Citation
How pedestrian friendly are we

Material Information

Title:
How pedestrian friendly are we pedestrian accidents and safety in the city and county of Denver
Creator:
Kuhlmann, Anne K. Sebert
Publication Date:
Language:
English
Physical Description:
xii, 208 leaves : illustrations, maps ; 28 cm

Subjects

Subjects / Keywords:
Pedestrian accidents -- Case studies -- Colorado -- Denver ( lcsh )
Pedestrian accidents ( fast )
Colorado -- Denver ( fast )
Genre:
Case studies. ( fast )
bibliography ( marcgt )
theses ( marcgt )
non-fiction ( marcgt )
Case studies ( fast )

Notes

Abstract:
Being a pedestrian is a risky activity. In the United States pedestrians are more likely to be killed per kilometer traveled than are passengers of public transportation, passenger cars and vans, and commercial airlines. The United States also ranked worse than Western Europe, Australia, and Canada in terms of pedestrian danger. Using a cross-sectional, retrospective design, this exploratory study analyzed patterns of pedestrian/motor vehicle accidents within the City and County of Denver by census block group and census tract over a four-year period from 2000 to 2003. The study integrated data from traffic accident reports, liquor licenses, land use, street structure, and the 2000 U.S. census to look at characteristics of the built and social environments associated with rates of pedestrian/motor vehicle accidents. At both the census block group and tract levels, three variables (proportion of labor force that walks or takes public transportation to work, population density, and liquor license establishments per 1,000,000 square feet) contributed significantly to a model of pedestrian/motor vehicle accidents per 1,000,000 square feet. Each of these variables has a positive, significant relationship to pedestrian/motor vehicle accidents per 1,000,000 square feet. The results of the study have implications for future research directions, public policy for pedestrian safety and road use, and public health programs aimed at decreasing unintentional injury from pedestrian accidents and increasing walking as a routine physical activity to combat obesity and its related complications.
Bibliography:
Includes bibliographical references (leaves 196-208).
General Note:
Department of Health and Behavioral Sciences
Statement of Responsibility:
by Anne K. Sebert Kuhlmann.

Record Information

Source Institution:
|University of Colorado Denver
Holding Location:
|Auraria Library
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
777957405 ( OCLC )
ocn777957405
Classification:
HE5614.4.D46 K84 2006a ( lcc )

Full Text
HOW PEDESTRIAN FRIENDLY ARE WE:
PEDESTRIAN ACCIDENTS AND SAFETY IN THE
CITY AND COUNTY OF DENVER
A thesis submitted to the
University of Colorado at Denver and Health Sciences Center
in partial fulfillment
of the requirements for the degree of
Doctor of Philosophy
Health and Behavioral Sciences
2006
by
Anne K. Sebert Kuhlmann
B.S., Emory University, 1998
M.P.H., University of Michigan, 2000


This thesis for the Doctor of Philosophy
degree by
Anne K. Sebert Kuhlmann
has been approved
by

Sunmin Lee
7-J Date
V


Sebert Kuhlmann, Anne K. (Ph.D., Health and Behavioral Sciences)
How Pedestrian Friendly Are We: Pedestrian Accidents and Safety in the City and
County of Denver
Thesis directed by Assistant Professor John Brett
Being a pedestrian is a risky activity. In the United States pedestrians are more likely to
be killed per kilometer traveled than are passengers of public transportation, passenger
cars and vans, and commercial airlines. The United States also ranked worse than
Western Europe, Australia, and Canada in terms of pedestrian danger. Using a cross-
sectional, retrospective design, this exploratory study analyzed patterns of pedestrian/
motor vehicle accidents within the City and County of Denver by census block group and
census tract over a four-year period from 2000 to 2003. The study integrated data from
traffic accident reports, liquor licenses, land use, street structure, and the 2000 U.S.
census to look at characteristics of the built and social environments associated with rates
of pedestrian/ motor vehicle accidents. At both the census block group and tract levels,
three variables (proportion of labor force that walks or takes public transportation to
work, population density, and liquor license establishments per 1,000,000 square feet)
contributed significantly to a model of pedestrian/ motor vehicle accidents per 1,000,000
square feet. Each of these variables has a positive, significant relationship to pedestrian/
motor vehicle accidents per 1,000,000 square feet. The results of the study have
implications for future research directions, public policy for pedestrian safety and road
use, and public health programs aimed at decreasing unintentional injury from pedestrian
accidents and increasing walking as a routine physical activity to combat obesity and its
related complications.
This abstract accurately represents the content of the candidates thesis. I recommend its
publication.
ABSTRACT


ACKNOWLEDGEMENTS
I would like to acknowledge the chair of my dissertation, Dr. John Brett, as well as
the members of my committee for all their assistance with this work: Dr. Craig Janes,
Dr. Sunmin Lee, Dr. Steve Sain, and Dr. Deborah Thomas. I would also like to
acknowledge the many people in various local and state agencies who facilitated my
interest in this topic and my access to the publicly available data: Dave Weaver,
Department of Public Works, City and County of Denver; Joan Vecchi, Motor
Vehicle Business Group, Colorado Department of Revenue; Holly Hedegaard, Injury
Epidemiology, Colorado Department of Public Health and the Environment; Becky
Picaso, Colorado State Demographers Office; and Sandie Norman, Department of
Public Safety, City and County of Denver.
IV


CONTENTS
Abbreviations.................................................viii
Figures.......................................................ix
Tables..........................................................x
CHAPTER
1: INTRODUCTION.................................................1
Literature Review.........................................4
Pedestrian Accidents and Safety....................4
Obesity and Diabetes..............................13
Walkability.......................................16
Built Environment/ Neighborhood Level Effects....19
Livable Communities/ Urban Sprawl.................20
Conclusion...............................................22
2: THEORETICAL PERSPECTIVES....................................24
Liquor License Establishments and Public Health Outcomes.29
Social Ecology...........................................33
Strengths and Weaknesses of Social Ecology........37
Social Ecology as a Framework for Pedestrian Accidents.. ..38
Conclusion...............................................42
3: METHODS.....................................................43
Research Design..........................................44
Study Location...........................................46
Study Period.............................................47
v


Data Sources.................................................47
Traffic Accident Report Data..........................48
Liquor License Establishment Data.....................56
Land Use Data.........................................58
Street Structure Data.................................60
Census Data...........................................62
Exploratory Data Analysis....................................64
Multivariate Data Analysis...................................67
Human Subjects...............................................68
4: POINT-LEVEL RESULTS..............................................69
Pedestrian/ motor vehicle accidents..........................69
Liquor license establishments................................80
Discussion...................................................83
5: CENSUS BLOCK GROUP-LEVEL RESULTS................................88
Exploratory Analysis: Pedestrian/ motor vehicle accidents....88
Exploratory Analysis: Liquor license establishments..........96
Exploratory Analysis: Land use data.........................102
Exploratory Analysis: Street structure data.................106
Exploratory Analysis: Socio-demographic variables...........106
Bivariate Analyses..........................................110
Multivariate Analyses...................................... 117
Conditional Autoregressive Model............................121
Discussion..................................................122
6: CENSUS TRACT-LEVEL RESULTS......................................125
Exploratory Analysis: Pedestrian/ motor vehicle accidents...125
vi


Exploratory Analysis: Liquor license establishments.......133
Exploratory Analysis: Land use data.......................139
Exploratory Analysis: Street structure data...............142
Exploratory Analysis: Socio-demographic variables.........145
Bivariate Analyses........................................149
Multivariate Analyses.....................................154
Conditional Autoregressive Model..........................160
Discussion................................................161
7: DISCUSSION AND CONCLUSION.....................................169
Discussion................................................170
Limitations...............................................176
Future Research Directions...............................179
Policy Recommendations...................................188
Conclusion................................................191
APPENDIX...............................................................193
REFERENCES CITED.......................................................196
vii


ABBREVE ATION S
BMI Body Mass Index
BRFSS Behavioral Risk Factor Surveillance Survey
CAR Conditional Auto-Regressive Model
CDC U.S. Centers for Disease Control and Prevention
CDPHE Colorado Department of Public Health and the Environment
DALYs Disability Adjusted Life Years
DHHS U.S. Department of Health and Human Services
DIA Denver International Airport
DPW Department of Public Works of the City and County of Denver
FARS Fatality Accident Reporting System
FIPS Federal Information Processing System
GIS Geographic Information System
MPH Miles per Hour
NCSA National Center for Statistics and Analysis
NEWS Neighborhood Environment Walkability Scale
NHANES National Health and Nutrition Examination Survey
NHTSA National Highway Traffic Safety Administration
PDI Pedestrian Danger Index
PTA Parent Teacher Association
STPP Surface Transportation Policy Project
TAR Traffic Accident Report
TIGER Topographically Integrated Geographic Encoding and Referencing
USCB U.S. Census Bureau
WHO World Health Organization
vm


