bicyclist crash hotspots (exclusively at intersections) and corridors with high levels of pedestrian and bicyclist crashes. These corridors also rely on roadway length to generate the number of pedestrian and bicyclist crashes per mile. In addition to reporting the location of pedestrian and bicyclist crash hotspots, the report explored environmental and operational characteristics of the crashes. The report also made recommendations for treatments in terms of engineering, enforcement, and education, although not specific to any individual hotspot.
We took the results from these existing pedestrian and bicyclist crash analyses and imported them into GIS. We assigned one point for each hotspot and one line for each high-crash corridor.
63


We took the total number of crashes at these locations and calculated the number of crashes per year.
Crash Cluster Analysis
While the proactive safety analysis was designed to examine child pedestrians and bicyclists, the existing pedestrian and bicyclist safety reports examined users of all ages. In order to examine spatial patterns of child crashes, we pulled all child pedestrian and bicycle crashes that occurred between 2010 and 2014 from DRCOG’s Regional Data Catalog. We analyzed the spatial relationship between these child crashes to identify statistically significant crash clusters. We then explored the spatial layout of the crashes in these clusters to identify the geographic extent of the deviational ellipses.
We utilized the Optimized Hotspot Analysis tool in Esri’s ArcMap to identify statistically significant clusters. The tool functions to identify both hotspots and cold spots in incident data. When inputting the data, we considered each crash point as a single equally-weighted incident. The tool automatically aggregates all incidents that are found to be clustered into a mean centroid point and outputs the Getis-Ord Gi* statistic in the form of a confidence level bin for each identified cluster. This statistic indicates the statistical significance of the spatial cluster. We considered clusters that were in the 95% confidence bin (two standard deviations for normally-distributed data) and had at least three incidents in this study. Once these aggregated clusters were identified, we returned to the original crash point layer and assigned each crash to its appropriate cluster (Figure 1).
64


Figure 1. Cluster analysis methodology: A) Input all applicable crashes; B) Utilize Optimized Hotspot Analysis tool to locate cluster and query statistically significant clusters (in red); C) Utilize Direction Distribution tool to form standard deviational ellipses around significant clusters.
We then utilized the Directional Distribution tool in Esri’s ArcMap to form standard deviational ellipses around the identified clusters. Standard deviational ellipses measure the dispersion and orientation of the crashes around the mean center of the cluster by using the standard deviation of the distance between the mean center and the x- and y-coordinates to define the ellipses’ axes (Schneider, Ryznar, and Khattak 2003). We considered the ellipses that consisted of two standard deviations that — in two-dimensional space — have been shown to correlate to 95% confidence in normally-distributed data (Esri 2018). With 95% confidence, we can say that the space enclosed by the ellipses are within a crash cluster.
65


Proactive Analysis
While we found the locations of crash hotspots and high-crash corridors for pedestrians and bicyclists of all ages and paired those with the clusters of child pedestrian and bicyclist crashes, we hypothesize that examining only crashes neglects child pedestrian and bicyclist safety issues — namely, places that are perceived as so unsafe they have reduced levels of exposure. To address these gaps, we measured the number of child pedestrian and bicyclist trips that are suppressed because of safety concerns. This forms our proactive measure of safety. To accomplish this, we first administered a parental survey to quantify how perceptions of roadway characteristics influence suppression of children’s pedestrian and bicycle trips. We then completed a GIS network analysis to find the shortest path distance between the estimated home location of each child in Denver and their closest school. Finally, we applied the trip suppression rates to the road network and ran the network analysis once again. This allowed us to identify children who are within an appropriate walkshed or bikeshed under ideal conditions but are forced to travel beyond their walkshed or bikeshed because of safety concerns. This is what we define as our suppressed trips.
A survey of parents of children in elementary or middle school provided us with the rate at which children’s pedestrian and bicyclist trips are being suppressed because of safety concerns. The 1,298 survey respondents provided us with 924 complete responses accounting for 1,331 children. The survey was based on the methodology developed for the Denver Perceptions of Safe Routes to School Survey and was available for one month exclusively online in both English and Spanish (Ferenchak & Marshall 2018). The survey presented parents with a variety of scenarios consisting of varying roadway design characteristics and asked whether they would allow their child to walk or bike to school on each roadway (Figure 2). Parents were able to answer “No”, “Yes, with trusted adult supervision”, and “Yes, without adult supervision”. We included five randomly selected walking questions and five randomly selected bicycling questions from a pool of twenty walking
66


questions and twenty bicycling questions. Parents provided information regarding child age, gender, and physical activity levels. Roadway design characteristics included number of lanes (2, 3, or 4 lanes), posted speed limits (25 mph, 35 mph, or 45 mph), the presence of sidewalks and bike lanes (none or on one or both sides), and vehicle volumes (low or high volumes).
67


8. Would you allow your child to use this roadway on foot to get to school?
25 mph Speed Limit 3 Lanes Sidewalks
Low Vehicle Volume
35 mph Speed Limit 2 Lanes Bike Lanes Low Vehicle Volume
Figure 2. Example of survey questions.


Since the number of roadway scenarios in Denver exceeded the number of scenarios that we could feasibly include on the survey, we created linear regressions and used them to derive a suppression rate for each roadway in Denver (Table 1). We accomplished this by coding all forty roadway scenarios featured in the survey by their four predictor variables and designating the outcome variable as the percentage of parents that would not allow their children to use the roadway. Using these linear regressions, we then utilized the street-level design characteristics for each scenario that was not featured in the survey and derived the corresponding suppression rate. More details on the survey and the trip suppression model can be found in Ferenchak & Marshall (2018).
Table 1. Linear Regression Coefficients for Trip Suppression from Parental Survey
Walk (R2 = 0.9654) Bike (R2 = 0.9227)
Intercept 0.037 -0.076
Speed (mph) 0.015*** 0.010***
Lanes 0.046* 0.105***
Facilities -0.248*** -0.086***
Volume 0.131*** 0.230***
*p<0.10 ** p<0.05 *** p<0.01
Our goal was to then derive the number of suppressed trips. We accomplished this by first deriving the shortest path distance between each child’s estimated home location and their closest school and then figuring out how much distance would be added to those trips if roads perceived as unsafe were avoided. We included children outside of Denver if their closest school was located in Denver. Because of privacy issues, we approximated the location of children’s homes on the block group level by creating one random point for each child living in each block group. We assigned random points to residential areas so that trips would originate in somewhat realistic patterns. While
69


the assumption that each child will attend their closest school is known to be faulty because of the Colorado Open Enrollment program that allows children to attend schools other than their originally assigned neighborhood school, privacy issues precluded us from knowing which school each child actually attends.
By connecting children to their closest school, we first determined the distance that children would walk or bike to school if safety or perceived safety were not an issue (i.e. the shortest path distance). Even if these children were not attending their closest school, these are the areas that we would also expect trips to playgrounds or parks, assuming that many of children’s trips originate from their home.
After determining shortest path trip distances without considering safety perceptions, we ran the network analysis once again with each road segment now being weighted according to safety-perception-based suppression rates. Roads for which more than 50% of parents reported they would not allow their children to walk or bike on were weighted so as to be avoided in the second running of the network analysis. Since roadways had different suppression rates for walking and biking, we ran this weighted network analysis twice, once for each mode. We then derived the distance of each trip and compared the weighted distance to the original shortest path distance. We defined a trip as being suppressed when a child who would have been within a half-mile walkshed or a one mile bikeshed for their shortest path was forced to exceed that walkshed or bikeshed for their weighted path (Figure 3). For example, a child’s trip would not be counted as suppressed if they lived 0.40 miles from their closest school and avoiding a road perceived as unsafe caused the trip to become 0.45 miles. Flowever, if a child lived 0.40 miles from their closest school and avoiding a road perceived as unsafe caused the trip to become 0.55 miles, the walking trip was considered suppressed. If a child lived 0.55 miles from their closest school in the first place, we did not consider that child as a possible pedestrian.
70


IV*
500
Child
X School Barrier Routes
Shortest Path (2074') Weighted Distance (3004')
1,000
Ford
(Barney L) Park
t*tRP A V N
2,000 l Feet
\
on Park
Figure 3. Visualization of a suppressed pedestrian trip.
After identifying suppressed trips, we accounted for the relationship between intersection density and walking and bicycling to school. We found the intersection density within a half-mile buffer around each road segment and applied odds-ratios based on high- and low-density areas from past research of children’s active transport to school behavior (Schlossberg et al. 2006). In other words, if there are similar numbers of suppressed trips in both a high-density and a low-density neighborhood, we would expect more trips in the low-density neighborhood to be stifled because of network characteristics and not as many trips to be suppressed because of safety concerns. Once we identified suppressed trips, we identified the start point of the trip in order to locate the areas that would be most impacted. We then ran a kernel density to illustrate the areas with high concentrations and to enable spatial comparison of the proactive and reactive analyses.
71


Results
We first examined results from existing reactive crash-based traffic safety analyses (all pedestrians and bicyclists). We then utilized these methods and complemented the all-age results with our own reactive crash-based cluster analysis of child pedestrian and bicyclists. This enabled us to pursue a direct comparison between a reactive and proactive analysis. We then performed a proactive safety analysis based on trips that are suppressed because of safety concerns for child pedestrians and bicyclists. Finally, we compared the reactive analyses to the proactive analysis and observed how the perspectives align and vary in their interpretation of traffic safety.
Reactive Analyses
Two safety reports from Denver Public Works and one from DRCOG provided a perspective on the reactive methodologies that existing traffic safety analyses employ. While these analyses were for pedestrians and bicyclists of all ages, they provided a methodological framework for us to complete our own crash-based cluster analysis of child pedestrians and bicyclists. We performed a cluster analysis of DRCOG’s child pedestrian and bicyclist crashes in order to formulate a reactive child traffic safety perspective. All crash-based reactive analyses (for both children and all ages) indicated that the bulk of pedestrian and bicyclist traffic safety issues in Denver are proximate to downtown Denver, Federal Boulevard, and Alameda Avenue.
Existing Reports
Denver Public Works identified five bicyclist crash hotspots and six bicyclist high-crash corridors in their 2016 safety report (Denver Public Works 2016). They then identified six pedestrian hotspots and five pedestrian high-crash corridors in their 2017 safety report (Denver Public Works 2017). Table 2 shows that the worst pedestrian hotspot reported by Denver Public
72


Works (located at 20th Street and Market Street) experienced seventeen pedestrian collisions between 2011 and 2015, while the intersection of Colfax Avenue and Colorado Boulevard was the weakest hotspot noted in the report with ten pedestrian crashes. The worst bicyclist hotspot reported by Denver Public Works (located at N. Broadway and Colfax Avenue) experienced nine bicyclist collisions between 2008 and 2012, while three hotspots with seven collisions each were the weakest bicyclist hotspots noted in the report.
Denver consists of approximately 155 square miles of land. However, as Figure 4 shows, pedestrian and bicyclist crash hotspots were concentrated in a relatively small area proximate to downtown Denver and along a limited number of major corridors. The residential neighborhoods that comprise the rest of Denver have relatively few crash hotspots identified. Nearly all crash hotspots were found on at least one major roadway such as Colfax Avenue, Broadway, Alameda Avenue, or Federal Boulevard. While Denver Public Works did not control for levels of pedestrian and bicyclist exposure, high levels of pedestrian and bicyclist activity may be expected in these areas — especially in downtown — and may be one reason for the abundance of crashes. Regardless of whether high levels of non-motorized exposure exist, these downtown crash hotspots and high-crash corridors merit traffic safety interventions.
Table 3 shows the twelve pedestrian crash hotspots and eight pedestrian high-crash corridors that DRCOG identified in Denver along with four bicyclist crash hotspots and six bicyclist high-crash corridors. Colfax Avenue was identified as the worst pedestrian corridor in the Denver-metro area in terms of overall pedestrian crashes, with 418 crashes over the five study years. Colfax Avenue was also identified as the worst bicyclist corridor in the Denver-metro area with 163 crashes over the five study years. When considering crashes per mile as the safety metric, the Colfax Avenue corridor was ranked as the worst pedestrian corridor in the Denver-metro area, but several
73


corridors in Boulder surpassed Colfax Avenue in terms of bicyclist crashes per mile. All but two of the sixteen crash hotspots identified in the DRCOG report were found on high-crash corridors.
Figure 5 shows that the DRCOG hotspots and high-crash corridors are also found near downtown Denver and along high-crash corridors such as Federal Boulevard, Alameda Avenue, and Sheridan Boulevard. The residential neighborhoods that comprise the rest of Denver had few hotspots identified. Like the Denver Public Works reports, exposure was not accounted for in the DRCOG report. Because of the location of the crash hotspots, high levels of pedestrian and bicycling activity may play a role in the high number of bicyclist crashes, especially in downtown.
Findings from both reports suggest similar spatial patterns. When considering the traffic safety of pedestrians and bicyclists in Denver, based on these reactive approaches, an emphasis on downtown Denver and other major corridors is the suggested focus. Other neighborhoods found throughout Denver have few identified pedestrian or bicyclist hotspots.
74


Table 2. Pedestrian and Bicyclist Intersection Hotspots Identified by Denver Public Works and DRCOG
Report
Source Mode Years Ranking Street 1 Street 2 Crashes/Year
Denver PW Pedestrian 2011-2015 1 20th Street Market St 3.4
2 Colfax Ave Broadway 3.0
3 13th Ave Broadway 2.6
4 Federal Blvd Kentucky Ave 2.2
4 Colfax Ave Franklin St 2.2
6 Colfax Ave Colorado Blvd 2.0
Bicycle 2008-2012 1 Broadway Colfax Ave 1.8
2 Lincoln St Colfax Ave 1.6
3 Lipan St Evans Ave 1.4
3 Broadway 12th Ave 1.4
3 Kalamath St Alameda Ave 1.4
DRCOG Pedestrian 2003-2007 1 Colfax Ave Sheridan Blvd 2.2
2 Colfax Ave Broadway 2.0
3 Federal Blvd Louisiana Ave 1.6
3 Colfax Ave High St 1.6
5 Federal Blvd Alameda Ave 1.4
5 Colfax Ave Mariposa St 1.4
5 Colfax Ave Ogden St 1.4
5 Colorado Blvd Mississippi Ave 1.4
5 Colfax Ave Colorado Blvd 1.4
5 10th Ave Sheridan Blvd 1.4
5 Colfax Ave Lipan St 1.4
5 Lincoln St 8th Ave 1.4
Bicycle 2003-2007 1 Alameda Ave 1-25 1.8
2 Alameda Ave Kalamath St 1.6
3 Broadway Colfax Ave 1.4
4 Cherry Creek Dr Monaco Pkwy 1.2
75


Table 3. Pedestrian and Bicyclist High-Crash Corridors Identified by Denver Public Works and
DRCOG Report Crashes
Source Mode Years Ranking Street Segment /Yr/Mile
Denver PW Pedestrian 2011-2015 1 E Colfax Ave 1.4
2 N Broadway 1.3
3 S Federal Blvd 0.9
4 W Colfax Ave 0.7
5 N Federal Blvd 0.5
Bicycle 2008-2012 1 12th Ave 3.5
2 15th St 3.0
3 E 16th Ave 2.8
4 E Colfax Ave 2.0
4 Lincoln St 2.0
4 Broadway 2.0
DRCOG Pedestrian 2003-2007 1 Colfax Ave Union to Buckley 4.6
2 Broadway C-470 to Brighton 3.1
3 Federal Blvd Bowles to 104th 2.5
3 Colorado Blvd Hampton to 1-70 2.5
5 Alameda Ave Union to University 2.0
5 Peoria St Parker to 56th 2.0
7 Mississippi Ave Wadswrth-Brdway Parker to Buckley 1.9
8 Sheridan Blvd Hampton to 104th 1.7
Bicycle 2003-2007 1 Colfax Ave Union to Buckley 1.8
2 Broadway C-470 to Brighton 1.4
3 Mississippi Ave Colfax Ave 1.2
4 Alameda Ave Monaco Pkwy 1.1
5 Colorado Blvd Wadswrth-Brdway Parker to Buckley 1.0
6 Federal Blvd Bowles to 104th 0.7
76


Figure 4. Pedestrian (above in red) and bicyclist (below in blue) crash hotspots and high-crash corridors as identified by the Denver Public Work’s 2016 and 2017 crash-based safety reports (all ages and all trip purposes).
77


Figure 5. Pedestrian (above in red) and bicyclist (below in blue) crash hotspots and high-crash corridors as identified by DRCOG’s 2012 crash-based safety report (all ages and all trip purposes).
78


Crash Cluster Analysis
The results from existing safety reports focused on pedestrians and bicyclists of all ages. We wish to explore the spatial patterns of child pedestrian and bicyclist crashes so that we can compare them to our proactive analysis. Would a reactive crash-based safety analysis of child pedestrians and bicyclists focus on the same areas as our existing reports? We use the existing methodologies to create our own analysis.
In order to explore the above question, we pulled child pedestrian and bicyclist crashes from DRCOG’s Regional Data Catalog. We utilized all 342 child pedestrian crashes and 208 child bicyclist crashes that were reported within Denver between 2010 and 2014. Of these 342 child pedestrian crashes, there were eleven clusters identified consisting of a total of 44 crashes. Of these eleven identified clusters, Figure 3 shows the six clusters were significant at 95% confidence and consisted of at least three crashes. These six clusters consisted of 24 child pedestrian crashes. The largest cluster had six crashes and the smallest had three. Of the 208 child bicyclist crashes, there were fifteen clusters identified consisting of a total of eighty crashes. Of these 33 identified clusters, Figure 3 shows the eleven clusters were significant at 95% confidence and consisted of at least three crashes. These eleven clusters consisted of 67 child bicyclist crashes. The largest cluster had twelve crashes and the smallest had three.
We then formed deviational ellipses around these significant clusters of child pedestrian and bicyclist crashes (Figure 6). These deviational ellipses spatially define where we would expect to find high levels of crash incidences. Figure 7 shows the pedestrian crash ellipses are focused in an area around South Federal Boulevard. The bicyclist crash ellipses are focused in the area proximate to downtown Denver (Figure 7). The rest of the residential neighborhoods that largely comprise Denver have few identified clusters. The majority of crashes occur in this central area near downtown as well.
79


0 0.125 0.25
i Miles
Figure 6. Detailed maps of child crash clusters for 2010-2014. (Data Source: DRCOG Regional Data Catalog)
80


Figure 7. Deviational ellipses around child pedestrian crashes (above) and child bicyclist crashes (below). (Data Source: DRCOG Regional Data Catalog)
81