FIGURES
Figure
2.1: Social ecological framework for pedestrian safety.......................41
3.1: Streets within the City and County Denver...............................52
3.2: Census Block Groups within the City and County of Denver................53
3.3: Census Tracts within the City and County of Denver......................54
4.1: Pedestrian/ Motor Vehicle Accident Locations in Denver..................70
4.2: Liquor License Establishment Locations within the City and Count of Denver..81
5.1: Pedestrian/ Motor Vehicle Accidents by Census Block Group................92
5.2: Pedestrian Accidents per 1,000,000 Square Feet by Census Block Group.....93
5.3: Pedestrian Accidents per 10 Roadway Miles by Census Block Group..........94
5.4: Pedestrian Accidents per 1,000 People by Census Block Group..............95
5.5: Total Liquor License Establishments by Census Block Group...............100
5.6: Liquor Licenses per 1,000,000 Square Feet by Census Block Group.........101
5.7: Percent of Land Dedicated to Business Activity by Census Block Group....105
5.8: Percent of Labor Force that Walks or Takes Public Transport to Work by
Census Block Group......................................................109
6.1: Pedestrian/ Motor Vehicle Accidents by Census Tract.....................129
6.2: Pedestrian Accidents per 1,000,000 Square Feet by Census Tract..........130
6.3: Pedestrian Accidents per 10 Roadway Miles by Census Tract...............131
6.4: Pedestrian Accidents per 1,000 People by Census Tract...................132
6.5: Total Liquor License Establishments by Census Tract.....................136
6.6: Liquor Licenses per 1,000,000 Square Feet by Census Tract...............138
6.7: Proportion of Land Dedicated to Business Activity by Census Tract.......141
6.8: Proportion of Streets Designated as Local Streets by Census Tract.......144
IX


6.9: Percent of Labor Force that Walks or Takes Public Transport to Work by
Census Tract............................................................


TABLES
Table
1.1: Pedestrian fatality rates by location........................................6
1.2: Demographic characteristics of pedestrian fatalities in Colorado by county.9
3.1: Data type by source.........................................................48
3.2: Calculated variables based on traffic accident report data..................55
3.3: Calculated variables based on liquor license establishment data.............58
3.4: Land use variables..........................................................59
3.5: Street structure variables..................................................61
3.6: Variables calculated from the 2000 U.S. census data.........................63
3.7: Variable transformations at the census block group and census tract levels.66
4.1: Streets with 20 or more pedestrian/ motor vehicle accidents................72
4.2: Description of when pedestrian/ motor vehicle accidents occur..............74
4.3: Conditions under which pedestrian/ motor vehicle accidents occur...........76
4.4: Pedestrians involved in pedestrian/ motor vehicle accidents................79
4.5: Categorization of liquor license establishments............................83
5.1: Descriptive statistics of pedestrian/ motor vehicle accidents per census block
group......................................................................90
5.2: Global Morans I calculations for potential outcome variables...............96
5.3: Descriptive statistics of liquor licenses per census block group............98
5.4: Descriptive statistics of land use per census block group..................103
5.5: Descriptive statistics of socio-demographic variables per census block group.. 107
5.6: Correlations between built environment explanatory variables...............111
5.7: Pearsons correlations for socio-demographic variables at census block group
level.....................................................................113
5.8: Correlations between built environment and selected social environment
explanatory variables.....................................................114
5.9: Correlations between outcome variables at census block group...............115
xi


5.10: Correlations of potential explanatory variables to potential outcome
variables..................................................................116
5.11: Simultaneous linear regression for pedestrian/ motor vehicle accidents per
1.000. 000 sq ft in census block group....................................119
5.12: Simultaneous linear regression for pedestrian/ motor vehicle accidents per
1.000. 000 sq ft in census block group, stratified by unemployment level..120
5.13: Conditional autoregressive model for pedestrian/ motor vehicle accidents per
1.000. 000 sq ft in census block group....................................122
6.1: Descriptive statistics of pedestrian/ motor vehicle accidents per census tract...127
6.2: Global Morans I calculations for potential outcome variables per census tractl28
6.3: Descriptive statistics of liquor licenses per census tract................135
6.4: Descriptive statistics of land use per census tract.......................139
6.5: Descriptive statistics of street structure per census tract...............142
6.6: Descriptive statistics of socio-demographic variables per census tract....146
6.7: Correlations between built environment explanatory variables..............149
6.8: Pearsons correlations for socio-demographic variables per census tract...151
6.9: Correlations between built environment and selected social environment
explanatory variables.................................................... 152
6.10: Pearsons correlations between selected explanatory variables and potential
outcome variables..........................................................153
6.11: Correlations between outcome variables...................................154
6.12: Simultaneous linear regression for pedestrian/ motor vehicle accidents per
1.000. 000 sq ft in census tract..........................................158
6.13: Simultaneous linear regression for pedestrian/ motor vehicle accidents per
1.000. 000 sq ft in census tract, stratified by poverty level.............159
6.14: Conditional autoregressive model for pedestrian/ motor vehicle accidents per
1.000. 000 sq ft in census tract..........................................160
xii


CHAPTER 1
INTRODUCTION
Pedestrian/motor vehicle accidents are an important cause of unintentional injury. To
date, however, pedestrian/motor vehicle accidents have received only limited
attention in the public health literature. While substantial literature discusses the risks
surrounding and prevention efforts related to motor vehicle crashes, only a small
proportion of this literature discusses pedestrian risks, accidents, and safety. This
lack of information and data about pedestrian accidents and safety is particularly
unsettling in light of the numerous public health efforts that promote walking for
health and wellness.
In response to the rising rates of obesity, diabetes, and other chronic diseases in the
Untied States, public health campaigns promote walking among adults for pleasure,
exercise, and everyday errands as well as among children as transportation to and
from school. Available data indicate that being a pedestrian can be a very risky
activity, much riskier than traveling in a motor vehicle. Furthermore, surveys
indicate that a substantial percentage of parents perceive traffic danger as a barrier to
allowing their children to walk or bike to school. Therefore, a better understanding of
pedestrian accidents and safety could play a key role both in safety improvements and
in injury prevention, as well as public health efforts to combat chronic disease.
The existing public health literature regarding pedestrian/motor vehicle accidents
suggests that the burden of morbidity and mortality resulting from pedestrian/motor
vehicle accidents is disproportionately distributed both geographically and
demographically within the population. Analyses of pedestrian/motor vehicle
1


accidents in Montreal and Mexico City reveal a concentration of accidents in just a
few census tracts and at a handful of intersections, respectively. Reports from
Atlanta, Los Angeles, and Washington, D.C. indicate that Latinos and Blacks are
significantly more likely to be killed or injured as pedestrians than are Whites. A
review of the public health literature revealed no information on pedestrian safety or
accidents in the Denver metropolitan area. Given the changing demographics of the
area and the nationwide public health endorsement of walking as a healthy, simple
mechanism for combating obesity, diabetes, and other chronic health conditions, a
better understanding of pedestrian safety and accidents in Denver would provide
useful information for the public health, public safety, and urban planning
communities.
This study investigated patterns of pedestrian/ motor vehicle accidents within the City
and County of Denver. All pedestrian/ motor vehicle accidents were included in the
study, not just fatal accidents or injury accidents which have received most of the
attention in the literature to date. The study was an exploratory, ecological study
using a retrospective, cross-sectional design. It covered a 4-year period from January
1, 2000 through December 31, 2003. The study sought to answer two main research
questions: 1) What are the spatial patterns of pedestrian/ motor vehicle accidents
within the City and County of Denver?; 2) What is the relationship between
characteristics of the built and social environments and levels of pedestrian/ motor
vehicle accidents at the census block group and census tract geographic units of
analysis within the City and County of Denver? As the research questions indicate,
the study integrated both spatial and population data in order to provide a more
comprehensive picture of pedestrian/ motor vehicle accidents within the City and
County of Denver.
2


For the purposes of this study, pedestrian accidents refer to events involving a
collision between a motor vehicle and a person walking or jogging. Specifically, the
study focuses on pedestrian/ motor vehicle accidents that occur on public roadways.
Accidents that occur in private parking lots or entirely within private driveways are
not considered. Accidents that occur in public alleys or on limited-access highways
were also excluded, as will be explained in more detail later. Accidents involving
public transport vehicles, such as buses or light rail trains, also were not included.
Likewise, bicycle/motor vehicle collisions and bicycle/pedestrian collisions are not
included in this study. Therefore, this study presents a very conservative overall
picture of pedestrian/ motor vehicle accidents within the City and County of Denver.
This dissertation initially provides an overview of the background literature that
speaks to the importance of gaining a better understanding of pedestrian/ motor
vehicle accidents followed by a discussion of the theoretical perspectives that provide
a framework for considering pedestrian/ motor vehicle accidents within a place-based
context and details the methods used to answer the research questions. The results
are presented in sequential levels of geographic aggregation from the point location,
to the census block group, to the census tract so that readers can follow the
progression of patterns of pedestrian/ motor vehicle accidents over increasing
geographic units of aggregation. Each results chapter provides an immediate
discussion of the results at that geographic unit of analysis at the end of the chapter.
The final chapter provides a discussion of the results overall and across geographic
units of aggregation, the limitations of the study, thoughts on future research ideas
generated from the results, and revisits the theoretical perspectives as these help
provide insight into the studys findings and future research ideas.
3


Literature Review
Pedestrian Accidents and Safety
Being a pedestrian is a risky activity. Pedestrians are up to nine times more likely
than those traveling in cars to be killed per kilometer traveled (Koomstra 2003; Miller
et al. 1999). Both walking and bicycling are more dangerous than car travel in the
United States (Pucher & Dijkstra 2003). In 2001, the pedestrian fatality rate for the
United States was 20.1 fatalities per 100 million miles traveled, making walking
substantially more dangerous than travel by public transit (0.75 fatalities per 100
million miles traveled), passenger cars or trucks (1.3), or commercial airlines (7.3)
(Ernst 2004). As these fatality rates were calculated from 2001 data, the rate for
commercial airlines is unusually high due to September 11, 2001 (Ernst 2004). In the
United States, nearly 175,000 pedestrians were killed in accidents with motor
vehicles between 1975 and 2001 (NCSA 2003). These numbers represent only
pedestrian fatalities. They do not include those pedestrians who were injured in
motor vehicle crashes nor those fortunate to escape unharmed. Despite these
statistics, awareness of pedestrian accidents and safety as a major public health issue
is just beginning to arise.
The World Health Organization (WHO) designated the theme of World Health Day
2004, an annual event marking the founding of the organization and used to raise
awareness globally of a single public health issue, as road safety the first time
such a theme has been selected (WHO 2004). A joint report by the WHO and the
World Bank calls for increased attention of public health to road safety, including
pedestrian safety. According to estimates, road traffic injuries will rank as the third
leading cause of disability adjusted life years (DALYs) globally by 2020, up from
4


ninth in 1990 (Murray & Lopez 1996). Many of these injuries will be among
pedestrians. The WHO report acknowledges that, historically, road safety issues have
been relegated to the responsibility of the transportation and urban planning sectors
but emphasizes that the public health sector has much to gain by playing an
important, collaborative role with other sectors in efforts to improve road safety and
raise awareness throughout the world (Peden et al. 2004). Better road injury
prevention would not only lead to fewer hospitalizations and reduced severity of
injuries, but guaranteeing safer conditions on the roads for pedestrians and cyclists
might also encourage more people to adopt the behaviors of walking and cycling as
part of a healthier lifestyle (Peden et al. 2004).
In Colorado, 78 pedestrians were killed by motor vehicles during 2002, according to a
preliminary analysis using the Fatality Analysis Reporting System (FARS), a national
census of all fatal motor vehicle crashes on public traffic ways supported by the
National Highway Traffic Safety Administration (NHTSA). Pedestrian fatalities are
the third leading cause of unintentional injury deaths in the state of Colorado behind
only motor vehicle occupant fatalities and fatalities from unspecified falls (CDPHE
2002). Over an 18-month period in 2003-2004, the Denver Police Department
received approximately 800 reports of motor vehicle accidents involving pedestrians
in the City and County of Denver (unpublished data, Denver Police Department,
2004). Clearly, pedestrian safety is an important public health issue in Colorado as
well as a transportation and urban planning issue.
Contrary to what one might expect, however, pedestrian/motor vehicle accidents are
not random events. If they were, they would occur equally among different
population groups and across various locations within a city or a region. This is not
the case, however. A number of published articles and reports demonstrate the non-
uniform geographic and socio-demographic distribution of pedestrian accidents.
5


Disparities in pedestrian/motor vehicle accidents appear at every geographic unit, all
of which are important to understand if public health is truly to contribute to
improving road safety throughout the world. Globally, Africa experiences the highest
rate of road injuries while Western Europe and Australia experience the lowest rate.
The Pan-American region has rates higher than Western Europe and Australia but
lower than those of Africa, Asia, and the former Soviet-bloc (Peden et al. 2004). At
the national level, discrepancies exist in rates of pedestrian/ motor vehicle accidents
between specific countries, such as the Netherlands, Germany, and the United States
- with the United States having the highest rates (Pucher & Dijkstra 2003; Fletcher &
McMichael 1997). Table 1.1 illustrates these differences in pedestrian fatality rates.
Within North America, Canada has a much lower rate of pedestrian injuries than the
United States and has had more success in reducing its accident fatality rate over the
past 25 years (Peden et al. 2004). Regionally within the United States, reports
indicate that cities in the South and the West of the country tend to have the highest
pedestrian danger levels when compared to cities in other regions (NCSA 2003; STPP
2002). Clearly, some places have been much more successful in ensuring and
improving pedestrian safety than other locations.
Table 1.1 Pedestrian Fatality Rates by Location
Location Pedestrian Fatalities per 100 Million Kilometers Traveled
United States a 14.0
European Union b 6.4
Germanya 4.4
Netherlands a n a 2.5 n r\:*i aaao b i r\r\r\A
Source:a = Pucher & Dijkstra 2003; b = Peden et al. 2004
While these global and national differences in pedestrian fatality and injury rates are
certainly interesting and provide a context within which to consider local disparities,
6


Hijar and colleagues argue that looking at the geographic distribution of pedestrian
accidents at the city level can provide important information for improving pedestrian
safety locally (2003). Research from Montreal, Mexico City and Atlanta reveals
geographic disparities in pedestrian accidents at the local level (Joly et al. 1991; Hijar
et al. 2003; Hanzlick et al. 1999). For example, pedestrian and bicyclist accidents
among children in Montreal are concentrated in just a handful of census tracts (Joly et
al. 1991). In Mexico City, where pedestrian fatalities account for the majority of
motor vehicle fatalities, Hijar and colleagues found that these fatalities are
concentrated in just ten neighborhoods, despite Mexico City being one of the worlds
largest cities (2003). Furthermore, six specific intersections within these ten
neighborhoods experience a particularly high concentration of these pedestrian
fatalities (Hijar et al. 2003). In Atlanta, the two counties most central to the
metropolitan area incurred the highest rate of pedestrian fatalities between 1994 and
1998 (Hanzlick et al. 1999). This research from other cities in North America
suggests that pedestrian/motor vehicle accidents in Denver may show distinct
geographic patterns as well.
Most analyses of pedestrian accidents in the United States are based on publicly
available data from the FARS. Preliminary analysis of the 2002 Colorado FARS data
demonstrates uneven geographic of pedestrian fatalities across the state. The Denver
metropolitan area incurred the largest proportion of the 78 pedestrian fatalities in
Colorado during 2002, with the City and County of Denver experiencing 28 of those
78 deaths (36%). According to the 2000 Census, however, the City and County of
Denver account for only 13% of Colorados 4.3 million people (U.S. Census Bureau
2004). Therefore, Denver had an index of approximately 5.0 pedestrian fatalities for
100,000 people while the index for the state as a whole was only 1.8 per 100,000
people. The 28 pedestrian fatalities in Denver during 2002 may be a slight elevation
from a typical year, as the National Center for Statistics and Analysis (NCSA 2003)
7


reports the average number of pedestrian fatalities was 23 for the City and County of
Denver from 1998 through 2000. Still, 23 pedestrian fatalities would result in index
of 4.1 pedestrian fatalities per 100,000 more than two times the index for the state
as a whole.
Multiple pedestrian fatalities also occurred in Adams, Arapahoe, and Jefferson
counties within the Denver metropolitan area. Outside the Denver metropolitan area,
El Paso County (Colorado Springs), Larimer County (Fort Collins), and Pueblo
County (Pueblo) were the only counties to experience more than 2 pedestrian
fatalities in 2002. Table 1.2 provides an overview of the FARS data by county in
Colorado from 2002. When looking at FARS data, it is important to remember that
FARS only represents fatal accidents. As will be shown in results, pedestrian/ motor
vehicle accidents resulting in a pedestrian fatality account for only approximately 4%
of all pedestrian/ motor vehicle accidents in Denver.
8