These reactive safety analyses — the existing traffic safety reports from Denver Public Works and DRCOG as well as the child-focused crash cluster analysis — identify downtown Denver and a limited number of other major corridors as the areas of Denver that have the most pressing pedestrian and bicyclist safety issues. None of these analyses, however, identify safety issues in the other neighborhoods away from these major corridors of Denver.
Proactive Analysis
To perform the proactive safety analysis based on trips suppressed because of safety concerns, we first determined child pedestrian and bicyclist suppression rates and then completed closest facility network analyses using shortest paths and suppression-weighted paths.
Based on survey results from parents, the most important roadway characteristic in terms of walking trip suppression was the presence of sidewalks. This factor was a 5.5 times stronger predictor of trip suppression than the next most significant factor. There are limited sidewalk gaps in central Denver, while there are more sidewalk gaps present in east and north Denver (Figure 8). The most important roadway characteristic in terms of bicycling trip suppression was vehicle volumes followed closely by the presence of bicycle lanes. High volume roadways are found throughout Denver while bike lanes are primarily located downtown and in an east-west orientation to the northeast of downtown Denver (Figure 8).
82


Figure 8. Location of pertinent roadway characteristics in Denver. (Data Source: Denver Data Catalog and DRCOG Regional Data Catalog)
There were 136,138 children included in the proactive analysis with 112,648 children living in Denver and 23,490 children living in municipalities directly bordering Denver. Figure 9 shows the highest concentrations of children in the residential areas to the west and northeast of downtown Denver. We included 217 public elementary and middle schools — both inside and outside Denver — in the analysis.
83


Figure 9. Child population concentrations (red - higher concentration) and school locations in Denver. (Data Source: U.S. Census Bureau)
Within GIS, the network analysis tool was able to derive a pathway to each child’s closest school. Of the children considered in the proactive analysis, 56.8% of children had a shortest path of 0.5 miles or less to their closest school (via network distance as opposed to Euclidian distance) and 93.1% of children had a shortest path of one mile or less (Figure 10). We focused on children who — without accounting for safety perceptions — had a half-mile or shorter walk or a one-mile or shorter bike ride.
84


Figure 10. Half-mile (walkshed) and one-mile (bikeshed) network buffers around schools in Denver.
Next, we weighted the network analysis with trip suppression rates and ran the analysis once again for both walking and bicycling trips. We selected children who were within the half-mile walkshed and one-mile bikeshed for their shortest path trip but were outside the walkshed and bikeshed once safety concerns were accounted for. We found 3,876 children who had their walking trips suppressed and 20,364 children who had their biking trips suppressed once safety concerns were used to weight the network analysis. Figure 11 shows that these suppressed trips were found throughout several neighborhoods primarily concentrated in west, east, and northeast Denver for walking trips and in east Denver for biking trips. There were relatively few suppressed trips found in central and southeast Denver. These trends are most likely related to the lower concentrations of children and fewer safety perception issues (there are few sidewalk gaps and many bike lanes) found in those areas. The large concentration of walking suppression in east Denver coincides with high
85


concentrations of children and sidewalk gaps (Figure 8 & Figure 9). The areas of high bicycling suppression similarly coincide with high concentrations of children and unfavorable roadway characteristics.
86


Figure 11. Suppressed child pedestrian (above) and bicyclist (below) trip origins.
87


Comparison
The reactive pedestrian and bicyclist safety analyses identify areas in central Denver and along major corridors as the areas of most concern. Safety issues in these areas have manifested in the form of high levels of crashes. However, Figure 12 shows that — in addition to the pedestrian analysis identifying S. Federal Boulevard — the proactive analysis identifies areas in east, northeast, and west Denver as having perceived traffic safety issues. These safety concerns may be hidden by a lack of non-motorized activity (a result of trip suppression), which precludes crashes from occurring in the first place and from being visible in a conventional reactive analysis.
88


Reactive Analyses
0 Denver PW Hotspot O DRCOG Hotspot Denver PW Corridor DRCOG Corridor â–¡ Crash Deviational Ellipse Denver
I Miles
Proactive Analysis
Suppressed Trip Concentrations
â–¡ Low High
Reactive Analyses
0 Denver PW Hotspot 0 DRCOG Hotspot — Denver PW Corridor DRCOG Corridor □ Crash Deviational Ellipse | Denver
Proactive Analysis Suppressed Trip Concentrations
â–¡ Low
I High
Figure 12. Spatial comparison of crash-based reactive analyses and proactive analysis (pedestrian on top; bicycle on bottom).
89


The proactive analysis presents fine-grained results with meaningful implications. By identifying a single point as a crash hotspot through a reactive analysis, we do not understand the larger context of the issue. Similarly, by identifying a lengthy high-crash corridor, we don’t know exactly where the issue is or what is causing it. However, with the proactive analysis, an analyst can observe where trips would be occurring, and exactly what is blocking them. Figure 13 shows that there are distinct clusters of suppressed trips throughout neighborhoods in Denver that do not necessarily fall on a major corridor or at a major intersection. One single gap in a sidewalk could possibly suppress all children over a three-block area. While a single missing sidewalk does not constitute a major safety issue from a crash-based perspective, it is a major safety issue for this particular neighborhood and can be picked up through a proactive analysis. Not only are the results of the proactive analysis new and important findings, but the form of the findings presents a more practical way of addressing safety.
90


Figure 13. Example of suppressed pedestrian trip origins and paths.
91


Full Text

PAGE 1

CHILD PEDESTRIAN AND BICYCLIST SAFETY: A PROACTIVE APPROACH VIA SAFETY PERCEPTIONS by NICHOLAS NATHAN FERENCHAK B.A., Lafayette College, 2010 M.A., West Chester University, 2013 A dissertation submitted to Faculty of the Graduate School of the University of Colorado in partial fulfillment of the requirements for the degree of Doctor of Philosophy Engineering and Applied Science Program 2018

PAGE 2

ii © 2018 NICHOLAS NATHAN FERENCHAK ALL RIGHTS RESERVED

PAGE 3

iii This thesis for the Doctor of Philosophy degree by Nicholas Nathan Ferenchak has been approved for the Engineering and Applied Science Program by Bruce Janson, Chair Wesley E. Marshall, Advisor Carolyn McAndrews Christopher M. Weible Deborah S. K. Thomas Date: May 12, 2018

PAGE 4

iv Ferenchak, Nicholas Nathan (PhD, Engineering and Applied Science Program ) C hild Pedestrian and Bicyclist Safety: A Proactive Approach via Safety Perceptions Thesis directed by Associate Professor Wesley E. Marshall ABSTRACT The purpose of this dissertation is to develop a more comprehensive manner of considering non motorized transportation safety. I accomplish this through three papers. First, I perform a traditional crash based analysis that while identifying new urban areas that deserve attention in terms of pedestrian safety sheds light on the shortcomings of reactive traffic safety approaches. Then, I build our theoretical framework for proactive safety analyses, develop a set of methodologies, and administer a sur vey to explore the relationship between safety perceptions and pedestrian and bicyclist travel behavior. Finally, I apply this proactive pedestrian and bicyclist safety approach and compare its results to reactive analyses . I f our goal is to get more people walking and biking safely as opposed to simply reducing the number of crashes then findings suggest that proactive analyses can contribute new and meaningful perspectives on pedestrian and bicyclist safety. The form and content of this abs tract are approved. I recommend its publication. Approved: Wesley E. Marshall

PAGE 5

v A CKNOWLEDGEMENTS I would like to first thank my adviser , Wesley Marshall , whos e tireless guidance and keen transportation insights ha ve done much to aid the develop ment of both my career and me as a person . I cannot thank you enough for taking the time to encourage and advise me through the last four years. I would also like to express my gratitude to everyone in the Civil Engineering Department at the University of Color ado Denver , especially Bruce Janson whose help through numerous classes, personal correspondences, and as the chair of this committee has not only done much to strengthen my quantitative abilities but also shown me new and perceptive ways of looking at t ransportation . I would be remiss to not mention the guidance I have received from work being done by the Urban and Regional Planning Department as well as everyone on this committee , which helped provide direction for this work. I am also indebted to colleagues too numerous to mention here with whom I have collaborated and learned from over the last four years as I advanced through this program. I would like to express my deep gratitude to Laura for being the mo st incredibly supporting and loving partner anyone could ask for . Thank you for always being willing to discuss ideas when I was stuck, help ing me to see things from different perspectives, aiding whenever there was a deadline looming, and always being th ere for me in every aspect . These have been four of the most enjoyable and rewarding years of my life, and it was in a very large part thanks to you. Thank you for everything. To my parents , thank you for always being there for me throughout my life and for instilling in me the value of hard work and persistence . Those characteristics were certainly vital over the last four years. In addition to support from the rest of my family, I would also like to hon or Jennifer Jacksits who provided motivation for my work and kept me going when I might otherwise have given up .

PAGE 6

vi The COMIRB submission for protocol 17 1624 for the Safe Routes to School survey ( Submission ID APP001 1 ) was awarded an exemption on September 21, 2017. This work was supported by the Mountain Plains Consortium University Transportation Center (grant number s MPC 515 and MPC 557). The Mountain Plains Consortium was not involved in study design, analysis, or writing of the paper s . Wesley Marshal l was co author for each paper .

PAGE 7

vii TABLE OF CONTENTS CHAPTER I. INTRODUCTION ................................ ................................ ................................ ...................... 1 II. REDEFINING THE CHILD PEDESTRIAN SAFETY PARADIGM Introduction ................................ ................................ ................................ ................................ ... 4 Data ................................ ................................ ................................ ................................ ................. 6 Crash Dat a ................................ ................................ ................................ ............................... 7 Exposure Data ................................ ................................ ................................ ........................ 7 Child Friendly Destinations Data ................................ ................................ ........................ 8 Methods ................................ ................................ ................................ ................................ ........ 10 Phase I Study Methodology ................................ ................................ ................................ 10 Phase II Study Methodology ................................ ................................ ............................... 12 Results ................................ ................................ ................................ ................................ ........... 13 Conclusion ................................ ................................ ................................ ................................ .... 14 III. QUANTIFYING SUPPRESSED CHILD PEDESTRIAN AND BICYCLE TRIPS Introduction ................................ ................................ ................................ ................................ . 18 Theory ................................ ................................ ................................ ................................ ........... 19 Data ................................ ................................ ................................ ................................ ............... 22 Parental Perceptions Data ................................ ................................ ................................ ... 22 Population and Built Environment Data ................................ ................................ .......... 27 Methods ................................ ................................ ................................ ................................ ........ 27 Suppression Rates ................................ ................................ ................................ ................. 28 Network Analysis ................................ ................................ ................................ .................. 2 9 Suppressed Trips ................................ ................................ ................................ ................... 31

PAGE 8

viii Results ................................ ................................ ................................ ................................ ........... 32 Trip Suppression Factors ................................ ................................ ................................ .... 32 Trip Suppression Rates ................................ ................................ ................................ ........ 36 Number of Suppressed Trips ................................ ................................ .............................. 43 Discussion ................................ ................................ ................................ ................................ .... 44 Conclusion ................................ ................................ ................................ ................................ .... 49 IV. SUPPRESSED CHILD PEDESTRIAN AND BICYCLE TRIPS AS AN INDICATOR OF SAFETY Introduction ................................ ................................ ................................ ................................ . 53 Theory ................................ ................................ ................................ ................................ ........... 55 Dat a ................................ ................................ ................................ ................................ ............... 58 Reactive Analyses ................................ ................................ ................................ .................. 59 Existing Safety Reports ................................ ................................ .......................... 59 Crash Cluster Analysis ................................ ................................ ............................ 60 Proactive Analysis ................................ ................................ ................................ ................. 61 Methods ................................ ................................ ................................ ................................ ........ 62 Reactive Analyses ................................ ................................ ................................ .................. 63 Existing Reports ................................ ................................ ................................ ...... 63 Crash Cluster Analysis ................................ ................................ ............................ 64 Proactive Analysis ................................ ................................ ................................ ................. 66 Results ................................ ................................ ................................ ................................ ........... 7 2 Reactive Analyses ................................ ................................ ................................ .................. 72 Existing Reports ................................ ................................ ................................ ...... 72 Crash Cluster Analysis ................................ ................................ ............................ 79

PAGE 9

ix Proactive Analysis ................................ ................................ ................................ ................. 82 Comparison ................................ ................................ ................................ ........................... 88 Equity Analysis ................................ ................................ ................................ ...................... 92 Conclusion ................................ ................................ ................................ ................................ .... 9 3 V. CONCLUSION ................................ ................................ ................................ .......................... 9 5 REFERENCES ................................ ................................ ................................ ................................ ......... 9 8

PAGE 10

1 C HAPTER I INTRODUCTION The goal of this dissertation is to develop a more comprehensive traffic safety analysis methodology for pedestrians and bicyclists through a proactive approach . I developed this proactive approach through the administration of a survey intending to better understand the relationship between demographics, roadway characteristics, traffic safety perceptions, and behavior. I then compare results from th is proactive analysis to results from a reactive crash based analysis to determine how t he safety perspectives of the different approaches align and vary . Traffic safety is a pressing public health issue. In 201 6 , the number of people killed on American roadways rose to 37,461, a level not seen since 2007. This translates to m ore than 102 people ( on average ) killed every day in the United States . In addition, t here were an estimated 2,443,000 reported injur ies resulting from motor vehicle crashes in 2015 ( NHTSA 2016 ) . Th ese safety issues are especially serious for vulnerable populations such as children and non motorized users . Traffic fatalities are the leading cause of death for America ns between the ages of 10 and 24 years , while t here were 5 , 987 pedestrians and 840 bicyclists killed in 2016, numbers that overrepresent the share of t rips made by these modes ( CDC 2017; NHTSA 2017 ) . If we are t o improve these traffic crashes, injuries, and fatalities , we must better understand them. Where are they occurring? What is causing them? Who is most affected? A nalyz ing crash data to identi fy patter n s can help us to answer these questions . By understanding where, why, and when crashes are occurring, we can work towards avoiding them . Certainly, examining crashes has done much to improve traffic safety on our streets . However, examining pedestrian and bicyclist crashes is a reactive approach , one that necessitates that w e wait for a crash to occur and then try to understand why it happened . In other words, w e must wait for these safety issues to manifest themselves before we can identify and fix

PAGE 11

2 them. Another way to think about this reactive perspective is that the only pedestrians and bicyclists we are consider ing in a crash based analysis are those who are on the roads in the first place the people that have determined that the road is safe enough to try to walk or bike on . What about the people that because of traffic safety concerns decide not to walk or bike in the first place? These people are not getting hit and are therefore not considered in reactive analyses. Furthermor e, the only locations that are considered when taking a reactive crash based approach are those that people deem safe enough to walk and bike on in the first place . What about the areas where because of traffic safety concerns levels of walking and biking have been suppressed ? To identify these people and places, w e need a proactive approach to identify safety issues before they manifest themselves , or even before any walking or biking occurs at all. Such a proactive pedestrian and bicyclist approach would be able to identify areas where safety concerns have suppressed walking and biking trips, thereby lowering levels of walking and biking activity , lowering the quantity of crashes, and hiding safety issues from the objective eye. Trips suppressed specifically because of traffic safety concerns becomes our new proactive safety metric with which we can then compare reactive and proactive safety approaches. We accomplish th ese objectives through three papers. First , in Chapter II, we complete a reactive crash based analysis of child pedestrian fatalities in six of the quickest growing American cities. Along with identifying parks as an urban area that deserve s additional focus in terms of child pedes trian safety , this study reveals some of the drawbacks of such a reactive approach to safety . Then , in Chapter III, we develop a proactive analysis framework, create a methodology, and administer a survey to explore the relationship between traffic safety perceptions and child pedestrian and bicyclist trip suppression in Denver, Colorado. Findings show that sidewalks are the most important roadway factor for walking trip suppression, vehicle volumes and bike lanes are the most important factors for biking trip suppression, and trip suppression is concentrated near

PAGE 12

3 network characteristics such as barrier connections and tributary configurations . These results are used in Chapter IV to implement a proactive child pedestrian and bicyclist safety analysis in Denver. We finally compare results from this proactive safety analysis with results from reactive analyses to understand what additional insights a proactive approach may be able to lend . Findings suggest that proactive analyses add new and important perspectives on pedestrian and bicyclist safety. Specifically, such approaches identify areas where trips would be expect ed to occur if it were not for the presence of inadequate facilities . This is opposed to results from a reactive crash based approach that focuses on areas with high levels of existing exposure. Proactive methods are also capable of providing fine grained results within neighborhoods as opposed to reactive results focused p rimarily on major roads. If our goal is to reduce the number of pedestrian and bicyclist crashes, then a crash based analysis makes sense. However, if our goal is to get more people walking and biking safely, then we also need to consider safety from a p roactive perspective.

PAGE 13

4 C HAPTER II REDEFINING THE CHILD PEDESTRIAN SAFETY PARADIGM : IDENTIFYING HIGH FATALITY CONCENTRATIONS IN URBAN AREAS 1 Introduction Child pedestrians are some of the most vulnerable users of our transportation systems, and they deserve particular attention when we consider traffic safety. The objective of this work is to identify urban locations in which child pedestrians are at particular risk for fatal collisions with vehicles. This paper examines thirty years of crash data f or six American cities in order to locate areas with high child pedestrian fatality concentrations. Phase I of the study, which examines Denver, CO, reveals higher concentrations of child pedestrian fatalities around parks as compared to other areas that children have been shown to frequent. In Phase II of the study, we specifically examine fatality concentrations near parks as compared to schools. Statistical analyses suggest that, once exposure is controlled for, child pedestrian fatalities concentrate around parks in densities 1.04 to 2.23 times higher than around schools. Also, the concentration of child pedestrian fatalities around parks is 1.16 to 1.81 times higher than the respective citywide concentration. Traffic risks for children around parks deserve further examination as we pursue the goals of Vision Zero and child safety on our streets. W alking for transportation during childhood has important health and social benefits as it encourages physical activity and independence ( Larsen, Buliung , & Faulkner 2013; Loukaitou Sideris & Sideris 2009). Yet, children are often not able or allowed to safely and comfortably walk to their destinations. Traffic safety is one of the primary barriers to such active transportation in children (Centers for Dise ase Control and Prevention 1999) . Motor vehicle collisions are the leading cause 1 Portions of this chapter were previously published in Injury Prevention (Issue 23 6, 2017) and are included with the permiss ion of the copyright holder.