Table 1.2: Demographic Characteristics of Pedestrian Fatalities in Colorado by County, 2002
COUNTY TOTAL <20 AGE 20- 60 >60 ETHNICITY Non- Hispanic Hispanic White RACE Black Other GENDER Male Female
Adams 8 2 6 0 3 3 4 1 , 7 1
Arapahoe 8 3 3 2 1 4 5 0 0 4 4
Baca 1 0 0 1 0 1 1 0 0 0 1
Boulder 1 0 1 0 0 1 1 0 0 0 i
Broomfield 1 0 1 0 0 0 0 0 0 1 0
Denver 28 4 19 5 11 12 22 0 2 19 9
Douglas 1 1 0 0 0 0 0 0 0 1 0
Eagle 1 0 1 0 1 0 I 0 0 1 0
El Paso 6 2 3 1 1 2 3 0 0 5 1
Jefferson 4 0 3 1 0 3 2 0 1 1 3
Larimer 7 0 3 4 0 6 6 0 0 4 3
Mesa 2 0 2 0 0 2 2 0 0 i 1
Montezuma 1 0 0 1 1 0 1 0 0 i 0
Montrose 1 0 0 1 0 1 1 0 0 1 0
Pueblo 6 0 5 1 2 2 3 1 0 6 0
Weld 2 1 1 0 0 0 0 0 0 2 0
TOTALS 78 13 48 17 20 37 52 2 . 4 54 24
Different agencies and organizations in the United States utilize various means for
ranking metropolitan areas according to pedestrian danger. Most indices are based on
data from FARS combined with demographic information available from the U.S.
Census Bureau. Depending on which index is used, Denver ranks between 12th and
26th among U.S. cities on measures of danger for pedestrians, with first being the
most dangerous. At the time of the 2000 Census, Denver was the 25th largest city in
the United States (USCB 2004). The Surface Transportation Policy Project (STPP)
calculates a pedestrian danger index (PDI) based on the rate of pedestrian deaths
relative to the amount people walk in a community (STPP 2002; Ernst 2004).
According to this STPP index, in 2000-2001 the Denver-Boulder-Greeley
consolidated metropolitan statistical area ranked 23rd of 49 metropolitan areas in
9


pedestrian danger, three spots worse than its ranking of 26th in 1997-1998. The
Denver-Boulder-Greeley metropolitan area ranked fourth among U.S. metropolitan
areas in terms of greatest declines in pedestrian safety between 1994-1995 and 2002-
2003 (Ernst 2004). Likewise, a 2003 report published by the National Center for
Statistics and Analysis of the National Highway Safety Traffic Administration
(NHSTA) ranked Denver 12th of 245 U.S. cities in pedestrian fatality rates per
100,000 population with a rate of 4.21 (NCSA 2003). Despite their difference, all the
indices indicate that pedestrian fatalities are a serious concern in Denver.
Interestingly, in both the STPP and the NCSA rankings, most of the cities that rank
the worst in pedestrian safety are located in the South and the West of the United
States (NCSA 2003; STPP 2002; Ernst 2004). Many of these cities tend to have
lower density development patterns and large numbers of wide, high-speed arterial
streets (STPP 2002; Ernst 2004). Arterial streets are those streets designed to provide
a high degree of mobility to, from, and within urban areas (City & County of Denver
2006). These findings underscore the importance of understanding the geographical
and spatial characteristics of pedestrian accidents.
While FARS data provide useful insights into pedestrian accidents, one must
remember that the database tracks only pedestrian fatalities. Several studies indicate
that pedestrian fatalities show biases towards certain types of pedestrian/motor
vehicle accidents, such as those involving high velocity or head-on impact (Anderson
et al. 1997). Therefore, FARS data provide only a limited amount of information
about pedestrian accidents in general.
Non-fatal pedestrian accidents should be a public health concern, as well. Pedestrian
injuries are up to 16 times more common than are fatalities (Ewing et al. 2003). As a
result, non-fatal pedestrian accidents may play a significant role in peoples
10


perceptions of safety and walkability. Thus, an analysis of all pedestrian accidents,
fatal and non-fatal, at the local level should produce information beneficial to the
public health, public safety, and urban planning sectors.
While pedestrian accidents show uneven geographic distribution, there also appear to
be disparities in who suffers the greatest burden of morbidity and mortality as a result
of pedestrian/motor vehicle accidents. Globally, those from poorer socio-economic
groups experience greater all-cause injury risk, including injuries from
pedestrian/motor vehicle collisions (Nantulya & Reich 2002; Odero et al. 2003;
Evans & Brown 2003; Roberts & Power 1996). Pedestrian and bicycle accidents
among children in Montreal show markedly lower rates in higher-income census
tracts (Joly et al. 1991). Like many other public health issues in the United States,
pedestrian fatalities show disparities according to race/ethnicity. Nationally, blacks
account for more than 20% of all pedestrian fatalities, yet only about 12% of the total
population (STPP 2002). In Atlanta from 1994 to 1998, the pedestrian fatality rate
for Hispanics was 6 times greater than the rate for non-Hispanic whites. For non-
Hispanic Blacks, the pedestrian fatality rate was 2 times the rate of non-Hispanic
whites (Hanzlick et al. 1999). Newspaper investigative reports from Washington,
D.C. and Los Angeles also indicate a greater than expected number of pedestrian
deaths among Latinos relative to their proportion of the population (Marosi 1999;
Moreno & Sipress 1999). These studies indicate that the burden of morbidity and
mortality for pedestrians is not shared equally among different populations with in the
United States.
Demographic patterns of pedestrian fatalities in Colorado during 2002 mirrored
patterns of disparities noted in other reports. Of the seventy-eight pedestrians killed
on public roadways in Colorado during 2002, persons aged 20 years or older
represented more than three-quarters of the pedestrian fatalities. Twenty of the 57
11


fatalities for which ethnicity information was available were classified as Hispanic
(see Table 1.2). In the City and County of Denver, nearly half (11 of 23) of
pedestrian fatalities with ethnicity information available occurred among Hispanics
despite Hispanic being just over 30% of the total population (USCB 2004).
Throughout the state, males were disproportionately represented in pedestrian
fatalities. Males accounted for just over 67% of pedestrian fatalities in the City and
County of Denver, while in Colorado as a whole just over 69% of pedestrian fatalities
were males. Although these analyses of pedestrian fatalities are interesting and
certainly important, a broader analysis that includes all pedestrian accidents should
provide additional insight into pedestrian safety.
Injuries to pedestrians resulting from accidents with motor vehicles are classified as
unintentional injuries. Unintentional injuries are distinct from intentional injuries,
such as homicide or suicide, where the intent is to cause harm (CDPHE 2002). While
unintentional injuries are commonly referred to as accidents, in reality many of
these events can be prevented (CDPHE 2002). This is true for pedestrian injuries as
well as for other types of unintentional injuries. Unintentional injuries are the leading
cause of death in the United States for those aged 1-34 years old, whereas in Colorado
unintentional injuries are the leading cause of death for those aged 1-44 years old
(CDPHE 2002). As with numerous other health conditions, both unintentional injury
and mortality rates show disparities by socioeconomic status (Roberts & Power 1996;
Marcin et al. 2003). A retrospective study from California revealed that children
from neighborhoods with lower socioeconomic status had both higher injury
hospitalization and higher injury mortality rates than children from neighborhoods
with higher socioeconomic status (Marcin et al. 2003). One of the main goals of both
the national Healthy People 2010 Initiative and the local Healthy Denver 2010
Initiative is to eliminate health disparities among segments of the population,
including differences that occur by gender, race or ethnicity, education or income,
12


disability, geographic location, or sexual orientation (Dept, of Environmental Health
2002). This goal applies to disparities in pedestrian injuries as well as to other health
conditions.
Preliminary analysis of the FARS data available for Colorado suggests that the state
mirrors the trends in disparities and geographic variations presented in the literature.
The literature review turned up no studies of pedestrian accidents or safety in Denver,
however. Yet, studies from other cities in North America recommend that it is
important to look at variations in pedestrian accidents and safety at the local level
(e.g., Hijar et al. 2003; Joly et al. 1991). Studying the phenomena at the local level
should provide information that will help improve pedestrian safety and reduce
unintentional injuries and fatalities from pedestrian/ motor vehicle accidents.
Obesity and Diabetes
The obesity and diabetes rates in the United States are increasing dramatically, both
among adults and children. Complications from obesity are now the second leading
cause of mortality among U.S. adults behind only smoking, which it is rapidly
overtaking (Allision et al. 1999). Self-reported obesity, a body mass index (BMI)
equal to or greater than 30 and determined by dividing weight in kilograms by height
in meters squared, among adults increased 61% from 1991 to a 2000 prevalence rate
of 19.8% (Mokdad et al. 2001). This represents a total of 38.8 million adults in the
United States (Mokdad et al. 2001). And these numbers may actually underestimate
the true magnitude of the problems since self-reported obesity data tends to produce
lower numbers than estimates based on measurement (Mokdad et al. 1999). Results
from the 1999-2000 National Health and Nutrition Examination Survey (NHANES),
a survey that measures height and weight among other health indicators in a mobile
13


examination center, indicate that the prevalence rate of obesity in U.S. adults might
be as high as 30.9% (Flegal et al. 2002). Although Colorado historically has had the
lowest rate of obesity among the 50 states, the obesity rate in the state is increasing
just as it is in every other state (Mokdad et al. 2001). Furthermore, these numbers do
not reflect the number of adults who are overweight, only those already classified as
obese. By 1994 approximately 63% of men and 55% of women aged 25 years or
older were classified as overweight in the United States (Must et al. 1999).
The growing obesity rates in the United States are not limited to adults; children are
also experiencing rapid increases in obesity. Childhood obesity in the United States
is at its highest rate ever. Moreover, an estimated 16% of children between the ages
of 6 and 11 years old are considered overweight and an additional 14% are at risk of
becoming overweight (St Onge et al. 2003). Obesity in children is not simply an
image concern. Overweight children experience elevated blood pressure, fasting
insulin, and cholesterol concentration levels when compared to normal-weight
children (St Onge et al. 2003). Furthermore, obese children experience increased
morbidity as adults, including chronic disease (Must et al. 1999; Freedman et al.
1999). Clearly, increasing body weight has become a major medical and public health
issue for all ages in the United States.
The increase in the diabetes epidemic in the United States mirrors the extraordinary
growth of the obesity epidemic. While the prevalence of adult obesity increased 61%
from 1991 to 2000, the prevalence rate of self-reported diabetes also increased
dramatically up 49% between 1990 and 2000 (Mokdad et al. 2001). By 2000,
approximately 15 million adults in the United States reported they had been
diagnosed with diabetes (Mokdad et al. 2001). Rates of diagnosed adult diabetes in
Colorado are similar to those of the nation as a whole with about 6% of the adult
population reporting having been diagnosed with diabetes (Mokdad et al. 2001). The
14


public health community has a vast interest in curbing the rising rates of obesity and
diabetes in the Colorado and the United States in an effort to improve health and
decrease health-care costs.
As with pedestrian/motor vehicle accidents and other health conditions, obesity and
diabetes show disparities by ethnicity. Blacks and Hispanics suffer a disproportionate
burden from these health conditions. Among respondents to the 2000 Behavioral
Risk Factor Surveillance System (BRFSS), 29.3% of Blacks and 23.4% of Hispanics
reported being obese compared to 18.5% of Whites (Mokdad et al. 2001). Similarly,
Blacks and Hispanics reported higher rates of diabetes, 11.1% and 8.9% respectively,
compared to Whites, 6.6% (Mokdad et al. 2001). Increases in diabetes and obesity
are seen among all ethnic groups, age groups, and sexes, however, despite showing
disparities in the total prevalence rates (Fiegal et al. 2002). The same population
groups that bear the burden of excessive morbidity and mortality from pedestrian/
motor vehicle accidents suffer disproportionately from obesity and diabetes.
Researchers focusing on obesity and diabetes often recommend increasing physical
activity, especially walking, as a mechanism to prevent these health conditions and
reign in the rising epidemics. They advocate walking as a simple activity to be
incorporated into everyday activities to help control the rise in obesity and diabetes
(e.g., Koplan & Dietz 1999). Doing so, however, may put people at increased risk of
injury or fatality as a pedestrian. As walking is promoted, corresponding
improvements in pedestrian safety need to be made as well.
To date there has been little discussion on risk and perception of risk as a factor in
peoples ability and willingness to walk. Mokdad and colleagues write, ... urban
policymakers must provide more sidewalks, bike paths, and other alternatives to cars
while Hill and Peters argue for the importance of recognizing and dealing with
15


environmental contributions to the obesity epidemic (Mokdad et al. 2001:1199; Hill
& Peters 1998). Yet, for all the suggestions for changing the environment in ways
more conducive to physical activity, there is little mention of pedestrian safety (Hill
& Peters 1998). Even Wickelgren, who argues that the focus of public health efforts
should really be on increasing physical activity instead of on obesity per se, does not
mention anything about the safety or danger of walking as that physical activity
(1998). As we advocate walking as a way to combat disease and promote health, we
must ensure that we make walking as safe as possible. Otherwise, people are trading
one set of risks for another and may be discouraged or disinclined to walk.
Walkabilitv
Walking, viewed as a basic physical activity that most adults and children can
incorporate relatively easily and inexpensively into their everyday lives, provides
health benefits even at modest exercise levels. As a result, the public health
community has become increasingly interested in walking as a physical activity that,
if done on a large-scale, regular basis, may help curb the tremendous increases in
obesity rates and associated health conditions such as diabetes, heart disease,
hypertension, and stroke in the United States. Although the public health literature
contains only limited information specifically about pedestrian accidents and safety,
there is a large body of literature on walking as physical activity and what is broadly
referred to as walkability those features of safety, convenience, and appeal that
seem to increase the likelihood of people walking in any given location.
Numerous studies have tried to operationalize the characteristics that contribute to the
walkability of a geographic area. While each study includes a different set of
variables in its walkability index, they all incorporate variables aimed at measuring
16


general categories such as physical features, social characteristics, safety, and
aesthetics (e.g., Brownson et al. 2001; Humpel et al. 2002; Craig et al. 2002; Saelens
et al. 2003). Saelens and colleagues, for example, measure eight characteristics:
residential density, land use mix-density, land use mix-access, street connectivity,
walking/cycling facilities, aesthetics, pedestrian/automobile traffic safety, and crime
safety (2003). Craig and colleagues, on the other hand, measure a total of eighteen
characteristics (2002). While the number and definition of characteristics changes
from index to index, pedestrian accidents and safety are a component of all these
constructions of walkability.
Healthy People 2010 includes two goals around increasing the proportion of children
walking and bicycling to school (U.S. DHHS 2001). As a response, public health
campaigns across the nation are promoting the idea of children walking or bicycling
to school. These efforts aim to improve the health of the nations children, including
curbing the rising rates of obesity and diabetes among youth. Pedestrian and bicycle
safety appears to be a very real concern for parents, however. In a survey conducted
by the U.S. Centers for Disease Control and Prevention (CDC) about barriers to
children walking or biking to school, nearly 40% of parents cited traffic as a major
barrier. This ranked traffic second of all barriers, behind only distance (Dellinger &
Staunton 2002). On the other hand, children whose parents listed no barriers to them
walking or bicycling to school were 6 times more likely to walk or bike to school than
children whose parents listed one or more barriers (Dellinger & Staunton 2002). In
response to the perception of traffic as a danger to young pedestrians, the concept of a
walking school bus or a bike train has been promoted in various cities (e.g.,
Bricker et al. 2002; Staunton et al. 2003; Walk to School Day USA 2004; National
SAFE KIDS Campaign 2004). A walking school bus refers to a group of students
and parents walking to and from school together for safety and social support.
Similarly, a bike train involves a group of students riding their bikes together to and
17