PAGE 14

5 of death for individuals from the age of 4 through the age of 24 in the United States, with pedestrians being the second most at risk user type (Centers for Disease Control a nd Prevention 2014) . Every hour, an average of 40 children die on roadways around the world, most of whom are vulnerable road users such as pedestrians (Toroyan & Peden 2007) . Despite the unfortunate road safety statistics of child pedestrians and the kn own health benefits of childhood walking, our transportation networks remain alarmingly dangerous for the few children that still walk independently. The question addressed through this work is: are there other land uses where we should be focusing our re sources beyond our traditional focus on schools to alleviate large concentrations of child pedestrian fatalities? Many researchers and practitioners have exerted considerable effort exploring child pedestrian safety around schools. These researchers and practitioners have found success when the necessary resources are allotted to combat the problem near school grounds. For example, reduced speed limits in school zones have been shown to lower vehicle speeds while projects funded by the Safe Routes to School program have reduced child pedestrian injury rates (Graham & Sparkes 2010; Abdul Hanan, King , & Lewis 2011; Dumbaugh & Frank 2007; DiMaggio & Li 2013; Orenstein, Gutierrez, Rice, et al. 2007) . However, other locations within our cities that are fr equented by children remain relatively unexplored (Kattan, Tay , & Acharjee 2011; Tay 2009) . The scant literature on the subject suggests that the areas around trails have relatively few child pedestrian crashes, while other research found that areas with few child pedestrian injuries contained a prevalence of parks and play areas, and similarly, that areas at high risk for traffic crashes involving pedestrians under the age of 15 were characterized by an absence of parks (Stutts & Hunter 1999; Kraus, Hoote n, Brown, et al. 1996; Joly, Foggin & Pless 1991) . Furthermore, an analysis of child injuries associated with playground visits in the United States found that pedestrian injuries were so uncommon that a statistical analysis was not possible (Phelan, Khou ry, Kalkwarf, et al. 2001) . This current work will

PAGE 15

6 fill the gap in the literature by further exploring concentrations of child pedestrian fatalities throughout our urban areas. In Phase I of the study, we use spatial and statistical analyses to compare c hild pedestrian fatality concentrations around schools to concentrations around other areas that children may frequent such as recreation centers, parks, and trails in Denver, CO using 31 years of crash data (Lee, Booth, Reese Smith, et al. 2005; Louka itou Sideris 2003) . In Phase II, based on findings from the first analysis, we then examine parks in more depth relative to schools in six different cities. The goal of these research activities is to identify the location of high concentrations of child pedestrian transportation systems. D ata The study cities were selected because focusing on rapidly growing cities would allow for the examination of current development patterns. While early U.S. cities were designed with pedestrians and streetcars in mind, those developed over the last century were primarily designed to cater to the automobile. Studying these modern auto centric cities will allow th e results to inform current building practices. By having a clearer understanding of the implications of our current community designs, we can build safer places for even the most vulnerable road users. According to Census data, the South was the quickes t growing region between 2000 and 2013, while the West was close behind (Cohen, Hatchard, & Wilson 2015) . Therefore, cities from these two regions became the focus of this study. Of the 25 most populous places across the United States, Austin had the l argest percentage increase in population from 2000 to 2013, Charlotte had the 2 nd largest increase, Denver had the 3 rd largest increase, and Dallas had the 10 th largest increase (Cohen, Hatchard, & Wilson 2015) .

PAGE 16

7 Houston had the 2 nd largest growth in total population while Los Angeles had the 4 th largest total growth (Cohen, Hatchard, & Wilson 2015) . These cities with substantial population growth are important to study because they are installing new infrastructure in new and unique land use configuration s. The safety outcomes of these new land use configurations are what we hope to explore through this work. Other cities with high growth rates were not used because comprehensive schools, parks, trails, and crash data were not available due to a lack of data collection. Crash Data We acquired child pedestrian fatality locations from the National Highway Traffic Safety data was available from 1982 to 2000 in address format, and from 2001 until 2012 with latitude and longitude coordinates. Crashes from 1982 to 2000 were geocoded on either the address level or, if the data did not contain enough detail, to the street level. Children were defined as persons un der Bureau through their Topologically Integrated Geographic Encoding and Referencing (TIGER) products. Exposure Data Due to consistency issues, finding reliable child pedestrian exposure data in geographically broad studies has historically been difficult (Wier, Weintraub, Humphreys, et al. 2009) . The best option for this particular study, when numerous exposure approaches were assessed, was to use a pop ulation based exposure metric such as that used by DiMaggio and Li (2013) in their safety examination of the Safe Routes to School program. In their study, DiMaggio and Li (2013) used the

PAGE 17

8 number of pedestrian crashes in selected census tracts and the numb er of persons living in those same census tracts to create a rate of crashes per 10,000 population for each of the tracts. While past studies modeled pedestrian exposure using proxy factors such as road network characteristics, land use, and socio economi cs, population based exposure metrics are also common and have proven useful for preliminary and/or geographically broad work (Wier, Weintraub, Humphreys, et al. 2009; Jacobsen 2003) . Given that our study fits both of these conditions, a population based metric facilitated a consistent child pedestrian exposure metric to study road safety across six U.S. cities. Although there were a number of limitations associated with this population based exposure metric (which will be detailed in the Conclusion secti on), analysis on the block group level did allow for finer grained contexts to be considered. The exposure variable for the analysis was the number of children living within the analysis zones. This variable was created by pulling the child populations fo r each block group from the 2010 Census and creating a random point for each child resident. This served as an indicator of the total number of children and a proxy for the relative level of child pedestrian traffic exposure in the study areas. This popu lation based exposure approach allows for a conservative analysis in terms of parks due to the fact that exposure around schools is typically higher than around parks. Almost all children attend school while not all children use parks. Also, schools get usage from weekday school trips and recreational trips for playgrounds and sports fields on the grounds while parks only experience usage for recreational purposes. Using the same population based exposure approach for all areas ensured a thoroughly conse rvative analysis of the risk around parks. Child Friendly Destinations Data We chose child friendly destinations because past research identified them as public places that children frequent as both recreational and physical activity resources (Lee, Boot h, Reese Smith,

PAGE 18

9 et al. 2005; Loukaitou Sideris 2003) . We obtained locational data for the child friendly destinations from the publicly available 2015 open data catalogs for the respective cities. The number of schools and parks within the study cities ranged widely (Table 1). Buffers were then created bas ed on the location of the buildings for schools and recreation centers and based on the parcel boundary for parks. This facilitated a more accurate representation of access than, for instance, if parks were based on a single point. The buffers for the tr ails were drawn adjacent the entire trail; however, this may not be representative of the actual access points. Study areas were designated by quarter mile buffers around the facilities. This quarter mile buffer size was chosen because it has been shown to be an appropriate access threshold for children, or the longest distance that children are typically allowed or able to independently walk to their destinations (Wolch, Wilson, & Fehrenbach 2005) . Also, the shortest service area with which parks and re creation areas are typically designed is one quarter of a mile (Cohen, Ashwood, Scott, et al. 2006) . For instance, regional parks are normally designed to serve entire cities, while pocket parks may be designed to serve just the surrounding blocks. Becau se every park has at least a quarter mile service area, this is an effective buffer size to use. in the buffer area. Since there were no fatalities within the parks , erasing the park area did not impact the number of fatalities, but ensured that the exposure variable was not inflated. Thus, the park buffer consisted of only the land one quarter mile outside of each park.

PAGE 19

10 T able 1 . Descriptive Statistics for St udy Cities Population Schools Park Area (Hectare) Rec. Centers (count) Trails (km) Austin 931,840 226 6,742.6 Charlotte 827,121 263 8,016.2 Dallas 1,300,082 221 7,618.6 Denver 682,545 227 7,583.2 30 142 Houston 2,298,628 1,180 10,236.5 Los Angeles 3,971,896 3,689 25,868.4 M ethods In Phase I, we examined child pedestrian fatality concentrations at four destinations that children frequent (i.e. schools, recreation centers, trails, and parks) in Denver, CO. The other study cities were omitted in Phase I because of data limitations. In Phase II, we investigated schools and parks in more detail across six study cities. Phase I Study Methodology Upon completion of the data collection and formatting, we initiated spatial analy sis by defining the study area buffers and calculating the number of child pedestrian fatalities in those

PAGE 20

11 F igure 1 . Child pedestrian fatalities relative to park buffers in Denver. After the number of child pedestrian fatalities was derived through spatial queries for each of the zones, the same procedure was run again to find the total number of children living within those able, allowing for a rate of fatalities per 10,000 children to be operationalized. There were no child pedestrian fatalities around recreation centers in Denver (Table 2). This suggests that recreation centers are not a primary problem for child pedestria n safety. Trails had rates similar to schools and parks. However, it is not clear if children use trails in the same manner that they use parks and schools. Access to trails is typically constrained, and the location of the child pedestrian fatalities n ear trails did not appear to necessarily correlate with trail access points. Trails

PAGE 21

12 typically have limited access points while child friendly destinations such as parks have more permeable access along their borders (Cutts, Darby, Boone, et al. 2009; Kriz ek, El Geneidy, & Thompson 2007 ; Krizek, Barnes, & Thompson 2009; Price, Reed, & Muthukrishnan 2012 ) . Parks were of interest due to the fact that they had the highest fatality rates. We therefore examined parks in Phase II by comparing their fatality rates to the fatality rates around schools, which have been the traditional focus. T able 2 . Child Pedest rian Fatality Rates per 10,000 Children Near Child Friendly Locations Schools Rec Centers Trails Parks Fatalities near child friendly locations 3.51 per 10,000 children 0.00 per 10,000 children 3.58 per 10,000 children 3.64 per 10,000 children Phase II Study Methodology Based on findings from the preliminary study, a second analysis of child pedestrian safety around parks was warranted. Parks and schools were therefore examined in more detail for six cities: Austin, TX; Charlotte, NC; Dallas, TX; Denver, CO; Houston, TX; and Los Angeles, CA. Using the same procedure from the previous analysis, the child populations and the number of child pedestrian fatalities were derived for analysis (Table 3). These variables wer e considered for areas near schools, areas near parks, areas near schools or parks, and areas near neither schools nor parks. The level of risk was derived for each city, location type, and year within the study. Confidence intervals were then computed.

PAGE 22

13 T able 3 . Child Pedestrian Fatality Statistics & Child Resident Statistics Near Destinations Child Pedestrian Fatalities Child Population (thousands) Total Schools (%) Parks (%) Total Schools (%) Parks (%) Austin 32 5 (15.6%) 15 (46.9%) 173 33 (18.9%) 70 (40.4%) Charlotte 62 12 (19.4%) 42 (67.7%) 177 23 (12.7%) 66 (37.4%) Dallas 108 13 (12.0%) 62 (57.4%) 325 57 (17.6%) 122 (37.6%) Denver 37 17 (45.9%) 29 (78.4%) 129 48 (37.6%) 80 (61.8%) Houston 172 39 (22.7%) 45 (26.2%) 564 128 (22.8%) 121 (21.5%) Los Angeles 417 246 (59.0%) 167 (40.0%) 872 486 (55.8%) 208 (23.9%) R esult s The results suggest that, for all of the study cities, child pedestrian fatality rates are significantly higher in areas near a school or a park than in areas near neither a school nor a park (Table 4). Fatality rates in areas that are near a park or a sc hool are significantly higher than the average citywide rates for five of the six study cities and not significantly different for one of the study cities. T able 4 . Child Pedestrian Fatality Rates per 10,000 Children Living Around Schools or Parks or Neither Schools nor Parks with 95% Confidence Intervals Citywide Schools or Parks Neither Schools nor Parks % Difference* Austin 1.85 (1.71, 1.99) 2.14 (1.91, 2.37) 1.57 (1.40, 1.74) 36.3% Charlotte 3.51 (3.28, 3.74) 5.77 (5.36, 6.18) 1.72 (1.56, 1.88) 235.5% Dallas 3.32 (3.17, 3.47) 4.36 (4.10, 4.62) 2.39 (2.22, 2.56) 82.4% Denver 2.87 (2.73, 3.01) 3.34 (3.15, 3.53) 1.52 (1.30, 1.74) 119.7% Houston 3.05 (2.95, 3.15) 3.60 (3.43, 3.77) 2.69 (2.57, 2.81) 33.8% Los Angeles 4.78 (4.58, 4.98) 5.34 (5.11, 5.57) 3.73 (3.53, 3.93) 43.2% *Statistically Significant Percent Differences from Schools or Parks to Neither Schools nor Parks are Bold

PAGE 23

14 Risk was found to be higher around parks than around schools for all of the study cities (Table 5). Dallas has the largest difference between schools and parks in terms of risk, with child pedestrians being over twice as likely to experience a fatality wi thin close proximity to a park than within close proximity to a school. All of the fatality rates around parks are significantly higher than the rates around schools except for in Denver. Rates around parks are also higher than the average rates citywide for all study cities and are significantly higher for each city except for Austin. Rates around schools are higher than the average citywide rates for just three of the six study cities, and only two of these are significantly higher. T able 5 . Child Pedestrian Fatality Rates per 10,000 Children Living Around Schools and Parks with 95% Confidence Intervals Schools Parks % Difference* Austin 1.53 (1.31, 1.75) 2.14 (1.90, 2.38) 40.5% Charlotte 5.33 (4.79, 5.87) 6.35 (5.87, 6.83) 19.1% Dallas 2.27 (2.02, 2.52) 5.07 (4.74, 5.40) 123.3% Denver 3.51 (3.20, 3.82) 3.64 (3.42, 3.86) 3.7% Houston 3.04 (2.87, 3.21) 3.71 (3.49, 3.93) 22.0% Los Angeles 5.06 (4.82, 5.30) 8.01 (7.51, 8.51) 58.3% *Statistically Significant Percent Differences from Schools to Parks Are Bold C onclusion While past efforts to ensure child pedestrian safety have focused primarily around schools, findings from this work suggest that parks may be an important location to focus on as well. In all of the six study cities, risk for child pedestrian fatalities i s higher around parks than around schools, although not all of these differences were statistically significant. The risk around parks has, prior to this research, been largely overlooked. Reasons for higher rates around parks may include unsafe

PAGE 24

15 streets along with a general lack of awareness, focus, education, and engagement in terms of the transportation safety issues present. There are two perspectives through which we may interpret solutions to this problem: transportation and urban design. Taking a transportation approach to the problem would have us lowering vehicle speeds and making drivers aware of child pedestrians through street design changes such as traffic calming, road diets, or pedestrian crossing treatments. A broader urban design approa ch would focus on the siting of our parks. If we site a park next to a 6 lane roadway with a high design speed, few transportation treatments would be able to help. Within the study cities, it was not uncommon to have a park separated from the community that it serves by roadways with four or six lanes. Some of these roadways have been documented with vehicle speeds greater than 70mph next to the adjacent park (Marshall 2015) . Siting parks on slow and narrow local roads within neighborhoods may help all eviate safety issues and thereby induce higher levels of independent walking. The most effective solution to the problem may very well lie in a combination of both of these approaches. We will need to ensure that parks are sited safely within neighborhoo ds and pedestrian infrastructure is included in a cohesive network to ensure safe access. In addition to these built environment improvements, other approaches such as child education, driver education, and enforcement methods may prove effective. There were several limitations present in this study. Many of the limitations were related to the measurement of child pedestrian exposure. A consistent exposure metric was necessary, which led to a population based exposure metric. We considered conduc ting a survey in order to measure exposure, but survey data have been found to significantly underrepresent child pedestrian exposure, and low response rates may introduce self selection issues (Routledge, Repetto Wright, & Howarth 1974; Roberts, Keall, & Frith 1994 ) . We also considered observational data, but observational data fails to properly consider potential endogeneity issues between perceived risk and exposure; in other

PAGE 25

16 words, a road perceived to be dangerous could be the cause of the low exposure and result in a seemingly good safety record. This would violate the independence assumption of most statistical models (Cho, Rodriguez, & Khattak 2009) . Moreover, observational data is difficult to acquire across multiple cities in large enough numbers to ensure sample sizes that reach statistical significance and are representative of actual conditions (Stevenson 1991) . For these reasons, a population based exposure metric was utilized. The exposure metric assumes that individual children will be exp osed to traffic dangers at similar rates across the study cities. While this assumption is not necessarily ideal, most children walking to a child friendly destination such as a school or park would likely live within a quarter mile of that school or park (Wolch, Wilson, & Fehrenbach 2005) . Examining finer geographic levels and exploring different methods of operationalizing child pedestrian exposure will be necessary in order to obtain a better understanding of the issue. The fact that children of all ages are assumed to act similarly and experience similar risk is another limitation of the exposure metric. In other words, the risk to a 5 year old pedestrian walking independently to a park is most likely higher than the risk to a 13 year old walking ind ependently to a park. However, the 5 year old pedestrian is more likely to be accompanied by a parent, typically alleviating some of the risk. This relationship between age and risk is complex and deserves more attention. Also, examining risk for child pedestrian injuries around parks would provide larger sample sizes and more robust statistical analysis than child pedestrian fatalities. Focusing on finer geographic levels may allow for an injury specific analysis. A further limitation was the lack of k nowledge pertaining to installation dates of schools and parks. It should also be noted that results may be exclusive to the generally warm climates of the study cities, and generalizability of the findings should not be assumed for other contexts. Other factors that may prove to be of importance include social factors such as population density,

PAGE 26

17 poverty, and crime, and built environment factors such as travel lanes, vehicle speeds, and cartway width. Child pedestrians, being highly vulnerable users of our transportation systems, find themselves at substantial risk as they move about our cities. Ensuring their safety is of the utmost importance. However, in order to ensure that safety, one must understand where safety risks are located. This study ha s shown that, opposed to traditional beliefs, there are higher concentrations of child pedestrian fatalities around parks than around schools. A shift in the child traffic safety paradigm is now needed to focus treatment efforts around our parks.