from school. Clearly, safety and perceived safety is an important component of
increasing pedestrian activity.
While the walkability literature does incorporate some aspects of pedestrian safety,
especially perceptions of safety and danger, the issues seem to be more complex than
has been dealt with in the walkability literature. For example, two sample items help
measure perceived pedestrian/ automobile traffic safety subscale in Saelens and
colleagues (2003) Neighborhood Environment Walkability Scale (NEWS): 1) The
speed of traffic on most nearby streets is usually slow (30mph or less); 2) There are
crosswalks and pedestrian signals to help walkers cross busy streets in my
neighborhood. A similar scale about perceived safety from traffic utilized 4 items: 1)
there is so much traffic along the street I live on or nearby that it makes it difficult or
unpleasant to walk in my neighborhood; 2) the speed of traffic on the street I live on
nearby is usually slow (40kph or less); 3) most drivers exceed the posted speed limits
while driving in my neighborhood; and 4) when walking in my neighborhood there
are a lot of exhaust fumes (De Bourdeaudhuij et al. 2003). These items focus almost
entirely on perceptions of safety around walking.
While these are certainly important items to consider, they do not capture the entire
complexity of pedestrian safety nor do they account for some of the knowledge about
pedestrian accidents that is available in the public health literature. Joly and
colleagues discovered that among children in Montreal, pedestrian accidents actually
occur most often under good traffic safety conditions good visibility and on
straight, residential streets with lower speed limits (1991). Furthermore, Langlois and
colleagues found that while crosswalks and pedestrian signals are necessary for
pedestrian safety among older adults, they are not sufficient for safety, as less than
1% of the pedestrians aged 72 or older in their study could cross the street within the
amount of time allotted by the traffic signal (1997). These examples highlight the
18


need to consider pedestrian accidents and safety independently, not just as a part of
walkability. They also point to the need to investigate objective measures of
pedestrian safety, not just perceptions of safety.
Built Environment/ Neighborhood Level Effects
As alluded to in the discussion of walkability, neighborhood design and the built
environment play an important role in the incorporation of walking as transportation
for everyday activities. Walking as a mode of transportation, not just for exercise,
may increase ones overall physical activity level, thereby contributing to improved
health status. In a survey of physical activity and weight status among adults living in
neighborhoods classified as highly walkable and those classified as having low
walkability, Saelens and colleagues found no difference in leisure time physical
activity (2003). Adults residing in high walkability neighborhoods reported higher
levels of overall physical activity, however, due to their incorporation of walking for
everyday activities (Saelens et al. 2003). De Bourdeaudhuij and colleagues report
similar findings from Belgium (2003). While vigorous physical activity was
associated with activity amenities in the home and convenient activity facilities
outside the home regardless of neighborhood design, minutes of walking and
moderate physical activity were associated with characteristics all related to
neighborhood design quality of sidewalks, accessibility of shopping, and
accessibility of public transportation (De Bourdeaudhuij et al. 2003). Addy and
colleagues also found that having sidewalks available in the neighborhood was
associated with increased walking behaviors (2004). For those where sidewalks were
not readily available, using a mall for walking was associated with increased walking
activity (Addy et al. 2004). Clearly, ones immediate physical surroundings either
encourage or deter one from walking as exercise or transport.
19


Perceived physical safety and violent crime in a geographic area may also contribute
to the conduciveness of an environment for walking or other physical activity. For
example, results from the BRFSS indicate that those who perceived their
neighborhood to be less safe are more likely to be physically inactive. This
association was strongest for those 65 years or older and for racial/ethnic minorities
(Weinstein et al. 1999). Community members perceptions of pedestrian safety and
danger may reflect issues of physical safety and violent crime as well as issues related
to traffic and the built environment. These findings about the association between
neighborhood design/ built environment and increased walkability support Koplan
and Dietzs (1999:1580) call that automobile trips that can be safely replaced by
walking or bicycling offer the first target for increased physical activity in
communities. In order to heed this call, however, more attention needs to be paid to
pedestrian safety and accidents as a public health issue both as part of the larger
concept of walkability and as unintentional injury prevention.
Livable Communities/ Urban Sprawl
Over the past few decades many American cities have developed in a largely
uncontrolled, unmanaged fashion that has resulted in extensive, low-density
development sprawl. By 2000 land development was roughly double what it had
been a decade before (Chen 2000). Out of a multitude of local issues, Americans said
they were most concerned about sprawl and traffic (Chen 2000). Sprawl can be
defined as an environment characterized by 1) a population widely dispersed in low-
density residential development; 2) rigid separation of homes, shops, and workplaces;
3) a lack of distinct, thriving activity centers; and 4) a network of roads marked by
very large block size and poor access from one place to another (Ewing 1997; Ewing
20


et al. 2003). As many cities have grown rapidly, their sprawling development
requires reliance on the automobile with little or no regard for pedestrians or
walkability.
Therefore, it is probably no coincidence that many U.S. cities with the highest rates of
pedestrian fatalities year after year are younger cities in the South and West of the
country cities whose development and infrastructure boom came after the advent of
the automobile. Eight out of ten U.S. cities with the highest pedestrian fatality rates
are in the South, while five of the top twenty are in the West including Denver and
Pueblo in Colorado (NCSA 2003). Ewing and colleagues computed a sprawl index
in order to determine associations between sprawl and both motor vehicle occupant
and pedestrian fatalities. In general, the more compact, less sprawling a city was the
fewer motor vehicle and pedestrian fatalities it had (Ewing et al. 2003). Not
surprisingly, except for one, the ten cities with the most compact urban forms
according to the sprawl index are located in the Northeastern portion of the United
States. These cities also have relatively low rates of all-mode traffic fatality
compared to cities with less compact urban form (Ewing et al. 2003). As Ewing and
colleagues conclude, .. .sprawl is a significant risk factor for traffic fatalities,
especially for pedestrians (2003: 1544). More compact urban design may contribute
to more locations where residents can walk to accomplish everyday errands as well as
safer environments for those who do choose to walk.
Much of the available literature on pedestrian accidents and safety resides in the
urban planning and transportation literature, although more is starting to appear in
public health as awareness of the importance of the issue increases. While issues of
urban development including pedestrian safety have traditionally been relegated
to the domain of urban planning and transport, Frumkin (2002) and others (e.g.,
Peden et al. 2004) argue that pedestrian safety represents an issue that should be at the
21


core of public health. As Frumkin reminds readers, public healths foundation was in
broad-scale interventions that benefited the population, such as sanitation and
immunizations. Over the years, public health has become less engaged in issues such
as urban design and development as the medical model of focusing on the individual
has become more prevalent in the discipline. Such automobile-dependent
development has numerous direct and indirect influences on health, however,
including pollution, motor vehicle crashes, pedestrian accidents, and a more sedentary
lifestyle (Frumkin 2002). Similarly, Peden and colleagues argue that the health sector
has much to gain from being involved in road injury prevention (2004). Frank and
Engelke, two urban planners, note that urban planning and public health originate
from similar roots of creating safe, livable environments for the population (2001).
These authors advocate for public health playing a larger, more involved role in the
prevention of pedestrian accidents and in building safe environments for pedestrian
activities a role that would return to the roots of public health.
Conclusion
A review of the literature highlights some critical points with regards to pedestrian
accidents and safety. First, pedestrian/motor vehicle accidents are a significant cause
of unintentional injury globally, domestically, and locally. So far, most research into
pedestrian accidents has looked exclusively at fatalities. This research has shown
disparities in pedestrian fatalities at all geographic units and by various socio-
demographic characteristics. Much less research looks at non-fatal accidents or all
pedestrian/motor vehicle accidents. Second, obesity along with its associated health
conditions, such as diabetes, is on the rise in the United States among both adults and
children. As a result, numerous public health interventions now promote walking as a
form of exercise or transportation for everyday activities in an effort to combat this
22


rise in obesity and its associated health conditions. Third, the walkability literature
mentions pedestrian safety, especially perceptions of pedestrian safety, as part of the
operationalization of walkability. Although the built environment is extremely
important for pedestrian safety and the promotion of walking, many neighborhoods
and cities in the United States are not designed to facilitate safe pedestrian activity.
23