PAGE 27

18 C HAPTE R III Q UANTIFYING SUPPRESSED CHILD PEDESTRIAN AND BICYCLE TRIPS Introduction Traditional pedestrian and bicycle safety analyses take a reactive approach to traffic safety by investigating crashes, injuries, or fatalities after they occur. Also examining trips that have been suppressed because of road safety concerns allows for a more proactive safety approach; however, a methodology must first be developed to estimate the number of pedestrian and bicycle trips that are suppressed specifically due to road safety concerns. To accomplish this, we examine child pedestrian and bicycle trips to and from schools in Denver, Colorado. By combining suppression rates derived from a survey examining parental perceptions of safety and the upper limit of trip frequenc ies derived from a GIS network analysis, we explore how grade level, gender, and adult supervision are related to childhood travel allowance in terms of street level design characteristics such as posted speed limits, vehicle volumes, presence of sidewalks and bike lanes, and the number of vehicle lanes. We then investigate how widespread these suppressed trips are by quantifying the number of children that are impacted and how their routes would be altered. We finally detect built environment characteris tics such as street level designs, network configurations, barriers, and destination siting linked with high levels of suppressed trips. By incorporating this tool into traditional traffic safety analyses, we hope to not only make the places where chi ldren are currently walking and bicycling safer, but to improve safety for all places where children want or need to walk and bike. school, they typically analyze c rashes, injuries, or fatalities while accounting for the number of child pedestrians and bicyclists that are on the street also known as exposure. However, this approach to traffic safety is a reactive one, only looking at pedestrians and bicyclists tha t have deemed the

PAGE 28

19 traffic environment safe enough to use, and only looking at the streets where those pedestrians and bicyclists are currently walking or biking. Therefore, these approaches neglect both the pedestrians and bicyclists who want to walk or b ike but do not feel safe enough to do so and the places where such trips are being suppressed. To create a more proactive traffic safety analysis, we also need to account for the pedestrian and bicycle trips that never occurred in the first place because of road safety concerns. How would we measure such suppressed trips? Which personal and built environment characteristics would be associated with road safety related trip suppression? How many children would be impacted by trip suppression, and how wou ld their routes be altered? While traditional mode choice models output the expected share of different modes, we create a model that instead predicts the percentage of trips that are suppressed due to road safety concerns in order to answer these researc h questions. To create this safety perception based mode choice model, we used results from a survey that we administered to parents of elementary and middle school students in Denver, Colorado, along with linear and logistic regressions to explore how g rade level, gender, adult supervision, and street level design characteristics (e.g. posted speed limits, sidewalks, bike lanes, number of lanes, vehicle volumes) are related to trip suppression rates. We then derived the total number of trips expected un der ideal conditions based on a GIS network analysis. Finally, we combined trip suppression rates with the upper limit of trip frequencies to determine the total number of trips being suppressed specifically due to road safety concerns. T heory Crashes, i njuries, and fatalities normalized to levels of user exposure are typically employed to analyze transportation safety of both motorized and non motorized users (TRB 2001; Waldheim, Wempe, & Fish 2015; FHWA 2006; Zegeer et al. 2010). However, this reac tive

PAGE 29

20 approach only accounts for individuals who are using the facility and neglects those individuals who have deemed the roadway too unsafe to use in the first place. Accounting for suppressed trips is especially prescient for pedestrian and bicycle safe ty analyses, where many possible users could be expected to be dissuaded because of road safety concerns (Schneider, Ryznar, & Khattak 2003). Proactive safety approaches account for such suppressed trips, with high levels of suppressed trips signaling roa d safety issues, regardless of the presence of objective outcomes (Schneider, Ryznar, & Khattak 2003; Nevelsteen et al. 2012). However, past proactive safety analyses have only examined specific areas an approach that is not generalizable to other roadw ays while accounting for a limited number of roadway characteristics. A more holistic method of measuring pedestrian and bicycle trips that have been suppressed because of traffic safety concerns that can be applied more widely is therefore necessary. F ew researchers have ventured to estimate the number of pedestrian and bicycle trips that are suppressed because of traffic safety concerns. Schneider et al. (2003) formulated an early approach by developing a survey to identify areas on the campus of the University of North Carolina at Chapel Hill that are perceived as unsafe in terms of traffic safety by pedestrians. By asking individuals to identify the three locations on campus that felt the least safe for pedestrians, the researchers were able to identify areas with poor road safety perceptions and theoretically high trip suppression. However, these results are not transferable because the perceptions were not associated with specific built environment characteristics (i.e. street design, network connectivity, land use, etc.). Also, this approach did not unitize the results (i.e. we may know which site is ranked as the least safe, but we do not know the number of actual trips that are being suppressed); therefore, it is difficult to compare levels of suppression between different sites. Bellemans et al. (2009) similarly used a travel diary that asked respondents to record trips that they had planned but never executed. While Bellemans et al. (2009) did associate the suppressed trips with built

PAGE 30

21 en vironment characteristics, they looked primarily at household and personal schedule factors, rather than the specific impact of road safety. Furthermore, this methodology does not capture suppressed trips that never reached the planning stage. Cho et al. (2009) related trip suppression rates to built environment factors but only examined macro level environmental characteristics (land uses and road network density) to estimate suppressed pedestrian and bicyclist trips. While these large scale factors are certainly important, we wish to examine the impact of individual roadway characteristics because, while it is not easy to change established land use or road network configurations, transportation planners and engineers can more feasibly alter street level design characteristics. Nevelsteen et al. (2012) related suppressed pedestrian and bicycle school trips to individual roadway characteristics; however, the researchers only examined two factors (speed limits and the presence of non motorized facilities). Furthermore, the study took place in the Flemish Region of Belgium, which, with its high levels of active mobility, presents a radically different context than that of the typical American city. All of these past studies used suppressed trip estimates to take a more proactive look at road safety. We will build upon this past work by using parental perceptions of roadway characteristics to determine trip suppression and then apply those results in a citywide analysis. If we were to estimate suppressed tri ps based on street level design characteristics, which characteristics would be important to consider? Past mode choice models found that pedestrian and bicycle facilities, crosswalks and crossing treatments, traffic volumes and speeds, traffic calming fe atures, and crossing guards are important roadway characteristics that predict child mode choice to school (Larsen, Buliung, & Faulkner 2013). Evers et al. (2014) found that, for walking trips to school, parents perceive a lack of sidewalks and the presen ce of large streets as particularly influential in their decision to allow their child to walk or not. While crossing guards are not

PAGE 31

22 included in our analysis because their location may change , and traffic calming features are not included in our analysis because they are captured through the inclusion of traffic volumes and speeds, all other pertinent factors are included in our analysis. It is important to account for as many influential factors as possible so that we may understand how these factors rea ct with each other in terms of suppressed trips (i.e. maybe vehicle speed is an important factor, but it becomes less important if sidewalks are present). D ata Through this work, we endeavor to design a trip suppression model based on parental perceptions of roadway characteristics. To do so, data regarding both road safety perceptions and trips are necessary. We garner perceptions through a survey and derive trips from a closest facility GIS analysis using child populations (origins) and school location s (destinations). We utilized the City and County of Denver to build our safety perception based model. (118,886 under 15 years of age) spread out over the surrounded by medium density neighborhoods laid out in predominantly gridded street networks. throughout the city. Parental Perceptions Data We targeted a survey at parents of children in grades pre kindergarten through 8 th grade to garner parental perceptions of traffic safety. The survey excluded parents of high schoolers because high school students have mor e independence than elementary and middle school students and would also be more likely to drive themselves or carpool with a friend. The survey was offered

PAGE 32

23 exclusively online and was marketed through newsletters, fliers, and social media by Denver Public Schools, parent teacher organizations, the City and County of Denver, and local advocacy groups. The survey was open for one month from October 5 th , 2017 until November 5 th , 2017. Since 36.8% of DPS students identify as Spanish speakers and 55.5% are Hi spanic, we provided the survey and promotional materials in both languages. Thus, respondents first answered whether they would like to take the survey in English or Spanish. We next asked parents how many children they would like to complete the survey for. Parents could complete the survey for up to four children simultaneously. Respondents then provided the grade level and gender of each child that was included in the survey response. The Leuven Travel Behavior of Children to Primary School Survey ( Nevelsteen et al. 2012) served as a prototype for the travel behavior questions on the survey. Parents answered whether they would allow their child to either walk or bike along ten different picture based roadway five scenarios for pedestrian questions and five for bicycle questions). While the survey from Nevelsteen et al. (2012) included posted speed limits and presence of active transportation facilities as explanatory factors, Larsen et al. (2013) determined t hat crossings and vehicle volumes are also important explanatory roadway characteristics. Because crossings at intersections can be complex and difficult to represent in a picture (e.g. varying phasing, signalization, markings, signage, turning movements, etc.), we chose to utilize a combination of variables including the number of lanes, posted speed limits, and vehicle volumes as a proxy for crossing risk. While not perfect, these variables are related to the amount of time spent exposed to traffic risk and the degree of that risk. Accordingly, each roadway scenario in our survey had four different characteristics that were identified for the parent: the speed limit of the roadway, the number of lanes, the presence of active transportation facilitie s (i.e. a sidewalk for walking questions or a bike lane for bicycling questions), and the approximate vehicular volume of the roadway.

PAGE 33

24 Two of the factors had three levels (speed limit: 25 mph, 35 mph, or 45 mph; and number of lanes: 2 lanes, 3 lanes, or 4 lanes) while two of the factors had two levels (presence of facility: yes or no; and volume of the roadway: low or high). This resulted in a total of 36 different scenarios for walking and 36 different scenarios for bicycling. We removed non existent or rare scenarios from each pool (e.g. 45 mph roadways with two lanes and low volumes), resulting in a total of 20 scenarios for each of the walking and bicycling question pools. Each parent respondent answered five random walking scenarios and five random b icycling scenarios. Each scenario had a picture of a roadway and asked if the parent would let their child/children walk or bike to school along the roadway (Figure 1). The available responses were also had the ability to leave open ended comments after the scenarios were presented. We then asked respondents about the amount of physical activity their child/children get on a weekly basis and gave them the choi ce to enter an email address for the chance to win one of ten $50 gift cards that were offered as a survey incentive. Of the 1,298 survey respondents, 924 provided complete responses. These 924 complete parent responses accounted for 1,331 children. Ther e was an appropriate distribution of responses across grade levels and gender, while the majority of surveys were completed for one or two children (Table 1 and Figure 2).

PAGE 34

25 F igure 1 . Picture based roadway scenarios from the survey we administered to pa rents.

PAGE 35

26 T able 1 . Survey Response Descriptive Statistics Gender Male 667 Female 658 Other 4 Number of Children for each Survey 1 431 2 330 3 54 4 10 Minutes you would like your child to be physically active during school 20 9 30 96 40 117 50 45 60 322 60+ 244 Days your child got 60 minutes of physical activity in the last week 0 7 1 19 2 3 148 4 5 314 6+ 351 F igure 2 . Number of parental allowance responses from the survey for all roadway scenarios by grade level.

PAGE 36

27 Population and Built Environment Data We collected age based population data on the block group level from the 2015 American Community Survey via the National Historical Geographic Information System (NHGIS) (Manson et al. 2017). The Denver Open Data Catalog provided the sidewalk network, roadway network, and off (DRC OG) Regional Data Catalog provided traffic volumes and school locations in point shapefile format. We created the bike lane network in polyline shapefile format based on the location of bike lanes per Google Maps, satellite imagery, and Google Street View . M ethods The goal of this work was to create a trip suppression model based on parental perceptions of roadway characteristics to determine which personal and street level design characteristics impact trip suppression. We then integrated trip suppressi on rates derived from the survey with the upper limit of trip frequencies the number of expected trips under ideal conditions as derived from a GIS network analysis to determine the number of active transport trips to school that are suppressed because of road safety concerns, how routes are altered, and which built environment characteristics suppressed trips are associated with. To accomplish these goals, we identified the pertinent roadway characteristics for each roadway segment and used survey res ults to determine the percentage of trips that we would expect as origins and their closest school as the corresponding destination, deriving optimal trip ro utes through a closest facility network analysis. Then, we used these optimal routes to derive the upper limit of trips that would theoretically utilize each roadway segment. After accounting for the impact of network connectivity on walking levels, we f inally combined the trip suppression rates (from the

PAGE 37

28 survey) with the upper limit of trips (from the network analysis) for each roadway segment and used this to answer our research questions. Suppression Rates With logistic regressions, we created the trip suppression model derived from the parental perceptions survey to determine which factors are associated with trip suppression. We then examined grade level, gender, adult supervision and street level design characteristics (e.g. posted speed limit, sidewalks, bike lanes, number of lanes, and vehicle volumes) to establish their relationship with trip suppression rates. For differences in suppression between gender and grade level, we created an independent logistic regression model for each student p opulation subset (i.e. 5 th grade males, 7 th grade females, etc.). When exploring the impact of adult supervision, we accounted for all students in a single logistic regression model for walking and a single logistic regression for bicycling. We next deter mined what percentage of trips would be suppressed due to road safety concerns for different roadway scenarios. We coded all forty roadway scenarios featured in the survey based on their four predictor variables while designating the outcome variable as t he percentage of parents that would not allow their children to use the roadway. Since the diversity of actual Denver roads exceeded what we were able to reasonably include in the parental survey, we created a linear regression using the four roadway char acteristics as predictor variables and the percentage of disallowance as the outcome variable (Table 2). Using these linear regressions, we then took the street level design characteristics for each scenario that was not featured in the survey and derived the corresponding suppression rate.

PAGE 38

29 T able 2 . Linear Regression Coefficients for Trip Disallowance Derived from Survey Results Walk (R 2 = 0.9654) Bike (R 2 = 0.9227) Intercept 0.037 0.076 Speed (mph) 0.015*** 0.010*** Lanes 0.046* 0.105*** Facilities 0.248*** 0.086*** Volume 0.131*** 0.230*** * p<0.10 ** p<0.05 *** p<0.01 Network Analysis We then deduced the maximum number of trips that would occur under ideal conditions through a network analysis in GIS. We utilized all public elementary and middle schools in Denver for analysis. DPS does not provide busing for elementary students that live less than a mile from their school. To capture these populations that would be more apt to pursue active mod es of transportation, we created a Euclidian distance one mile buffer (i.e. an as the crow flies buffer instead of a network buffer) around each of the elementary and middle schools and designated this as the study area. The study area included the majori ty of Denver, except for the far northeast portion of the city comprised of the airport. After clipping the roadway centerlines to the study area, we removed any limited access highways and merged off road trails into the layer. All divided roadways were represented by one line instead of two. We avoided edge issues by including roadways in neighboring municipalities that fell within one mile of a DPS school and ensuring that all stray road segments were connected to the larger network. Finally, we clea ned access points around the schools so that students in the model could approach their school from the same side they would in reality. To account for explanatory roadway characteristics, we utilized speed limits and the number of lanes that were provide

PAGE 39

30 and amending any errors. We utilized vehicle volumes provided by DRCOG, with any roadway having more than 1,000 vehicles per day being designated as high volume (Cornell Local Roads roadways with or without a sidewalk. Google Maps, satellite imagery, and Google Street View were utilized to identify roadways with bike lanes. Roadways with no s idewalks/bike lanes were given a value of zero, roadways with a sidewalk/bike lane on one side were given a value of one, and roadways with a sidewalk/bike lane on both sides were given a value of two. Once the roadway network was complete, we accounted for origins and destinations. Destinations were defined as public DPS elementary and middle schools within the City and County of Denver. Origins were based on child populations from the 2015 American C ommunity Survey. For each Census block group located within the study buffer, we created one random point for each child living in that Census block group. Because the average block group used in the analysis had an area of 248 acres, we created random p oints only in residential zones so as to realistically represent home origins. While it would be ideal to know exactly how many attending children live within one mile of each school and the home address of those children, that information was not availab le due to privacy concerns. There were 112,648 children in Denver and 23,490 in neighboring municipalities included in the analysis. The number of children included in the analysis is higher than the number of children attending DPS schools because the analysis also included children living in Denver that attend private schools or are home schooled. Also, some children living in Denver that were included in the analysis may be attending schools in neighboring municipalities or may be too young to attend school. The largest of the 587 block groups had 3,326 children while 23 block groups had no children. Census block groups with no children consisted of either undeveloped land or land uses other than residential.

PAGE 40

31 We then ran a closest facility network analysis using GIS. This takes the origin (child) and finds the shortest route to their respective destination (school). While pedestrians and bicyclists often do not use the shortest available path because of road safety and comfort concerns, we wanted to start with a baseline of how many trips could occur under ideal conditions and then derive the number of suppressed trips based on that value (Krizek, El Geneidy, & Thompson 2007). Alternatively, if we derived the number of trips being suppressed beca use of safety concerns from a trip count weighted on traffic safety concerns, this would result in multicollinearity issues. the western edge of the Great Plai ns. Therefore, no elevations were factored into the network analysis. One way streets were not accounted for in the pedestrian analysis but were accounted for in the bicycling analysis. We integrated the impact of roadway network connectivity on active transportation levels into the analysis by accounting for intersection density (Schlossberg et al. 2006). We did not account for crime because of mixed findings in terms of the relationship between objective crime and walking levels, mainly due to more wa lkable environments attracting different types of crimes (Foster et al. 2014). The network analysis resulted in 136,138 routes, as all origin/destination pairs were successfully connected. We then derived the number of routes that utilized each roadway se gment within the City and County of Denver. This resulted in the total number of children that could be expected to walk or bike on each segment under ideal conditions, assuming that children would be walking or biking within a mile to their closest schoo l. Suppressed Trips Now that each roadway segment had a set of roadway characteristics, a corresponding percentage of parents that would not allow their child to walk or bike, and the total number of

PAGE 41

32 possible trips under ideal conditions, we multiplied th e number of ideal trips by the percentage of disallowance to derive the number of trips that would theoretically be suppressed because of traffic safety concerns. This was done for each roadway segment in the study area. Doing so allowed us to perform sp atial analysis to understand where there are high numbers of suppressed trips and which built environment characteristics are associated with those high numbers of suppressed trips. R esults We first utilized logistic regressions to examine the impact of d emographic and roadway employed linear regressions to explore the percentage of trips that could be expected to be suppressed due to traffic safety concerns for a variety of different scenarios. Finally, we integrated these trip suppression rates with the upper limit of modal frequencies derived through GIS network analyses to discover areas of Denver that have high levels of trip suppression and found associate d built environment characteristics. T rip Suppression Factors Typically, the roadway characteristics utilized in our models (i.e. posted speed limits, number of lanes, presence of active transport facilities, and vehicular volumes) would be collinear. Ho wever, because a wide range of roadway scenarios were chosen for the survey instrument, multicollinearity was avoided. A variance inflation factor (VIF) threshold of 1.46 for the variables signaled low multicollinearity. A VIF of 5.0 or above is typicall y indicative of multicollinearity issues (Vatcheva et al. 2016). When simply looking at parental allowance in a binary fashion (i.e. allowance both with and without adult supervision are categorized together) for all children captured by the survey, gender

PAGE 42

33 was not significantly related to parental allowance for either walking or bicycling, but the variable did strengthen the model based on a lower AIC (Table 3). Other than the number of siblings and the days of physical activity in the bicycling model, all other explanatory variables were significant. The presence of sidewalks was by far the strongest predictor for the walking model, with a 14.079 odds ratio being interpreted as parents being approximately 14 times more likely to allow their child to walk on a roadway with sidewalks present than a roadway without sidewalks, with all other factors held constant. walking trips to school (Evers et al. 2014). Vehicle volumes were t he strongest predictor for the bicycling model, followed closely by the presence of bicycle facilities (bicycle lanes). Parents are not as concerned with vehicle speed or volumes when considering walking, as long as children have a sidewalk outside of the traffic lanes. We utilized the grade variable by individual grade level, so a 1.078 odds ratio means that a 5 th grade student would be 1.078 times more likely to be allowed to walk than a 4 th grade student, with all other factors held constant. T able 3 . Parental Allowance Logistic Regression Odds Ratios Walk (R 2 = 0.338) Bike (R 2 = 0.223) Grade 1.078*** 1.178*** Gender (male=0 ) 0.940 0.896* Speed ( 10mph increments ) 0.402*** 0.581*** Lanes 0.751*** 0.576*** Facilities 14.079*** 2.502*** Volume ( low/high ) 0.445*** 0.297*** Number of Siblings 0.808*** 1.045 Days of Physical Activity 1.143*** 1.052 * p<0.10 ** p<0.05 *** p<0.01 We then took a more thorough look at these results in terms of grade level and gender. Results suggest that sidewalks remain the strongest predictor for walking suppression across all

PAGE 43

34 grade levels and genders (Figure 3). Posted speeds are typically the s econd strongest predictor for walking, followed by vehicle volumes. In terms of parental allowance for bicycling, we found vehicle volumes to be the strongest predictor for younger children. However, for some higher grade levels, bike lanes become the mo st important street level design characteristic. For bicycling allowance, posted speed limits and the number of lanes are the least important factors. There are no significant differences in terms of gender, as the results above suggest. There are notab le odds ratio increases in 3 rd grade, 4 th grade, and 5 th grade, possibly signaling this as a time when parents feel that children may walk and bike.

PAGE 44

35 F igure 3 . Odds ratios of street level design factors in relation to suppression by grade level and gender for walking (above) and bicycling (below) (hollow bars are not significant at 90%).

PAGE 45

36 When we further parse the results by examining how factors are correlated in terms of whether parents reported trusted adult supervision as required, findings sugges t that both types of allowance follow similar patterns (Table 4). Namely, the presence of sidewalks remains the most important factor in terms of walking allowance, followed by vehicle volumes and child grade level. In terms of biking allowance, vehicle volume remains the most important factor, followed by the grade level of the child. These results suggest that those factors identified as important in the models not accounting for supervision are also important (although less so) when supervision is acc ounted for, with grade level becoming a stronger predictor, as we would expect. T able 4 . Parental Allowance Logistic Regression Odds Ratios Walk (R 2 = 0.452) Bike (R 2 = 0.374) Grade 1.814*** 1.697*** Gender (male=0 ) 0.823** 0.880 Speed ( 10mph increments ) 0.642*** 0.629*** Lanes 0.863** 0.795*** Facilities 4.177*** 1.055 Volume ( low/high ) 0.533*** 0.447*** * p<0.10 ** p<0.05 *** p<0.01 Trip Suppression Rates Once we understand which demographic and street level design factors are related to trip suppression, we can begin to explore how many children in Denver are impacted by trip suppression and how that trip suppression impacts route choice. Since we did not detect any non linear r elationships, linear regression was used in this analysis. We created a linear regression (Table 2) with the results from the survey to derive a mode choice model that outputs the rate of trip suppression due to traffic safety concerns for a variety of di fferent roadway scenarios (Table 5). For instance, while parents reported that only 11.7% of walking trips would be suppressed on a low -

PAGE 46

37 volume, 25 mph, 3 lane road with sidewalks, parents reported that the same road without sidewalks would see 59.4% of wa lking trips suppressed (Figure 4). Parents reported that for a high volume, 25 mph, 3 lane road with sidewalks, 16.4% of walking trips would not be allowed. Here we can see that the change in facilities had a much larger impact on allowance than the chan ge in vehicle volume. Rate suppression for bicycling was generally higher than for walking, as parents were less willing to let their children bicycle to school. Presence of facilities can be seen to have a large impact for both walking and bicycling. T here is a substantial increase in trip suppression rates when going from 25 mph to 35 mph for bicycling, while there is a similarly substantial increase when going from 35 mph to 45 mph for walking. Also, vehicle volumes have a greater impact on trip supp ression for bicycling than for walking. T able 5 . Percentage of Trips Suppressed Based on Survey and Linear Regression (Variables Held at 3 Lanes, Presence of Facilities, and Low Volume; Survey Values in Bold ) 25 mph 35 mph 45 mph Walk Bike Walk Bike Walk Bike Lanes 2 5.1% 23.8% 14.5% 33.1% 31.6% 41.2% 3 11.7% 28.7% 25.6% 42.0% 36.2% 51.7% 4 10.5% 42.2% 20.1% 52.2% 43.3% 67.9% Facilities No 59.4% 52.3% 65.8% 63.1% 85.7% 68.9% Yes 11.7% 28.7% 25.6% 42.0% 36.2% 51.7% Volume Low 11.7% 28.7% 25.6% 42.0% 36.2% 51.7% High 16.4% 49.8% 32.4% 64.0% 49.2% 74.7%

PAGE 47

38 Figure 4 25 mph 3 Lanes Sidewalks Low Volume 11.7% of child walking trips will not be allowed according to parent responses. 25 mph 3 Lanes No Sidewalks Low Volume 59.4% of child walking trips will not be allowed according to parent responses. 25 mph 3 Lanes Sidewalks High Volume 16.4% of child walking trips will not be allowed according to parent responses.

PAGE 48

39 F igure 4 . Examples of parent reported trip disallowance for different walking and biking roadway scenarios. 25 mph 3 Lanes Bike Lanes Low Volume 28.8% of child biking trips will not be allowed according to parent responses. 25 mph 3 Lanes No Bike Lanes Low Volume 52.3% of child biking trips will not be allowed according to parent responses. 25 mph 3 Lanes Bike Lanes High Volume 49.8% of child biking trips will not be allowed according to parent responses.

PAGE 49

40 We next identified to what extent children are encountering these roads with high levels of disallowance. More children in Denver encounter roads that parents have deemed unsafe for bicycling than encounter roads that parents have deemed unsafe for walking. Approximately 2.3% of children in Denver would encounter a road with 75% disallowance or greater for walking (a road that is perceived as particularly unsafe), assuming that they take the shortest route to school (Table 6). However, 31.8% of children in Denver would encounter a road with 75% disallowance or greater for bicycli ng. For roads with 50% disallowance (still perceived as relatively unsafe), more than half of children in Denver would encounter those roads for a bicycling trip and 12% would encounter them for walking trips. This converts to tens of thousands of trips each day, showing that the issue is widespread throughout Denver. T able 6 . Percentage of Children Encountering Roads with Varying Disallowance Rates 25% Disallowance 50% Disallowance 75% Disallowance Walk 40.5% 12.2% 2.3% Bike 64.9% 61.4% 31.8% Children that encounter roads that are perceived as unsafe for walking specifically those roads which more than half of parents would not allow their child to walk on are primarily found in two different areas: an area with sidewalk gaps near the borde r of the Montclair and East Colfax neighborhoods of Denver (the top right concentration) and an area near the border of the Mar Lee and Ruby Hill neighborhoods (the bottom left concentration) (Figure 5). This first neighborhood has high numbers of childre n that encounter roads perceived as unsafe because of gaps in the existing sidewalk network while the second neighborhood has high concentrations of children near Federal Boulevard and Florida Avenue, roads that are perceived as unsafe. Children that enco unter roads that are perceived as unsafe for bicycling are similarly found near Federal Boulevard, but also

PAGE 50

41 in the Montbello neighborhood, which contains high concentrations of children, a curvilinear roadway network, and a lack of bicycle facilities. It is apparent that, in general, there are more children that encounter roads that parents would not allow them to bicycle on than children that encounter roads that parents would not allow them to walk on. The neighborhoods that had the highest number of ch ildren who encounter roads perceived as unsafe have median household incomes that are 6.2%, 15.1%, and 46.7% below average for Denver.

PAGE 51

42 F igure 5 . Densities of children with negatively impacted routes for walking (top) and bicycling (below) (red represents higher densities of impacted routes). However, if a grid network is in place that presents pedestrians and bicyclists with different route options, children may be able to simply avoid these roads that are perceived as unsafe by using para

PAGE 52

43 not utilized? To answer this inquiry, we reran the network analysis after the roads that are perceived as unsafe particularly those with greater than 50% d isallowance were weighted so that they would be discouraged according to parental traffic safety concerns. We then compared the trip lengths under ideal conditions to those when roads perceived as unsafe were accordingly discouraged. For walking trips, the average trip length across the city increased from 2,728 feet under optimal conditions to 2,937 feet once roads perceived as unsafe were discouraged. For bicycling trips, the average trip length increased from 2,728 feet under optimal conditions to 3, 763 feet once roads perceived as unsafe were discouraged. Citywide, about 4,274 children were pushed out of a ½ mile walkshed when road safety perceptions were accounted for, while 23,429 children were pushed out of a ½ mile bikeshed. The greatest increa ses in distance were concentrated near Interstate 25, the South Platte River, and Sheridan Boulevard. These neighborhoods have curvilinear tributary roadway networks or limited route options because of barriers, resulting in large increases in trip distan ce (upwards of an additional 5,228 feet to avoid roads perceived as unsafe in the curvilinear tributary neighborhood). Areas with grid networks that saw large percentages of roads perceived as unsafe did not have similarly large increases in trip distance because of the ability for pedestrians and bicyclists to select alternate routes with little additional distance. Number of Suppressed Trips We then integrated the results from the mode choice model with the number of total possible trips to derive the n umber of trips that are suppressed due to road safety concerns for roadways in Denver and identified which roadways have the most suppressed trips (Table 7). We focused on roads with greater than 25% disallowance because these roads are perceived as unsaf e and are most likely in need of amendment. While some roads with less than 25% disallowance had

PAGE 53

44 high numbers of suppressed trips, it was only because of high levels of ideal exposure, not because of a lack of perceived safety. Therefore, these roads wer e not considered. The high number of low road segment. We utilize spatial analysis techniques in the Discussion to explore where these road segments wit h high numbers of suppressed trips are located. T able 7 . Number of Suppressed Trips per Road Segment Min Max Mean SD Walk 0 272 1.60 5.90 Bike 0 528 7.11 20.59 D iscussion This section presents a spatially oriented, rather than statistical, analysis of where roadways are located with high numbers of suppressed trips. Walking trip suppression and bicycling trip suppression displayed similar spatial patterns within the study area. We found high numbers of suppressed trips primarily either near a connection through a barrier in the roadway network or near a school. The barrier connections were at impediments that have limited pathways over or under them (e.g. limited access h ighways or bodies of water), for which high rates of trips would optimally funnel through the few available connections. Because these barrier connections primarily serve vehicles, they are usually wide, high speed roadways. Therefore, while such connect ions are vital to both motorists and non motorists, the connections are often built to accommodate vehicles and present non motorists with an option that is many times perceived as unsafe. Examples of these barrier connections include Lincoln Street where it goes under I 70 (25 mph, two sidewalks, no bike lanes, high volume, two lanes; 14.3% walking disallowance & 61.4% biking disallowance; 14 suppressed walking trips & 41 suppressed biking trips) and Dunkirk Street as it goes across First

PAGE 54

45 Creek (30 mph, t wo sidewalks, no bike lanes, high volume, four lanes; 31.1% walking disallowance & 87.4% biking disallowance; 99 suppressed walking trips & 286 suppressed biking trips) (Figure 6). It is also important to note that these numbers of suppressed trips are ba sed on each individual to/from trip to school. If we multiply the number of suppressed trips by 180 annual school days and by two for the departing and returning trips, 14 suppressed trips is equivalent to 5,040 annual suppressed trips. We found high num bers of suppressed trips at barrier connections to be rare because children often have schools within their own neighborhoods that are closer than those on the other side of the barrier. F igure 6 . Examples of barrier connections with high trip suppression (Central Park Boulevard over Prairie Meadows Drive on the left; Dunkirk Street over First Creek on the right).

PAGE 55

46 There were also high rates of trip suppression on roadways near schools. These roadways can be broken into two categories: those in tributary curvilinear loop networks and those in grid networks. While the majority of Denver consists of gridded networks, three of the top six areas for suppressed walking trips and two of the top six areas for suppressed biking trips near schools are f ound in the Montbello neighborhood of northeast Denver. Montbello has a predominately tributary curvilinear loop roadway network, as the bulk of its development occurred in the late 1960s and early 1970s. These networks consist of curved local streets th at channel onto collector or arterial streets. While pedestrians and bicyclists in grid networks have the option to use a variety of roadways to get to their destination, non motorized users in these curvilinear loop networks typically must channel to a m ain road that has been prioritized for vehicles. The four most extreme cases of suppressed trips in the tributary curvilinear network in Montbello are Andrews Drive (46.3% walking disallowance & 82.4% biking disallowance; 51 suppressed walking trips & 37 4 suppressed biking trips), Maxwell Place (54.8% walking disallowance & 87.4% biking disallowance; 135 suppressed walking trips & 215 suppressed biking trips), Gateway Avenue (46.3% walking disallowance & 82.4% biking disallowance; 75 suppressed walking tr ips & 133 suppressed biking trips), and 46 th Avenue (46.3% walking disallowance & 82.4% biking disallowance; 272 suppressed walking trips & 484 suppressed biking trips) (Figure 7). While the roads in these neighborhoods may not be perceived to be as dange rous as some in central Denver (the roadways have two sidewalks, do not have bike lanes, are signed at 25 mph or 30 mph, have four lanes, and are high volume), the high trip suppression rates in these northeast neighborhoods are being driven by the fact th at trips are concentrated on these main roads because of the street network configuration.

PAGE 56

47 F igure 7 . Examples of high trip suppression around a school in a curvilinear loop network (Blue dots are schools; image is of Andrews Drive). Relatively few areas with high trip suppression were found in the grid network, which is predominant across Denver. High trip suppression roadways that were found in the grid network

PAGE 57

48 typically occurred when a school was placed on or near a major road. Th is scenario occurred outside of the grid network as well. Examples include S. Sheridan Boulevard (71.8% walking disallowance & 97.4% biking disallowance; 127 suppressed walking trips & 421 suppressed biking trips) and S. Monaco Parkway (54.8% walking disa llowance & 87.4% biking disallowance; 80 suppressed walking trips & 314 suppressed biking trips) (Figure 8). While pedestrians and bicyclists in a grid network typically have options in regard to which roads they utilize, siting a school directly on an ar terial can force them to use an unsafe road, thereby dissuading walking or biking trips. F igure 8 . Examples of high trip suppression around schools near major roads (Sheridan on the left; Monaco on the right).

PAGE 58

49 C onclusion By combining trip suppression rates derived from a perception survey with the upper limit of trip frequencies from a GIS network analysis, our tool allows us to identify areas where trips are suppressed because of road safety concerns. Although identified areas are not technically objectively unsafe, only perceived as unsafe, identifying such areas will hopefully aid in the identification of road safety issues before they occur. We find that, for all children both with and without adult supervision, side walks are the strongest predictor in terms of walking allowance and vehicle volumes are the strongest predictor in terms of biking allowance. Parents may be more concerned with the presence of vehicles for biking rather than walking because their child wo uld likely be biking in the street. Gender of the child has a weak relationship with allowance. When we parse the results by age, sidewalks consistently remain the most important factor for walking allowance, but bike lanes are found to become more impor tant for higher grade levels in terms of biking allowance. When we look at the role adult supervision plays, we see sidewalks (for walking) and vehicle volumes (for biking) remaining the most important factors (although losing importance), while grade lev el becomes a stronger predictor. When looking at the pervasiveness of these issues, we find that over 61% of children encounter a road perceived as unsafe (50% or greater disallowance) for biking and over 12% encounter a similar road for walking. This ind icates that the problem is prevalent across Denver. neighborhoods experienced large increases in the distance that must be travelled, with children in some n eighborhoods needing to add an additional mile onto their route to avoid such roads. Areas with high numbers of suppressed trips were heavily concentrated around schools in parts of the city with curvilinear loop and tributary networks. Grid networks see m to help to

PAGE 59

50 alleviate high numbers of suppressed trips, provided that the school is not sited on or very near a major road. There are also high numbers of suppressed trips present at barrier connections. The segments of high trip suppression are typical ly not great in length, meaning that it may only take one or two blocks of conditions perceived as unsafe to dissuade a pedestrian or bicycle trip from occurring. Primary limitations of the work are focused on the origin and destination location of the s chool trips employed in the model. Because of privacy issues, we could not account for the actual trip of each student, and instead assumed that children would be most likely to attend their closest school. This is an assumption we know to be imperfect b policy. While this may have impacted the implementation of the tool, the theoretical methodology developed is still sound. Future work could improve upon the current approach by examining areas with clearer origins an d destinations. Furthermore, the number of parent respondents that took the survey in Spanish was low (3.6%) relative to the number of reported Spanish speaking students in DPS (36.8%). While we believe that some Spanish speakers took the survey in Engli sh, specifically concentrating on these populations in future efforts may result in more representative outcomes. The decision to allow a child to walk or bike is typically influenced by a combination of street design variables. Future work may explore di fferent explanatory variables utilized for the model. While we used more explanatory roadway characteristic variables than past studies (Cho, Rodriguez, & Khattak 2009; Schneider, Ryznar, & Khattak 2003; Nevelsteen et al. 2012), roadways are complex, and more variables may lend further strength to our models. For the sake of this work, we used a combination of variables including the number of lanes, posted speed limits, and vehicle volumes as an indicator of the exposure to risk a child would encount er when crossing a road. However, in terms of crossings, there are other factors (e.g. signalization, phasing, crosswalks, medians, etc.) that would also be important to account for. Sidewalk conditions also vary, and these

PAGE 60

51 varying conditions would be im portant to include in future models. Specifically for Denver, many neighborhoods have two foot wide sidewalks for which parents may have varying perceptions of road safety relative to current five foot wide sidewalk standards. Furthermore, examining the impact of actual vehicle speeds instead of simply assuming signed speed limits are reflective of actual conditions may improve future models. The vehicle volume factor could account for whether peak hour crests occur during times when children would be ex pected to be walking or biking on each road. Finally, future work may also account for school zones, which could improve traffic safety perceptions near schools. School zones in Denver are typically signed at 20 mph, but major roads use speed limits of u p to 40 mph in school zones, highlighting the bias towards motor vehicles and against non motorized trips. In terms of macro scale perspectives of the work, future analysis could account for varying land uses. While we were only concerned with trips to a nd from school, and it was therefore appropriate to only account for this one specific land use (Ewing, Schroeer, & Greene 2004), more holistic future examinations would be wise to account for the presence of other land uses and destinations. The impact o f crime on levels of walking and biking could also be accounted for but would necessitate a thorough examination of the types of crime occurring relative to the land uses throughout the study area. For example, while violent crimes in residential neighbor hoods may dissuade walking and biking, high rates of shoplifting may signal a strong commercial area that may have high levels of walking and biking (Foster et al. 2014). Such a thorough analysis of crime in Denver was outside the purview of this work. T his aspect of the work also hints at equity issues, namely that lower income or minority neighborhoods may have to more frequently deal with both crime and traffic safety issues, more so than their more affluent counterparts. The neighborhoods that had th e highest number of children who encounter roads perceived as unsafe were found to have median household incomes that are below average for Denver.

PAGE 61

52 Future work could see integration of this trip suppression tool into a proactive safety analysis by similarl y identifying areas with high trip suppression. Results from such a proactive analysis could be compared to a traditional reactive safety analysis. Outcomes from the two analyses would be expected to vary. Hopefully, we would be able to identify areas w ith high rates of trip suppression but low objective crashes or injuries. These would be areas with road safety issues that dissuade non motorized users enough to preclude objective outcomes, and therefore have thus far been neglected. It is recommende d that planners and engineers utilize such analysis approaches and also deploy recommendations from this work, namely employing grid networks, siting schools and other locations that children may be expected to frequent on more minor roads, and ensurin g that there are pedestrian and bicycle facilities present where there are vital connections across barriers. Pedestrian and bicycle trips that have been suppressed because of traffic safety concerns can be an important indicator of road safety in our tran sportation systems. The tool developed in this paper allows for the identification of roadways with high levels of suppressed trips in terms of street level design characteristics. This approach allows for the methodology to be applied widely, enabling u tilization by academics and practitioners alike. Through the application of this tool in Denver, we identified important personal and design characteristics that act as predictors of trip suppression, as well as the importance of grid networks, barrier co nnections, and destination siting. By identifying these areas with high numbers of suppressed trips, and by enabling others to do the same, we have facilitated the proactive identification of traffic safety issues on our roadways.