CHAPTER 2
THEORETICAL PERSPECTIVES
Acknowledging and articulating theoretical assumptions helps guide ones research,
process the analyses, and situate the findings within the literature. The issues
surrounding pedestrian accidents and safety are complex and multidimensional.
Therefore, theoretical perspectives must provide guidance for integrating these issues
in order to build a more comprehensive understanding of pedestrian accidents and
safety and to enhance interventions designed to increase pedestrian safety, to promote
walking and physical activity, and to prevent unintentional injury.
This study is based on the fundamental premise that place matters. Although public
health does not yet have the detailed theoretical models at the environmental level
that provide specific constructs and expected mechanisms of effect for neighborhood
or environmental influences on health and safety like it does at the individual level,
studies from a number of disciplines have shown that environmental factors do
influence human behavior and health. For example, one study showed that
individuals living in the most deprived neighborhoods were significantly shorter and
had larger waist circumferences, waist-hip ratios, and BMIs than individuals living
elsewhere (Eliaway et al. 1997). Other studies have found that opportunities for
physical activity and eating a healthy diet vary between different types of
neighborhoods (MacIntyre et al. 1993; Glyptis 1989). As noted in the literature
review section, a number of papers have analyzed differences in motor vehicle and
pedestrian/ motor vehicle accidents by various geographic areas (see, for example,
Joly et al. 1991; Hijar et al. 2003; Peden et al. 2004). This previous research provides
strong evidence that spatial factors are important influences on individual and
24


population health and safety despite the lack of theoretical models that help frame
such relationships.
The role that the socio-cultural and socio-demographic environments play in the
creation of healthy or unhealthy populations has received increased attention in public
health over the past decade. It is essential that the same type of awareness about the
importance of the physical and built environment also now occur. The disciplines of
geography and urban planning have been the leaders in such thinking about how the
physical and built environment can and does affect human behavior (e.g., Loukaitou-
Sideris 2006; Duncan et al. 1999; Fletcher & McMichael 1997; Frank et al. 2003).
This is the same type of thinking that was central to the early origins of public health,
but which in more recent decades has been overshadowed by a more individualistic
approach. It is important that public health continue to broaden its more narrow focus
of the recent decades as it has been doing in the areas of walkability and livable
communities and expand research and intervention efforts that deal with the
physical and built environments.
Some argue that social factors are likely the fundamental causes of disease and
advocate for the importance of researching and understanding these social factors as
fundamental causes (Link & Phelan 1995). These authors pose the question what
puts people at risk of risks? in order to spur consideration of social factors as
fundamental causes of disease. This type of thinking should be applied to the built
environment, as well.
Individually-based risk must be considered within both the larger social context and
the broader environmental context, especially when dealing with injury and safety.
Accidents and injuries, or the lack of these events, occur within a larger spatial
context in which the physical and built characteristics of this larger spatial context
25


place people at risk directly and at risk of risks. Just as it would at the social level,
understanding fundamental causes of disease and injury at the environmental level
may help address multiple diseases or injury outcomes at a single time. Furthermore,
effective interventions and modifications targeting fundamental causes can result in
the largest population impact using the lowest per capita investment of resources.
Following the premise that place matters, this study considered characteristics of both
the social environment and the built environment at various geographic units of
analysis. If the social and built environments are fundamental causes of disease and
injury, then the influences of these fundamental causes on disease and injury might be
operationalized through more proximal causes that are specific characteristics of the
social and built environments. Specific characteristics of the social environment
analyzed in the study include population density, age distribution, ethnic composition,
economic indicators, educational attainment and mode of transportation to work.
Specific characteristics of the built environment investigated include street structure,
density of liquor license establishments, and land use.
Looking at a combination of place-based characteristics of the social environment and
of the built environment seems especially important for a topic like pedestrian
accidents and safety. Every pedestrian/ motor vehicle accident in this study involves,
at a minimum, a human-made vehicle colliding with a person on a public roadway
within the context of a multitude of physical conditions. Aspects of the social
environment help determine who the pedestrians are, when they must be walking,
under what conditions, and for what purposes. While some have argued that
geographic or spatial variations in outcome measures may simply be the result of
particular types of people linked to the outcome of interest being more common in
certain places than others (Duncan et al. 1999), this argument seems somewhat less
convincing for pedestrian/ motor vehicle accidents because it is not just people who
26


are involved and because characteristics of the location likely affect the accident or
the probability of an accident as well.
Compositional and contextual effects provide another way to conceptualize
characteristics of the social and built environments when considering pedestrian/
motor vehicle'accidents in a place-based manner. Compositional effects are ...
differences in the kinds of people who live in these places with the implication that
these differences would be retained regardless of where the people live (MacIntyre &
Eliaway 2005:24). While this study cannot make the assumption that those
pedestrians involved in accidents with a motor vehicle reside in the census block
group or census tract where the accident occurred, the compositional effects of a
geographic area may influence pedestrian activity in that area and what types of
people are walking for what purposes. Compositional effects may also influence
public investment in pedestrian-related infrastructure and enforcement environment
as well as access to resources, including advocacy resources, to create a safer
pedestrian environment. Compositional effects in this study would be indicated
through the variables derived from census data representing characteristics of the
social environment: population density, age distribution, ethnic composition,
economic indicators, educational attainment, and mode of transportation to work.
On the other hand, contextual effects are those ... differences between the places
themselves with the implication that ... death rates [or accident rates] of...
individuals will vary depending on what sort of area they live in (MacIntyre &
Ellaway 2005:24). Contextual effects in this study would be indicated through the
variables representing characteristics of the built environment: density of liquor
license establishments, place-based street structure, and land use. Contextual effects
might influence pedestrian accidents and safety through creating places where 1)
pedestrians or motor vehicle-drivers under the influence of alcohol meet more often,
27


2) the street structure is geared towards high-volume, high-velocity motor vehicle
traffic in which pedestrian use is a lower priority, or 3) land use brings motor vehicles
and pedestrians into more frequent contact.
In their discussion of contextual and compositional effects on health and illness,
MacIntyre and Ellaway point out that the distinction between the two is somewhat
artificial, however, because people create places and places create people
(2005:26). This point seems particularly applicable to pedestrian/ motor vehicle
accidents. As noted above, both contextual and compositional effects of the
environment likely play an important role in determining the safety, or lack thereof,
of the pedestrian environment. These effects are likely highly integrated with each
other. Just as ... a crucial thing that income, wealth, education, and high social class
help people to buy is a residence in areas with pleasant environments, good schools
and other schools, and low crime rates (MacIntyre & Ellaway 2005:34), these
characteristics may also help buy safer places to walk or play and alternative modes
of transportation.
This research suggests that place matters in determining health status. Although a
single theoretical model does not exist that defines the constructs and mechanisms for
how spatial factors influence health, substantial evidence points to the important role
of environmental factors. King and colleagues note that while there is substantial
evidence of the relationship between the environment, physical activity, and health,
the field still lacks trans-disciplinary meta-theory to frame these relationships
(2002). Therefore, continuing exploration into the role that environmental factors
have on health and safety should contribute to enhancing the understanding of how
and why place matters for health.
28


Liquor License Establishments and Public Health Outcomes
Over the past decade much research has focused on the relationship between the
density of alcohol outlets within a geographic area and various health outcomes in
those same areas. While these studies are similar in their general focus of analyzing
the relationship between alcohol outlet density and health outcomes, they vary in the
type of health outcome studied, the geographic unit of analysis, the handling of
spatial autocorrelation in the data, and the measurement of alcohol outlet density. To
date, the results of these research studies have been mixed with some, but not all,
studies finding statistically significant positive associations between alcohol outlet
density and health outcomes within geographic areas.
The health outcomes investigated in these research studies include motor vehicle-
related crashes and injuries (Scribner et al. 1994, Meliker et al. 2004, Escobedo &
Ortiz 2002, LaScala et al. 2001), gonorrhea rates (Cohen et al. 2006), birth outcomes
(Farley et al. 2006), violence (Gruenewald et al. 2006, Scribner et al. 1999, Lipton &
Gruenewald 2002, Escobedo & Ortiz 2002, Reid et al. 2003), and child abuse and
neglect (Freisthler et al. 2004). Three of the studies analyzing motor vehicle-related
crashes and injuries found statistically significant positive associations with alcohol
outlet densities, including one study on pedestrian injuries (Scribner et al. 1994,
Escobedo & Ortiz 2002, LaScala et al. 2001). A more recent study failed to replicate
these earlier findings, however (Meliker et al. 2004).
The studies analyzing the relationship between the density of alcohol outlets and
health outcomes have utilized a variety of geographic units for analysis. Significant
positive associations have been found at geographic units of analysis ranging from
census tracts and zip codes (Cohen et al. 2006, Gruenewald et al. 2006, Freisthler et
al. 2004, Tatlow et al. 2000, Reid et al. 2003) to cities and counties (Scribner et al.
29