PAGE 62

53 CHAPTER IV SUPPRESSED CHILD PEDESTRIAN AND BICYCLE TRIPS AS AN INDICATOR OF SAFETY: ADOPTING A PROACTIVE SAFETY APPROACH IN DENVER Introductio n Traditional pedestrian and bicyclist safety analyses typically examine crashes, injuries, or fatalities. However, this reactive appr oach only accounts for the places where people are currently walking or biking and those who are doing so. Would a proactive approach examining areas where pedestrian and bicyclist activity is being suppressed because of safety concerns illuminate oth er previously neglected safety issues? The goal of this work is to compare results from reactive and proactive pedestrian and bicyclist safety analyses in Denver, Colorado. To accomplish this, we focus on child pedestrians and bicyclists because of the st ructured characteristics of their trips to school. We utilize conventional reactive analyses from Denver Public Works and the Denver Regional Council of Governments as well as our own cluster analysis of crashes. We then complete a proactive safety appro ach based on the number of trips that are suppressed due to traffic safety concerns. A parental perception survey forms the basis of the mode choice model we created to perform the proactive safety analysis. Findings suggest that reactive approaches ident ify downtown Denver and major corridors as unsafe, while the proactive analysis identifies neighborhoods in west, east, and northeast Denver. Due to an absence of crashes, the majority of these areas would not normally be considered unsafe for pedestrians and bicyclists based on the conventional approach. However, the fact that they are perceived as unsafe may be limiting usage and thereby limiting the number of crashes. In order to improve safety both where children are walking and bicycling as well a s where they want or need to walk or bike traditional analyses would benefit from augmentation by such a proactive safety approach.

PAGE 63

54 Researchers traditionally analyze crashes, injuries, or fatalities when examining traffic safety of walking and bicycling trips. However, the only people that are accounted for in this reactive approach to safety are those who are already walking or biking the people who have decided that those activities are safe enough to pursue. What about the people who because of traffic safety concerns have decided to not walk or bike in the first place? Furthermore, the only locations that are identified in such reactive approaches are the locations that have pedestrians and bicyclists present. What about the places where peo ple have decided not to walk or bike? How would we proactively identify these places as unsafe before a crash occurs or even before any walking or biking occurs? Past research has investigated how perceptions of traffic safety impact the choice to walk or bike. We propose building upon this theory by quantifying the number of walking and bicycling trips that are being suppressed due to traffic safety concerns and using this as an indicator of traffic safety risk. If high levels of pedestrian and bicycle trips are being suppressed because of traffic safety concerns, this suggests that there are traffic safety issues present, regardless of whether crashes are occurring. We then compare results from such a proactive safety analysis to results from tradition al reactive analyses. To accomplish these objectives, we reactively and proactively analyze the safety of walking and biking trips of children to and from school in Denver, Colorado. Children are some of the most vulnerable road users, who depend heavily on walking and biking. Their trips to school are highly structured, making this an ideal group to study. For the reactive approach, we use methodologies and results from existing safety reports focused on pedestrians and bicyclists of all ages to create our own crash cluster analysis specific to child pedestrians and bicyclists. To derive the number of child trips to school being suppressed because of safety concerns for the proactive approach, we combine trip suppression rates from a parental perception s survey with shortest path

PAGE 64

55 distances to derive safety perception weighted distances. Finally, we compare results from this proactive analysis of trips suppressed due to safety concerns to results from the traditional reactive analyses of crashes. If our goals include promoting walking and biking activity instead of just reducing crashes, such a proactive safety analysis may provide an important perspective on pedestrian and bicyclist safety. T heory Traditional safety analyses rely on crashes, injuries, or fatalities to identify unsafe roadways (Transportation Research Board 2001; Waldheim, Wemple, and Fish 2015). This approach is utilized for safety studies of vehicles as well as pedestrians and bicyclists (Federal Highway Administration 2006; Zegeer, N abors, Gelinne, Lefler, and Bushell 2010). Ideally, these crash based safety analyses account for exposure situations involving motor vehicles in the form of distance or time traveled, user c ounts, times crossing a street, or the product of pedestrian or bicyclist and vehicle volumes (Molino et al. 2012). However, the lack of reliable pedestrian and bicyclist exposure data makes such comparisons difficult (Turner et al. 2017). In the absence of appropriate exposure data, cluster analyses of crashes are a common approach used to identify areas of safety concern (Blackburn et al. 2017). The four most recent pedestrian and bicyclist safety analyses for Denver, Colorado that were completed by loc al and regional transportation agencies consisted of traditional crash based analyses that did not account for exposure (Denver Public Works 2016; Denver Public Works 2017; DRCOG 2012; DRCOG 2017). While such a focus on crashes can allow for the successfu l identification and reduction of those crashes, it is by nature a reactive approach that requires that a crash occur before any safety issues can be identified. Because a pedestrian or bicyclist crash cannot

PAGE 65

56 take place unless there is a pedestrian or bic yclist present, this conventional approach only accounts for people who are currently walking or biking and the places where they are doing so. A proactive approach to safety is one that identifies areas that are likely to experience crashes before those c rashes occur. It could also be one that identifies areas where safety concerns have caused low levels of pedestrian and bicyclist activity, and therefore effectively hides safety issues from the objective eye. For instance, proactive approaches can be eff ective when exposure levels and therefore crash levels are so low that traditional indicators do not provide an accurate representation of risk (Cho, Rodriguez, and Khattak 2009; Schneider, Ryznar, and Khattak 2003). Perceptions of safety have been sh own to correlate with exposure levels and may therefore proactively indicate safety issues (Noland 1995; Pucher, Dill, and Handy 2010). In other words, while a road perceived as unsafe may suppress walking and biking trips and therefore reduce or preclude pedestrian and bicycle crashes, these same negative safety perceptions can be used as an indicator that low levels of exposure have hidden safety issues. Past researchers have taken a number of approaches when using perceptions of safety to proactively id entify pedestrian and bicyclist safety issues. An early attempt surveyed pedestrians and drivers on the campus of the University of North Carolina at Chapel Hill, asking them to identify locations that posed safety issues to pedestrians (Schneider, Ryznar , and Khattak 2003). When the researchers compared these subjective perceptions to objective outcomes, it became apparent that there were areas perceived as unsafe that had no crashes occurring. Researchers determined that, while these areas were otherwi foundation for our work, the method of identifying unique hotspots is not scalable. In other words, because the perceptions were not tied to specific characteristics of the built environment, the survey

PAGE 66

57 would need to be re administered for every new area that may be studied in the future. We seek a methodology that can be generalized and ap plied to other areas. Cho et al. (2009) improved upon this scalability issue by creating a risk estimate based on built environment characteristics such as land use mix and street connectivity. They found that increased perceptions of risk for pedestrians and bicyclists reduced crash rates because of decreased usage. However, the methods did not account for street level risk factors such as roadway width, vehicle speeds, and vehicle volumes, which could have an impact on behavior. This approach and its r esults have limited implications for built environment improvements as road network connectivity and land use mixes are not as easily addressed as vehicle volumes, vehicle speeds, or sidewalk gaps. Nevelsteen et al. (2012) built upon these trip suppression studies by examining the relationship between perceptions of street level risk factors and pedestrian and bicyclist trip allowance for children traveling to school. However, the researchers only examined two factors: the presence of pedestrian and bicycl e facilities and vehicle speeds. Furthermore, the study was performed in a Belgian context that reports 40% of 11 and 12 year olds cycling to school a transportation culture that is vastly different from that of much of the world. Importantly, all of th e studies proactively examining safety perceptions showed that perceptions of safety and trip suppression differ from objective safety outcomes such as crashes. Specifically, Cho et al. (2009) and Schneider et al. (2003) found that perceptions of unsafe c onditions lowered use and therefore exposure, which improved objective safety outcomes, thereby hiding safety issues. However, these past proactive analyses were either focused on small areas, did not include all applicable roadway variables, or were in u nfamiliar contexts. We therefore create a proactive model that quantifies trip suppression for an entire American city based on all applicable roadway risk factors and then compare results to those from more conventional reactive crash based

PAGE 67

58 analyses. We hypothesize that the reactive and proactive analyses will illuminate different safety concerns and complement one another. are an important focus because they are some of the most vulnerable road users and they rely on walking and biking to travel independently, they also have relatively structured trips that allow for more accurate modeling. This provides us with a foundation for our approach and allows us to avoid needing to model every pedestrian and bicycle trip throughout a city for every age and trip purpose. D ata The goal of this paper is to compare reactive crash based pedestrian and bicyclist safety analyses (using existing reports for pedestrians and bicyclists of all ages as a guide to form a crash cluster analysis specific to children) to a proactive analysis (based on where child trips are being suppressed because of safety concerns) that we develop for Denver, Colorado. Data for the reactive a nd proactive safety analyses are detailed below. We utilized the City and County of Denver for our analysis. Denver has a population of 663,303 according to the 2016 American Community Survey. Because Denver has a variety of pedestrian and bicycle specif ic safety reports along with detailed crash data available and high enough population, pedestrian, and bicyclist levels to get significant samples, it was an ideal location for of our study. While there are suitable levels of walking and biking for this st udy, there are also pedestrian and bicyclist safety issues present. According to Denver Public Works (DPW) reports and DRCOG data, there were 1,508 pedestrians and 1,083 bicyclists hit by motor vehicles in Denver between 2011 and 2014 (Denver Public Works 2017). These collisions resulted in injuries for 918 pedestrians and

PAGE 68

59 791 bicyclists, with 72 pedestrians and seven bicyclists killed (Denver Public Works 2016; Denver Public Works 2017). These are pressing safety issues that should be addressed. Reacti ve Analyses We utilized existing pedestrian and bicyclist safety reports along with our own crash cluster analysis for the reactive crash based portion of the study. The existing safety reports focus broadly on pedestrians and bicyclists of all ages, whil e the focus of this research was on child pedestrians and bicyclists. Therefore, we supplemented these existing safety reports with results from our own child pedestrian and bicyclist crash cluster analysis. Existing Safety Reports There were two pedestrian and bicyclist safety reports published by Denver Public Works and two pedestrian and bicyclist safety reports published by the Denver Regional Council of Governments (DRCOG) the regional Metropolitan Planning Organization that analyzed pedes trian and bicycle safety and were considered in this study (Denver Public Works 2016; Denver Public Works 2017; DRCOG 2012; DRCOG 2017). All four of the pedestrian and bicyclist safety reports utilized conventional reactive crash based analysis methodolog ies. None of these agency produced reports accounted for levels of exposure. The 2016 Denver Public Works report considered bicyclist crashes that occurred between 2008 and 2012, while the 2017 Denver Public Works report considered pedestrian crashes that occurred between 2011 and 2015. Crashes were defined as motor vehicle collisions that involved a bicyclist or pedestrian and were obtained from the Denver Police Department. Crashes included only those that were reported by the police. The reports noted that there were likely unreported pedestrian and bicycle crashes due to the involved parties or the police not understanding the need

PAGE 69

60 to report or not wanting to report their crashes that were not considered in the analysis (Denver Public Works 2016; Denver Public Works 2017). Due to a lack of pedestrian and bicyclist count data, exposure was also not accounted for in these studies. The DRCOG reports similarly examined pedestrian and bicyclist crashes for all ages without accounting for exposure (DRCO G 2012). The DRCOG report identified pedestrian and bicyclist crash hotspots for the entire Denver region for the years between 2003 and 2007. A pedestrian crash was defined as a collision between a motor vehicle and a person on foot (including persons i n wheelchairs, skateboards, rollerblades, or other small personal conveyance devices), while a bicyclist crash was defined as a collision between a motor vehicle and a non motorized pedal cyclist (DRCOG 2012). Crashes were pulled from the Colorado Departm database. DRCOG also completed an updated pedestrian and bicyclist safety report in 2017 that similarly utilized a reactive crash based approach, but because it did not identify hotspots, it was not used in this analysis (DRC OG 2017). Crash Cluster Analysis To supplement the reactive crash based analyses from Denver Public Works and DRCOG, we performed our own crash cluster analysis of pedestrian and bicyclist crashes in Denver that is specific to children. Crashes were pull format for the years 2010 2014. We included any crash involving a pedestrian or bicyclist under the age of 14 in our analysis in order to match the proactive analysis survey. We did not account for expos ure, as reliable counts of child pedestrians and bicyclists were not available.

PAGE 70

61 Proactive Analysis The goal of the proactive safety analysis was to quantify the number of child pedestrian and bicyclist trips that were suppressed because of traffic safet y concerns and utilize this as an indicator of safety. There were two aspects of the proactive safety analysis: 1) suppression rates based on survey derived parental perceptions of roadway characteristics; and 2) shortest path distances to school based on a GIS closest facility network analysis. For the suppression rates, we administered a survey to parents of children enrolled in pre kindergarten through 8th grade in Denver, Colorado, asking them which roadway characteristics they would allow their child to walk or bike on. Roadway characteristics included number of lanes, posted vehicle speed limits, vehicle volumes, and the presence of sidewalks and bike lanes. The survey excluded parents of high school students because high school students have more i ndependence than younger students and would be more likely to drive themselves or carpool with a friend. The survey was offered exclusively online and was marketed through newsletters, fliers, and social media by Denver Public Schools, parent teacher orga nizations, the City and County of Denver, and local advocacy groups. The survey was open for one month in 2017 and resulted in responses for 1,331 children. To derive the shortest path distances to schools, we utilized a closest facility GIS network analy sis with child home locations as origins and schools as destinations. We approximated child origin locations on the block group level by creating a random point for each child with population numbers coming from the 2015 American Community Survey on the N ational Historical Geographic Information System (NHGIS) (Manson et al. 2017). These origin points were clustered Data Catalog in polygon shapefile format. Sc hool destinations were in point shapefile format from

PAGE 71

62 the analysis. forma t. This layer included posted speed limits and the number of lanes for each road segment. We utilized vehicle volumes provided by DRCOG, with any roadway having more than 1,000 vehicles per day being designated as high volume (Cornell Local Roads Program 2014). We noted the shapefile format. We accounted for bike lanes based on their location per Google Maps, satellite imagery, and Google Street View and acc ounted for the off road path network based on a layer provided by the Denver Open Data Catalog. M ethods The goal of this paper is to compare a proactive pedestrian and bicyclist safety analysis based on trips suppressed because of traffic safety concerns to a reactive safety analysis. In order are structured and more readily modeled than adult walking and bicycling trips. However, existing cra sh based analyses examined pedestrians and bicyclists of all ages. Therefore, we identified locations with high rates of child pedestrian and bicyclist crashes, just as the Denver Public Works and DRCOG analyses had done for all age groups. We show that r eactive analyses for both children and all ages provide similar results, and then compare both to our proactive analysis.

PAGE 72

63 Reactive Analyses For the reactive analyses, we first utilized existing crash based pedestrian and bicyclist safety reports. Sinc e these studies were for all ages, we complemented their results with our own crash cluster analysis of child pedestrian and bicyclist crashes. Existing Reports The first reports forming the basis of our reactive analysis were produced by Denver Public Wo rks. These pedestrian and bicycle safety reports identified locations that experienced the most pedestrian and bicyclist crashes and then make recommendations for engineering, education, enforcement, and evaluation treatments. The reports included maps a nd tables identifying hotspots and corridors where the highest number of pedestrian and bicyclist crashes occurred. Hotspots were exclusively at intersections. The corridors depend on roadway length as a form of exposure to generate a rate of crashes per mile. The reports examined pedestrian crashes and bicyclist crashes separately, but for all ages. The second report forming the basis of our reactive analysis was created by DRCOG. It similarly located all age pedestrian and bicyclist crash hotspots (ex clusively at intersections) and corridors with high levels of pedestrian and bicyclist crashes. These corridors also rely on roadway length to generate the number of pedestrian and bicyclist crashes per mile. In addition to reporting the location of pede strian and bicyclist crash hotspots, the report explored environmental and operational characteristics of the crashes. The report also made recommendations for treatments in terms of engineering, enforcement, and education, although not specific to any in dividual hotspot. We took the results from these existing pedestrian and bicyclist crash analyses and imported them into GIS. We assigned one point for each hotspot and one line for each high crash corridor.

PAGE 73

64 We took the total number of crashes at these l ocations and calculated the number of crashes per year. Crash Cluster Analysis While the proactive safety analysis was designed to examine child pedestrians and bicyclists, the existing pedestrian and bicyclist safety reports examined users of all ages. In order to examine spatial patterns of child crashes, we pulled all child pedestrian and bicycle crashes that occurred relationship between these child crashes to identify statistically significant crash clusters. We then explored the spatial layout of the crashes in these clusters to identify the geographic extent of the deviational ellipses. istically significant clusters. The tool functions to identify both hotspots and cold spots in incident data. When inputting the data, we considered each crash point as a single equally weighted incident. The tool automatically aggregates all incidents that are found to be clustered into a mean centroid point and outputs the Getis Ord Gi* statistic in the form of a confidence level bin for each identified cluster. This statistic indicates the statistical significance of the spatial cluster. We consider ed clusters that were in the 95% confidence bin (two standard deviations for normally distributed data) and had at least three incidents in this study. Once these aggregated clusters were identified, we returned to the original crash point layer and assig ned each crash to its appropriate cluster (Figure 1).

PAGE 74

65 Figure 1 . Cluster analysis methodology: A) Input all applicable crashes; B) Utilize Optimized Hotspot Analysis tool to locate cluster and query statistically significant clusters (in red); C) Utilize Direction Distribution tool to form standard deviational ellipses around significant clusters. deviational ellipses around the identified clusters. Standard deviational ellipses measure the dispersion and orientation of the crashes ar ound the mean center of the cluster by using the standard deviation of the distance between the mean center and the x and y coordinates to define of two s tandard deviations that in two dimensional space have been shown to correlate to 95% confidence in normally distributed data (Esri 2018). With 95% confidence, we can say that the space enclosed by the ellipses are within a crash cluster.

PAGE 75

66 Proactive Analysis While we found the locations of crash hotspots and high crash corridors for pedestrians and bicyclists of all ages and paired those with the clusters of child pedestrian and bicyclist crashes, we hypothesize that examining only crashes neglects ch ild pedestrian and bicyclist safety issues namely, places that are perceived as so unsafe they have reduced levels of exposure. To address these gaps, we measured the number of child pedestrian and bicyclist trips that are suppressed because of safety c oncerns. This forms our proactive measure of safety. To accomplish this, we first administered a parental survey to quantify how perceptions of roadway characteristics influence GIS network analysis to find the shortest path distance between the estimated home location of each child in Denver and their closest school. Finally, we applied the trip suppression rates to the road network and ran the network analysis once again. Thi s allowed us to identify children who are within an appropriate walkshed or bikeshed under ideal conditions but are forced to travel beyond their walkshed or bikeshed because of safety concerns. This is what we define as our suppressed trips. A survey of parents of children in elementary or middle school provided us with the rate at 1,298 survey respondents provided us with 924 complete responses accountin g for 1,331 children. The survey was based on the methodology developed for the Denver Perceptions of Safe Routes to School Survey and was available for one month exclusively online in both English and Spanish (Ferenchak & Marshall 2018). The survey pres ented parents with a variety of scenarios consisting of varying roadway design characteristics and asked whether they would allow their child to walk or adult walking questions and five randomly selected bicycling questions from a pool of twenty walking

PAGE 76

67 questions and twenty bicycling questions. Parents provided information r egarding child age, gender, and physical activity levels. Roadway design characteristics included number of lanes (2, 3, or 4 lanes), posted speed limits (25 mph, 35 mph, or 45 mph), the presence of sidewalks and bike lanes (none or on one or both sides), and vehicle volumes (low or high volumes).

PAGE 77

68 F igure 2 . Example of survey questions.

PAGE 78

69 Since the number of roadway scenarios in Denver exceeded the number of scenarios that we could feasibly include on the survey, we created linear regressions and used them to derive a suppression rate for each roadway in Denver (Table 1). We accomplished this by coding all forty roadway scenarios featured in the survey by their four predictor variables and designating the outcome variable as the percentage of parents that would not allow their children to use the roadway. Using these linear regressions, we then utilized the street level design characteristics for each scenario that was not featured in the survey and derived the corresponding suppression rate. More de tails on the survey and the trip suppression model can be found in Ferenchak & Marshall (2018). Table 1 . Linear Regression Coefficients for Trip Suppression from Parental Survey Walk (R 2 = 0.9654) Bike (R 2 = 0. 9 22 7 ) Intercept 0.037 0 . 076 Speed ( mph ) 0.015*** 0 .0 1 0 *** Lanes 0.046* 0 . 105*** Facilities 0.248*** 0 .0 86*** Volume 0 . 131*** 0 . 23 0 *** * p<0.10 ** p<0.05 *** p<0.01 Our goal was to then derive the number of suppressed trips. We accomplished this by first school and then figuring out how much distance would be added to t hose trips if roads perceived as unsafe were avoided. We included children outside of Denver if their closest school was located in group level by creating o ne random point for each child living in each block group. We assigned random points to residential areas so that trips would originate in somewhat realistic patterns. While

PAGE 79

70 the assumption that each child will attend their closest school is known to be f aulty because of the Colorado Open Enrollment program that allows children to attend schools other than their originally assigned neighborhood school, privacy issues precluded us from knowing which school each child actually attends. By connecting children to their closest school, we first determined the distance that children would walk or bike to school if safety or perceived safety were not an issue (i.e. the shortest path distance). Even if these children were not attending their closest school, these are the areas that we from their home. After determining shortest path trip distances without considering safety perceptions, we ran the network analysis on ce again with each road segment now being weighted according to safety perception based suppression rates. Roads for which more than 50% of parents reported they would not allow their children to walk or bike on were weighted so as to be avoided in the se cond running of the network analysis. Since roadways had different suppression rates for walking and biking, we ran this weighted network analysis twice, once for each mode. We then derived the distance of each trip and compared the weighted distance to the original shortest path distance. We defined a trip as being suppressed when a child who would have been within a half mile walkshed or a one mile bikeshed for their shortest path was forced to exceed that walkshed or bikeshed for their weighted path ( lived 0.40 miles from their closest school and avoiding a road perceived as unsafe caused the trip to become 0.45 miles. However, if a child lived 0.40 miles from their clo sest school and avoiding a road perceived as unsafe caused the trip to become 0.55 miles, the walking trip was considered suppressed. If a child lived 0.55 miles from their closest school in the first place, we did not consider that child as a possible pe destrian.

PAGE 80

71 Figure 3 . Visualization of a suppressed pedestrian trip. After identifying suppressed trips, we accounted for the relationship between intersection density and walking and bicycling to school. We found the intersection density within a half mile buffer around each road segment and applied odds ratios based on high and low density areas from words, if there are similar numbers of suppressed t rips in both a high density and a low density neighborhood, we would expect more trips in the low density neighborhood to be stifled because of network characteristics and not as many trips to be suppressed because of safety concerns. Once we identified s uppressed trips, we identified the start point of the trip in order to locate the areas that would be most impacted. We then ran a kernel density to illustrate the areas with high concentrations and to enable spatial comparison of the proactive and reacti ve analyses.

PAGE 81

72 R esults We first examined results from existing reactive crash based traffic safety analyses (all pedestrians and bicyclists). We then utilized these methods and complemented the all age results with our own reactive crash based cluster a nalysis of child pedestrian and bicyclists. This enabled us to pursue a direct comparison between a reactive and proactive analysis. We then performed a proactive safety analysis based on trips that are suppressed because of safety concerns for child ped estrians and bicyclists. Finally, we compared the reactive analyses to the proactive analysis and observed how the perspectives align and vary in their interpretation of traffic safety. Reactive Analyses Two safety reports from Denver Public Works and on e from DRCOG provided a perspective on the reactive methodologies that existing traffic safety analyses employ. While these analyses were for pedestrians and bicyclists of all ages, they provided a methodological framework for us to complete our own crash based cluster analysis of child pedestrians and bicyclists. We performed a in order to formulate a reactive child traffic safety perspective. All crash based reactive analyses (for both children and all ages) indicated that the bulk of pedestrian and bicyclist traffic safety issues in Denver are proximate to downtown Denver, Federal Boulevard, and A lameda Avenue. Existing Reports Denver Public Works identified five bicyclist crash hotspots and six bicyclist high crash corridors in their 2016 safety report (Denver Public Works 2016). They then identified six pedestrian hotspots and five pedestrian high crash corridors in their 2017 safety report (Denver Public Works 2017). Table 2 shows that the worst pedestrian hotspot reported by Denver Public

PAGE 82

73 Works (located at 20th Street and Market Street) experienced seventeen pedestrian collisions between 2011 and 2015, while the intersection of Colfax Avenue and Colorado Boulevard was the weakest hotspot noted in the report with ten pedestrian crashes. The worst bicyclist hotspot reported by Denver Public Works (located at N. Broadway and Colfax Avenue) experienced nine bicyclist collisions between 2008 and 2012, while three hotspots with seven collisions each were the weakest bicyclist hotspots noted in the report. Denver consists of approximately 155 square miles of land. However, as Figure 4 shows, pedestrian and bicyclist crash hotspots were concentrated in a relatively small area proximate to downtown Denver and along a limited number of major corridors. The residential neighborhoods that comprise the rest of Denver have relatively few crash hotspots identified. Nearly all crash hotspo ts were found on at least one major roadway such as Colfax Avenue, Broadway, Alameda Avenue, or Federal Boulevard. While Denver Public Works did not control for levels of pedestrian and bicyclist exposure, high levels of pedestrian and bicyclist activity may be expected in these areas especially in downtown and may be one reason for the abundance of crashes. Regardless of whether high levels of non motorized exposure exist, these downtown crash hotspots and high crash c orridors merit traffic safety in terventions. Table 3 shows the twelve pedestrian crash hotspots and eight pedestrian high crash corridors that DRCOG identified in Denver along with four bicyclist crash hotspots and six bicyclist high crash corridors. Colfax Avenue was identified as the worst pedestrian corridor in the Denver metro area in terms of overall pedestrian crashes , with 418 crashes over the five study years. Colfax Avenue was also identified as the worst bicyclist corridor in the Denver metro area with 163 crashes over the fiv e study years. When considering crashes per mile as the safety metric, the Colfax Avenue corridor was ranked as the worst pedestrian corridor in the Denver metro area, but several

PAGE 83

74 corridors in Boulder surpassed Colfax Avenue in terms of bicyclist crashes per mile. All but two of the sixteen crash hotspots identified in the DRCOG report were found on high crash corridors. Figure 5 shows that the DRCOG hotspots and high crash corridors are also found near downtown Denver and along high crash corridors such as Federal Boulevard, Alameda Avenue, and Sheridan Boulevard. The residential neighborhoods that comprise the rest of Denver had few hotspots identified. Like the Denver Public Works reports, exposure was not accounted for in the DRCOG report. Because o f the location of the crash hotspots, high levels of pedestrian and bicycling activity may play a role in the high number of bicyclist crashes, especially in downtown. Findings from both reports suggest similar spatial patterns. When considering the tra ffic safety of pedestrians and bicyclists in Denver, based on these reactive approaches, an emphasis on downtown Denver and other major corridors is the suggested focus. Other neighborhoods found throughout Denver have few identified pedestrian or bicycli st hotspots.

PAGE 84

75 Table 2 . Pedestrian and Bicyclist Intersection Hotspots Identified by Denver Public Works and DRCOG Source Mode Year s Report Ranking Street 1 Street 2 Crashes/Year Denver PW Pedestrian 2011 2015 1 20 th Street Market St 3.4 2 Colfax Ave Broadway 3 .0 3 13 th Ave Broadway 2. 6 4 Federal Blvd Kentucky Ave 2. 2 4 Colfax Ave Franklin St 2. 2 6 Colfax Ave Colorado Blvd 2. 0 Bicycle 2008 2012 1 Broadway Colfax Ave 1.8 2 Lincoln St Colfax Ave 1 . 6 3 Lipan St Evans Ave 1 . 4 3 Broadway 12 th Ave 1 . 4 3 Kalamath St Alameda Ave 1 . 4 DRCOG Pedestrian 2003 2007 1 Colfax Ave Sheridan Blvd 2 . 2 2 Colfax Ave Broadway 2 . 0 3 Federal Blvd Louisiana Ave 1. 6 3 Colfax Ave High St 1. 6 5 Federal Blvd Alameda Ave 1. 4 5 Colfax Ave Mariposa St 1. 4 5 Colfax Ave Ogden St 1. 4 5 Colorado Blvd Mississippi Ave 1. 4 5 Colfax Ave Colorado Blvd 1.4 5 10 th Ave Sheridan Blvd 1.4 5 Colfax Ave Lipan St 1.4 5 Lincoln St 8 th Ave 1.4 Bicycle 2003 2007 1 Alameda Ave I 25 1.8 2 Alameda Ave Kalamath St 1.6 3 Broadway Colfax Ave 1.4 4 Cherry Creek Dr Monaco Pkwy 1.2

PAGE 85

76 Table 3 . Pedestrian and Bicyclist High Crash Corridors Identified by Denver Public Works and DRCOG Source Mode Year s Report Ranking Street S egment Crashes /Yr /Mile Denver PW Pedestrian 2011 2015 1 E Colfax Ave 1 .4 2 N Broadway 1.3 3 S Federal Blvd 0 . 9 4 W Colfax Ave 0 . 7 5 N Federal Blvd 0 . 5 Bicycle 2008 2012 1 12 th Ave 3.5 2 15 th St 3.0 3 E 16 th Ave 2 . 8 4 E Colfax Ave 2 . 0 4 Lincoln St 2 . 0 4 Broadway 2.0 DRCOG Pedestrian 2003 2007 1 Colfax Ave Union to Buckley 4 . 6 2 Broadway C 470 to Brighton 3 . 1 3 Federal Blvd Bowles to 104 th 2 . 5 3 Colorado Blvd Hampton to I 70 2 . 5 5 Alameda Ave Union to University 2 . 0 5 Peoria St Parker to 56 th 2 . 0 7 Mississippi Ave Wadswrth Brdway Parker to Buckley 1. 9 8 Sheridan Blvd Hampton to 104 th 1. 7 Bicycle 2003 2007 1 Colfax Ave Union to Buckley 1.8 2 Broadway C 470 to Brighton 1.4 3 Mississippi Ave Colfax Ave 1.2 4 Alameda Ave Monaco Pkwy 1.1 5 Colorado Blvd Wadswrth Brdway Parker to Buckley 1.0 6 Federal Blvd Bowles to 104 th 0.7

PAGE 86

77 Figure 4 . Pedestrian (above in red) and bicyclist (below in blue) crash hotspots and high crash based safety reports (all ages and all trip purposes).

PAGE 87

78 Figure 5 . Pedestrian (above in red) an d bicyclist (below in blue) crash hotspots and high crash based safety report (all ages and all trip purposes).

PAGE 88

79 Crash Cluster Analysis T he results from existing safety reports focused on pedestrians and bicy clists of all ages. We wish to explore the spatial patterns of child pedestrian and bicyclist crashes so that we can compare them to our proactive analysis. Would a reactive crash based safety analysis of child pedestrians and bicyclists focus on the sam e areas as our existing reports? We use the existing methodologies to create our own analysis. In order to explore the above question, we pulled child pedestrian and bicyclist crashes from an crashes and 208 child bicyclist crashes that were reported within Denver between 2010 and 2014. Of these 342 child pedestrian crashes, there were eleven clusters identified consisting of a total of 44 crashes. Of these eleven identified clusters, Figu re 3 shows the six clusters were significant at 95% confidence and consisted of at least three crashes. These six clusters consisted of 24 child pedestrian crashes. The largest cluster had six crashes and the smallest had three. Of the 208 child bicycli st crashes, there were fifteen clusters identified consisting of a total of eighty crashes. Of these 33 identified clusters, Figure 3 shows the eleven clusters were significant at 95% confidence and consisted of at least three crashes. These eleven clust ers consisted of 67 child bicyclist crashes. The largest cluster had twelve crashes and the smallest had three. We then formed deviational ellipses around these significant clusters of child pedestrian and bicyclist crashes (Figure 6). These deviational ellipses spatially define where we would expect to find high levels of crash incidences. Figure 7 shows the pedestrian crash ellipses are focused in an area around South Federal Boulevard. The bicyclist crash ellipses are focused in the area proximate to downtown Denver (Figure 7). The rest of the residential neighborhoods that largely comprise Denver have few identified clusters. The majority of crashes occur in thi s central area near downtown as well.

PAGE 89

80 Figure 6 . Detailed maps of child crash clusters for 2010 2014. (Data Source: DRCOG Regional Data Catalog)

PAGE 90

81 Figure 7 . Deviational ellipses around child pedestrian crashes (above) and child bicyclist crashes (below). (Data Source: DRCOG Regional Data Catalog)

PAGE 91

82 These reactive safety analyses the existing traffic safety reports from Denver Public Works and DRCOG as well as the child focused crash cluster analysis identify downtown Denver and a limited number of other major corridors as the areas of Denver that have the most pressing pedestrian and bicyclist safety issues. None of these analyses , h owever, identify safety issues in the other neighborhoods away from these major corridors of Denver. Proactive Analysis To perform the proactive safety analysis based on trips suppressed because of safety concerns, we first determined child pedestrian and bicyclist suppression rates and then completed closest facility network analyses using shortest paths and suppression weighted paths. Based on survey results from parents , the most important roadway characteristic in terms of walking trip suppression was the presence of sidewalks. This factor was a 5.5 times stronger predictor of trip suppression than the next most significant factor. There are limited sidewalk gaps in c entral Denver, while there are more sidewalk gaps present in east and north Denver (Figure 8). The most important roadway characteristic in terms of bicycling trip suppression was vehicle volumes followed closely by the presence of bicycle lanes. High vo lume roadways are found throughout Denver while bike lanes are primarily located downtown and in an east west orientation to the northeast of downtown Denver (Figure 8).

PAGE 92

83 Figure 8 . Location of pertinent roadway characteristics in Denver. (Data Source: D enver Data Catalog and DRCOG Regional Data Catalog) There were 136,138 children included in the proactive analysis with 112,648 children living in Denver and 23,490 children living in municipalities directly bordering Denver. Figure 9 shows the highest c oncentrations of children in the residential areas to the west and northeast of downtown Denver. We included 217 public elementary and middle schools both inside and outside Denver in the analysis.

PAGE 93

84 Figure 9 . Child population concentrations (red = higher concentration) and school locations in Denver. (Data Source: U.S. Census Bureau) Within GIS, t school. Of the children considered in the proactive analysis, 56.8% of ch ildren had a shortest path of 0.5 miles or less to their closest school (via network distance as opposed to Euclidian distance) and 93.1% of children had a shortest path of one mile or less (Figure 10). We focused on children who without accounting for safety perceptions had a half mile or shorter walk or a one mile or shorter bike ride.

PAGE 94

85 Figure 10 . Half mile (walkshed) and one mile (bikeshed) network buffers around schools in Denver. Next, we weighted the network analysis with trip suppression rat es and ran the analysis once again for both walking and bicycling trips. We selected children who were within the half mile walkshed and one mile bikeshed for their shortest path trip but were outside the walkshed and bikeshed once safety concerns were ac counted for. We found 3,876 children who had their walking trips suppressed and 20,364 children who had their biking trips suppressed once safety concerns were used to weight the network analysis. Figure 11 shows that these suppressed trips were found th roughout several neighborhoods primarily concentrated in west, east, and northeast Denver for walking trips and in east Denver for biking trips. There were relatively few suppressed trips found in central and southeast Denver. These trends are most likel y related to the lower concentrations of children and fewer safety perception issues (there are few sidewalk gaps and many bike lanes) found in those areas. The large concentration of walking suppression in east Denver coincides with high

PAGE 95

86 concentrations o f children and sidewalk gaps (Figure 8 & Figure 9). The areas of high bicycling suppression similarly coincide with high concentrations of children and unfavorable roadway characteristics.

PAGE 96

87 Figure 11 . Suppressed child pedestrian (above) and bicyclist (below) trip origins.

PAGE 97

88 Comparison The reactive pedestrian and bicyclist safety analyses identify areas in central Denver and along major corridors as the areas of most concern. Safety issues in t hese are as have manifested in the form of high levels of crashes. However, Figure 12 shows that in addition to the pedestrian analysis identifying S. Federal Boulevard the proactive analysis identifies areas in east, northeast, and west Denver as having perce ived traffic safety issues. These safety concerns may be hidden by a lack of non motorized activity (a result of trip suppression), which precludes crashes from occurring in the first place and from being visible i n a conventional reactive analysis.

PAGE 98

89 Figure 12 . Spatial comparison of crash based reactive analyses and proactive analysis (pedestrian on top; bicycle on bottom).

PAGE 99

90 T he proactive analysis presents fine grained results with meaningful implications. By identifying a single point as a crash hot spot through a reactive analysis, we do not understand the larger context of the issue. Similarly, by identifying a lengthy high exactly where the issue is or what is causing it. However, with the proactive analysis, an anal yst can observe where trips would be occurring, and exactly what is blocking them. Figure 13 shows that there are distinct clusters of suppressed trips throughout neighborhoods in Denver that do not necessarily fall on a major corridor or at a major inter section. One single gap in a sidewalk could possibly suppress all children over a three block area. While a single missing sidewalk does not constitute a major safety issue from a crash based perspective, it is a major safety issue for this particular ne ighborhood and can be picked up through a proactive analysis. Not only are the results of the proactive analysis new and important findings, but the form of the findings presents a more practical way of addressing safety.

PAGE 100

91 Figure 13 . Example of su ppressed pedestrian trip origin s and paths.

PAGE 101

92 Equity Analysis While we have identified parts of our built environment that experience neglected road safety issues, it is important to explore who bears the brunt of these problems. A cursory examination of the neighborhoods that had the most children encountering roads perceived as unsafe shows that those neighborhoods had median household incomes 6.2%, 15.1%, and 46.7% below average for Denver. In order to further explore this relationship, we counted the number of suppressed walking and bicycling trips for each Census tract in Denver and normalized per the number of road segments in each Census tract. NHGIS then provided median household income, race, and ethnicity data on the Census tract level. Result s show that the neglected safety issues in Denver are concentrated in neighborhoods with high levels of non white, Hispanic, and low income populations (Table 4). The relationship is strongest with Hispanic populations, while the relationship between supp ressed walking trips and median household income is weakest and does not reach statistical significance. Because the socioeconomic factors had collinearity, we created separate models for each socioeconomic variable. We can see that these formerly unreco gnized safety issues are inequitably impacting lower income, Hispanic, and non white populations. These are the people that might be best helped from active transportation improvements , although their need may not be fully recognized through more traditio nal traffic safety approaches. Table 4 . Linear Regression of Census Tract Socioeconomics and Number of Suppressed Trips Median Household Income % Non White % Hispanic Walk 0.122 0.255*** 0.360*** Bike 0.281*** 0.290*** 0.435*** * p<0.10 ** p<0.05 *** p<0.01

PAGE 102

93 C onclusion Reactive analyses have been found to identify areas where levels of exposure enable safety issues to manifest themselves. The proactive analysis, on the other hand , identifies areas where safety concerns have lowered exposure enough to effectively reduce crashes and hide safety issues from sight. While the areas identified in the proactive analysis cannot be said to be objectively unsafe as there are few crashes occurring, they are perceived as unsafe, which is signif icant i f we not only value reducing crashes but also strive to enable more people to safely walk and bik e . While we provide a methodology for a proactive safety analysis tool, the methodology does employ some assumptions that can be improved through future work. Excluding levels of non motorized exposure was a limitation . Including exposure would have provided a clearer picture, especially by allowing for crash rates in the reactive analysis. However, non motorized exposure is difficult to obtain and was not used in the existing reports. Future work would benefit from counts or volumes to better understand where trips are occurring. Furthermore, pedestrian and bicyclist age and trip purpose were inconsistent among the different analyses. B oth reactive a nalyses , however, identified the same areas as crash hotspots. Because of the concentration of children in Denver and the location of roadway characteristics perceived as unsafe, we believe that suppressed trips would be similarly clustered regardless of trip purpose. Future work will hopefully move toward a more holistic model of all ages and trip purposes. Additionally , we could account for the characteristics of crossings (e.g. signalization, phasing, turning movements, crosswalks, refuge islands, etc .) in future work to gain a better understanding of safety perceptions. Finally, there were other factors such as crime and socioeconomics that may be influencing trip suppression and would be worthy to examine in future analyses.

PAGE 103

94 Findings suggest tha t proactive safety analyses can effectively complement traditional prudent to consider both manifested and latent traffic safety issues. While we rec ommend that each city administer their own survey to account for city specific traffic safety perceptions, this paper provides a framework to accomplish more holistic traffic safety analyses . The research fills a critical gap in the literature by showing the importance of proactive safety analyses and providing methods to accomplish such analyses. While reactive crash based analyses of non motorized safety are effective at identifying manifested safety issues, we find that proactive safety analyses based on suppressed trips can be effective at identifying hidden safety issues. This suggests that bo th types of safety analyses provide unique and important perspectives on traffic safety. If our goal is to increase the number of people who can safely walk and bike as opposed to simply reducing crashes, then it is important to consider traffic safety pr oactively.

PAGE 104

95 CHAPTER V C ONCLUSION Traffic safety is a pressing public health issue, especially for vulnerable populations such as children and non motorized road users . Better understanding traffic safety issues, both those that are manifested and those t hat are suppressed, can have substantial impacts on both safety outcomes and travel behavior. The results from this work provide important new insight for our pursuit of safe streets for all. Chapter II served two purposes: 1) To identify parks as a negl ected area in terms of child pedestrian safety; and 2) to illustrate the limitations of reactive crash based analyses. While Safe Routes to School has brought much attention and funding for child pedestrians and bicyclists on their trips to school, parks are another area lacking safe access for children who independently walk or bike or wish to do so . Improving access may enable more children to enjoy the benefits of physical activity and independence both in their travel and their subsequent use of the p arks. Improvements can embrace a range of approaches from improved facilities to education. It is important to note that facilities can only do so much, and appropriate siting of parks, schools, and other destinations is necessary to enable access for ch ild pedestrians and bicyclists. While this exploration of child pedestrian fatalities provided new and important insight, it also revealed some limitations of crash based approaches. By focusing on where crashes are occurring, we are only focusing on whe re children are already walking or biking. However, as we know from our everyday travel experiences, there are areas of our cities that are unsafe enough that people especially children will not or are not allowed to walk or bike and would therefore n ot be identified by a crash analysis. We therefore sought a proactive method of identifying child pedestrian and bicyclist safety. In Chapter III, a survey that I administered provided important insights into traffic safety perceptions and travel behavio r, and therefore provided a tool to develop

PAGE 105

96 a proactive safety analysis. Findings show that sidewalks are the most important roadway characteristic in terms of walking trip suppression while vehicle volumes and bike lanes are the most important factors fo r biking trip suppression. There was a sharp decrease in the odds of trip suppression around the 5 th grade, but little difference in terms of gender. Parents reported being less willing to allow their children to bike than to walk. The majority of children experience a roadway perceived as unsafe for biking. While trips are not altered greatly on aver age throughout Denver, certain neighborhoods must travel an extra mile to avoid roads perceived as unsafe. These suppressions are concentrated near major roads, barrier connections, and tributary network configurations. Taking this new knowledge of where child pedestrian and bicyclist trips are being suppressed because of traffic safety concerns, I was able to develop and implement a proactive safety analysis in order to understand the extent of safety issues that we are currently neglecting in Denver. I found that while reactive approaches focus on where crashes are occurring and therefore where the number of pedestrians, bicyclists, and vehicles is typically high, the proactive analysis was adept at identifying perceived safety issues on less used roads and within neighborhoods. The tool identified areas that while not objectively unsafe because of a lack of crashes were perceived as unsafe and were neglected by the crash based analysis. This included areas within neighborhoods that were not necess arily located on major roads. These findings suggest that cities would be apt to adopt proactive approaches as a supplement to their current crash based approaches. This work provides a framework for such a tool, although cities that feel their transport ation system or culture is significantly different from those of Denver may wish to administer their own survey as perceptions and behavior can change. All in all, it is important that we reconsider our objectives in terms of non motorized transportation safety. Our traditional goal has been to reduce the number of crashes occurring.

PAGE 106

97 However, how do our outcomes change if we shift our goal to that of enabling more children to walk and bike safely? Now we can consider the health benefits of physical acti vity, the developmental benefits of social activity, and the mental benefits of independence, in addition to improved safety. While reducing the number of crashes certainly is important and should be a continued focus , proactively enabling children to wal k and bike is a worthy supplement to our traditional safety approach .

PAGE 107

98 REFERENCES Abdul Hanan S, MJ King, and IM Lewis. (2011). Understanding S peeding in S chool Z ones in Malaysia and Australia using an E xtended T heory of P lanned B ehaviour: T he P otential R ole of M indfulness. J ournal of Australas ian Coll ege of Road Saf ety , 22(2) , 56 62. Bellemans T, K van Bladel, D Janssens, G Wets, and H Timmermans. (2009). Measuring and Estimating Suppressed Travel with Enhanced Activity Travel Diaries. Transportation Rese arch Record , 2105, 57 63. Beltz M, and H Huang. (1998). Bicycle/Pedestrian Trip Generation Workshop: Summary . Federal Highway Administration. FHWA RD 97 141. Blackburn L, C Zegeer, and K Brookshire. (2017). Guide for Improving Pedestrian Safety at Uncontrolled Crossing Locations . FHWA SA 17 072, Washington, DC. Centers for Disease Control and Prevention. (2002). Barriers to C hildren W alking and B iking to S chool: United States, 1999. Morb idity and Mortal ity W ee kly Rep ort , 51(32) , 701 703. Centers for Disease Control and Prevention. (2014). Ten Leading Causes of Death and Injury . Injury Prevention & Control: Data & Statistics . Atlanta, GA: CDC . Centers for Disease Control and Prevention. (2017). Ten Leading Causes of Death and Injury . Injury Prevention & Control: Data & Statistics . Atlanta, GA: CDC . Cho G, DA Rodriguez, and AJ Khattak. (2009). The Role of the Built Environment in Explaining Relationships Between Perceived and Actual Pedestrian and Bicyclist Safety. Accident Analysis and Preve ntion , 41, 692 702. Cohen DA, JS Ashwood, MM Scott, A Overton, KR Evenson, LK Staten, D Porter, TL McKenzie, and D Catellier . (2006). Public P arks and P hysical A ctivity A mong A dolescent G irls. J ournal of Pediatr ics , 118(5) , 1381 1389. Cohen DT, GW Hatchard , and SG Wilson. (2015). Population Trends in Incorporated Places: 2000 to 2013 . U.S. Department of Commerce, Economics and Statistics Administration, U.S. Census Bureau. Cornell Local Roads Program. (2014). Basics of a Good Road . New York Local Technical Assistance Program Center, CLRP No. 14 04. Cutts BB, KJ Darby, CG Boone, and A Brewis . (2009). City S tructure, O besity, and E nvironmental J ustice: A n I ntegrated A nalysis of P hysical and S ocial B arriers to W alkable S treets and P ark A ccess. Soc ial Sci ence & Med icine , 69 , 1314 1322. Denver Public Works. (2016). Bicycle Crash Analysis: Understanding and Reducing Bicycle & Motor Vehicle Crashes . Denver Public Works. (2017). Pedestrian Crash Analysis: Understanding and Reducing Pedestrian & Motor Vehicle Crashes . Denver Regional Council of Governments. (2012). Pedestrian and Bicycle Safety in the Denver Region . Denver Regional Council of Governments. (2017). Report on Traffic Crashes in the Denver Region . DiMaggio C , and G Li . (2013). Effectiveness of a Safe Rout es to School Program in P reventing S chool A ged P edestrian I njury. J ournal of Pediatr ics , 131 , 290 296.

PAGE 108

99 Dumbaugh E, and L Frank. (2007). Traffic S afety and Safe Routes to Schools: S ynthesizing the E mpirical E vidence. Transp ortation Res earch Rec ord , 2009 , 89 97. Environmental Systems Research Institute. (2018). How Directional Distribution (Standard Deviational Ellipse) Works. Evers C, S Boles, D Johnson Shelton, M Schlossberg, and D Richey. (2014). Parent Safety Perceptions of Child Walking Routes. Journal of Transport & Health , 1, 108 115. Federal Highway Administration. (2006). Pedestrian and Bicyclist Safety . FHWA HRT 05 090. Ferenchak NN, and WE Marshall. (2018). Quantifying Suppressed Child Pedestrian and Bicyclist Trips. Travel Behaviour and Society , Under Review. Foster S, M Knuiman, K Villanueva, L Wood, H Christian, and B Giles Corti. (2014). Does Walkable Neighbourhood Design Influence the Association Between Objective Crime and Walking? International Journal of Behavioral Nutrition and Physical Activity , 11(1), 100. Graham A, and P Sparkes. (2010). Casualty Reductions in NSW Associated with the 40 km/h School Zone Init iative. Canberra, Australian Capital Territory: 2010 Australasian Road Safety, Research, Policing and Education Conference . Jacobsen PL. (2003). Safety in N umbers: M ore W alkers and B icyclists, S afer W alking and B icycling. Inj ury Prev ention , 9 , 205 209. Jol y MF, P Foggin, and I Ples s . (1991). Socioecological D eterminants of the R isk of A ccidents in Y oung P edestrians. Rev ue Epidemiol ogie et de Sante Publique , 39(4) , 345 351. Kattan L, R Tay, and S Acharjee. (2011). Managing S peed at S chool and P layground Z ones. Accid ent Anal ysis and Prev ention , 43 , 1887 1891. Kraus JF, EG Hooten, KA Brown, C Peek Asa, C Heye, and DL McArthur . (1996). Child P edestrian and B icyclist I njuries: R esults of C ommunity S urveillance and a C ase C ontrol S tudy. Inj ury Prev ention , 2 , 212 218. Krizek KJ, G Barnes, and K Thompson. (2009). Analyzing the E ffect of B icycle F acilities on C ommute M ode S hare O ver T ime. Journal of Urban Planning and Development , 135(2) , 66 73. Krizek KJ, A El Geneidy, and K Thompson. (2007). A Detailed Analysis of How and Urban Trail Transportation , 34, 611 624. Larsen K, RN Buliung, and GEJ Faulkner. (2013). Safety and School Travel: How Does the Environment Along the Route Relate to Safety and Mode Choice? Transportation Resear ch Record , 2327, 9 18. Lee RE, KM Booth, JY Reese Smith, G Regan, and HH Howard . (2005). The P hysical A ctivity R esource A ssessment (PARA) I nstrument: E valuating F eatures, A menities and I ncivilities of P hysical A ctivity R esources in U rban N eighborhoods. Int ernational J ournal of Behav ioral Nutr ition and Phys ical Act ivity , 2(1) , 13. Loukaitou Sideris A. (2003). C ommon G rounds: A S tudy of I ntergroup R elations among C hildren in P ublic S ettings. J ournal of the Am erican Plann ing Assoc iation , 69(2) , 130 143.

PAGE 109

100 Loukaitou Sideris A, and A Sideris. (2009). What B rings C hildren to P arks? Analysis and M easurement of the V ariables A ffecting C U se of P arks. J ournal of the Am erican Plann ing Assoc iation , 76(1) , 89 107. Manson S, J Schroeder, D Van Riper, and S Ruggles. (2017). IPUMS National Historical Geographic Information System: Version 12.0 [Database]. Minneapolis: University of Minnesota . http://doi.org/10.18128/D050.V12.0 Marshall WE. (2015). Understanding the I mp acts of I ntegrating New Urbanist N eighborhood and S treet D esign I deals with C onventional T raffic E ngineering S tandards: T he C ase of Stapleton. J ournal of Urban ism , 8(2) , 148 172. Molino JA, JF Kennedy, PJ Inge, MA Bertola, PA Beuse , NL Fowler, AK Emo, and A Do. (2012). A Distance Based Method to Estimate Annual Pedestrian and Bicyclist Exposure in an Urban Environment . FHWA HRT 11 043, Washington, DC. Moudon AV, PM Hess, JM Matlick, and N Pergakes. (2002). Pedestrian Location Identi fication Tools: Identifying Suburban Areas with Potentially High Latent Demand for Pedstrian Travel. Transportation Research Record , 1818, 94 101. National Highway Traffic Safety Administration. (2016). Traffic Safety Facts: 2015 Motor Vehicle Crashes: Ove rview . U.S. Department of Transportation, DOT HS 812 318. National Highway Traffic Safety Administration. (2017). Traffic Safety Facts: 201 6 Fatal Motor Vehicle Crashes: Overview . U.S. Department of Transportation, DOT HS 812 456. Nevelsteen K, T Steenberghen, A Van Rompaey, and L Uyttersprot. (2012). Controlling Factors of Accident Analysis and Prevention , 45, 39 49. Noland RB. (1995). Perceived Risk and Modal Choice: Risk Compensation in Transportation Systems. Accident Analysis and Prevention , 27(4), 503 521. Orenstein MR, N Gutierrez, TM Rice, JF Cooper, and DR Ragland . (2007). Safe Routes to School Safety and Mobil ity Analysis . Safe Transportation Research & Education Center. Phelan KJ, J Khoury, HJ Kalkwarf, and BP Lanphear . (2001). Trends and P atterns of P layground I njuries in United States C hildren and A dolescents. Ambul atory Pediatr ics , 1 , 227 233. Price AE, JA Reed, and S Muthukrishnan. (2012). Trail U ser D emographics, P hysical A ctivity B ehaviors, and P erceptions of a N ewly C onstructed G reenway T rail. J ournal of Community Health , 37 , 949 956. Pucher J, J Dill, and S Handy. (2010). Infrastructure, Programs, and P olicies to Increase Bicycling: An International Review. Preventive Medicine , 50, S106 S125. Roberts IG, MD Keall, and WJ Frith. (1994). Pedestrian E xposure and the R isk of C hild P edestrian I njury. J ournal of Paediatr ics and Child Health , 30 , 220 223. Routledge DA, R Repetto Wright, and CI Howarth. (1974). A C omparison of I nterviews and O bservation to O btain M easures of C E xposure to R isk as P edestrians. Ergonomics , 17(5) , 623 638. Schlossberg M, J Greene, PP Phillips, B Johnson, and B Parker. (2006). School Trips: Effects of Urban Form and Distance on Travel Mode. Journal of the American Planning Association , 72(3), 337 346.

PAGE 110

101 Schneider RJ, RM Ryznar, and AJ Khattak. (2003). An Accident Waiting to Happen: A Spatial Approach to Proactive Pedestria n Planning. Accident Analysis and Prevention , 36, 193 211. Stevenson MR. (1991). Analytical A pproach to the I nvestigation of C hildhood P edestrian I njuries: A R eview of the L iterature. J ournal of Safety Res earch , 22 , 123 132. Stutts JC, and WW Hunter. (1999). Motor V ehicle and R oadway F actors in P edestrian and B icyclist I njuries: A n E xamination B ased on E mergency D epartment D ata. Accid ent Anal ysis and Prev ention , 31 , 505 514. Tay R. (2009). Speed C ompliance in S chool and P layground Z ones. ITE Journal , 79(3) , 36 38. Toroyan T, and M Peden. (2007). Youth and Road Safety. Geneva, Switzerland: World Health Organization. Transportation Research Board. (2001). Statistical Methods in Highway Safety Analysis . NCHRP Synthesis 295, Washington, DC. Turner S, I Se ner, M Martin, S Das, E Shipp, R Hampshire, K Fitzpatrick, L Molnar, R Wijesundera, M Colety, and S Robinson. (2017). Synthesis of Methods for Estimating Pedestrian and Bicyclist Exposure to Risk at Areawide Levels and on Specific Transportation Facilities . FHWA SA 17 041, Washington, DC. Vatcheva KP, M Lee, JB McCormick, and MH Rahbar. (2016). Multicollinearity in Regression Analyses Conducted in Epidemiological Studies. Epidemiology , 6(2), 227 236. Waldheim N, E Wemple, and J Fish. (2015). Applying Safety Data and Analysis to Performance Based Transportation Planning . FHWA SA 15 089. Wier M, J Weintraub, EH Humphreys, E Seto, and R Bhatia . (2009). An A rea L evel M odel of V ehicle P edestrian I njury C ollisions with I mplications for L and U se and T ransportation P lanning. Accid ent Anal ysis and Prev ention , 41 , 137 145. Wolch J, JP Wilson, and J Fehrenbach. (2005). Parks and P ark F unding in Los Angeles: A n E quity M apping A nalysis. Urban Geogr aphy , 26(1) , 4 35. Zegeer C, D Nabors, D Gelinne, N Lefler , and M Bushell. (2010). Pedestrian Safety Strategic Plan: Recommendations for Research and Product Development . FHWA 10 035.