1994, Escobedo & Ortiz 2002). Of the three studies that found statistically significant
positive associations between alcohol outlet density and motor vehicle-related crashes
and injuries, one analyzed geographic units by city (Scribner et al. 1994), one by
neighborhoods (LaScala et al. 2001), and one by county (Escobedo & Ortiz 2002).
The study that failed to replicate these earlier findings utilized census block groups as
the geographic unit of analysis (Meliker et al. 2004).
The research studies on alcohol outlet densities and health outcomes also have
differed in whether they utilized statistical controls or adjustments for the spatial
relationships inherent in the geographic data. Two studies analyzing motor vehicle-
related crashes and injuries utilized statistical controls or adjustments for the spatial
relationship in the data (LaScala et al. 2001, Meliker et al. 2004), while two others
did not (Scribner et al. 1994, Escobedo & Ortiz 2002). There does not appear to be a
relationship between studies that utilized statistical techniques to account for spatial
autocorrelation and those that found statistical significance.
The measurement of alcohol outlet density also varied between the studies. Some
studies simply utilized total alcohol outlets as the predictor while other studies
divided alcohol outlets into the type of liquor license such as outlets that sell alcohol
for on-premise consumption versus off-premise consumption. A few studies further
sub-divided alcohol outlets into classifications such as bars or restaurants (Scribner et
al. 1994, LaScala et al. 2001, Gruenewald et al. 2006, Lipton & Gruenewald 2002,
Freisthler et al. 2004). One of the studies about motor vehicle-related crashes or
injuries just looked at the density of bars (LaScala et al. 2001). Another just looked at
the density of total alcohol outlets (Escobedo & Ortiz 2002); two others looked at
both the total density and the density of specific types of alcohol outlets (Scribner et
al. 1994, Meliker et al. 2004). As with statistical techniques for spatial data, there
30


does not seem to be a pattern between studies that used total outlets or sub-divided
type of outlet and whether they found a statistically significant association.
Nearly all of the research studies on the density of alcohol outlets and health
outcomes are associational and based on cross-sectional, ecological data. Given that
these studies have produced mixed results, it is important to consider what the link or
mechanism along the causal pathway between alcohol outlets in a geographic area
and health outcomes would be. The link may be direct in that a greater density of
alcohol outlets signifies a greater consumption of alcohol in the area leading to more
drivers and pedestrians who have been drinking and whose judgment might be
impaired. Alternatively, the link may be indirect.
A number of potential theories and ideas have been proposed to articulate such
indirect links. These theories include social normative theories and theories of social
disorganization. Social norms are those rules and standards that guide what is
acceptable and unacceptable behavior within a society or group of people (National
Social Norms Resource Center, 2006). Social norms can be either behavioral the
most common actions or behaviors displayed by a social group or attitudinal the
most widely shared beliefs or expectations within a social group about how one ought
to behave under certain circumstances (National Social Norms Resource Center,
2006). Cohen and colleagues (2006) propose that one way the presence of alcohol
outlets in a neighborhood may affect high risk behaviors is because they change the
social norms around those high-risk behaviors. In such a scenario, neighborhoods
with a higher density of alcohol outlets may also see higher rates of unsafe driving
habits, pedestrians ignoring traffic regulations and/or the level and frequency of
public intoxication because those high-risk behaviors might be seen as more socially
acceptable.
31


Theories of social disorganization include Broken Windows theory. Such theories
suggest that the appearance of the physical environment signals how effectively, or
ineffectively, informal social controls are able to influence individual behavior in the
neighborhood. Broken windows that are not repaired in a neighborhood signal that
people do not care about or are not able to control what happens around them (Wilson
& Keeling 1989). Lipton and Gruenewald (2002) apply this idea to alcohol outlet
density proposing that bars and broken bottles in a neighborhood serve the same
function as broken windows by indicating social disorganization and a lack of social
control. In terms of linking the density of alcohol outlets to pedestrian/ motor vehicle
accidents, Broken Windows theory would suggest that an increased density of alcohol
outlets in a geographic area signals a lack of formal and informal enforcement of
traffic safety and regulations as well as greater acceptance of public intoxication.
In addition to the above theories, several authors have proposed the idea that alcohol
outlet density might simply be a proxy for other retail activity in the area
(Gruenewald et. al. 2006, Scribner et. al. 1999, Farley et. al. 2006). Retail activity in
general would bring people and motor vehicles together in more dense, public spaces.
A corresponding increase in health outcomes such as violence and motor vehicle
accidents may simply be a result of the denser congregation of people. While this
idea is conceptually intriguing, two studies that included measures of retail outlet
density in addition to alcohol outlet density failed to find a significant association
between their measure of retail outlet density and the health outcome of interest
(Scribner et. al. 1999, Farley et. al. 2006).
Much research in recent years has focused on the relationship between the density of
alcohol outlets and health outcomes within specified geographic areas. Since most of
this research has been associational and has produced mixed results, it is important to
consider what the link in the causal pathway between alcohol outlet density and
32


health outcomes may be. The causal mechanism may be direct, indirect, or a
combination of the two. A number of theories and ideas have been proposed to
explain potential indirect links. Particularly for an outcome like such as pedestrian/
motor vehicle accidents that involves people, moving vehicles, and the physical
environment, it seems likely that if there is a relationship between the density of
alcohol outlets and accidents the link is direct or a combination of direct and indirect
mechanisms. Regardless, previous research suggests that it is important to include
the density of alcohol outlets as one potential predictor in any model of the patterns of
pedestrian/ motor vehicle accidents. Therefore, this study includes the density of
liquor license establishments as one measure of the built environment.
Social Ecology
Social ecology provides one framework that integrates both social and environmental
influences for helping to analyze data and organize thinking around pedestrian
accidents and safety. Social ecology is not a detailed theoretical model with specific
variables and constructs. Instead, it is a general model, framework, or perspective
(Sallis & Owen 1997). As such, it provides an organizational tool or heuristic device
for thinking about and analyzing issues within a multi-level context.
Social ecology arose out of human ecology. It differs from human ecology, however,
by placing more emphasis on the social, institutional, and cultural context of peoples
lives as opposed to mainly on the biological processes and human-environment
interaction (Stokols 1992). Social ecology goes a step beyond traditional
environmental or ecological models by recognizing the importance of social and
cultural variables as well as ecological ones. Furthermore, social ecology posits that
social and ecological environments interact with individual factors to influence health
33


status while individual behavior also can modify the social and ecological
environments.
While health promotion in recent decades has emphasized individual-level models of
behavior change, ecological models of health are, historically, at the heart of public
health. Just as a traditional hallmark of public health is its focus on populations as
opposed to solely on individuals, considering the role of the environment and
intervening at the environmental level through policies and regulations also has been
a fundamental focus of public health (McLeroy et al. 1988; Green & Kreuter 1999).
The renewed interest in the role of environmental factors in health and safety returns
to the roots of public health.
Such an ecological approach to public health also allows for the reframing of a
problem. For example, Egger and Swinbum (1997:478) argue, [a]n ecological
approach regards obesity as a normal response to an abnormal environment, rather
than vice versa. While .. .ecological approaches in health promotion view health as
a product of the interdependence of the individual and subsystems of the ecosystems
(such as family, community, culture, and physical and social environment) (Green &
Kreuter 1999:22), social ecology places more emphasis on the social, institutional,
and cultural contexts of people-environment relations (Stokols 1992). A social
ecological framework allows for the integration of both the social environment and
the built environment.
Although social ecology has received increased attention in health-related disciplines
in recent years, some researchers have been using it for decades. Social ecology has
been applied to health at least since the mid-1970s when it was utilized in health
psychology (e.g., Moos 1979). Health promotion has used social ecology since at
least late 1980s (e.g., McLeroy et al. 1988, Stokols 1992, Stokols 1996; Green et al.
34


1996; Breslow 1996). It continues to be an important framework in health-related
disciplines today.
As social ecology is a flexible framework, various authors have defined the numbers
and labels of levels differently within the models of social ecology that they have
used. Some models are more complex and involve more levels than other models.
Sallis and Owen, for example, discuss a relatively simple framework of personal,
social, and environmental levels in ecological models (1997). On the other hand,
McLeroy and colleagues (1988) describe the interaction of five levels as determining
health and behavior: 1) intrapersonal factors, 2) interpersonal processes and primary
groups, 3) institutional factors, 4) community factors, and 5) public policy. Corbett
(2001) identifies five levels, as well, but defines the five levels differently: 1)
individual, 2) groups or social networks, 3) organizational, 4) community, and 5)
population. A taskforce on food and physical activity choices developed a model that
outlines seven layers of determinants: psychobiological core, cultural, social,
enablers, behavioral settings, proximal leverage points, and distal leverage points
(Booth et al. 2001; Wetter et al. 2001). The number and definition of levels in a
social ecological framework seems to depend on the health issues involved as well as
on author preferences.
Some of the levels referenced above may be more or less relevant to pedestrian safety
while other levels not yet identified may also prove important in the future. For
example, psychobiological core would not seem applicable to pedestrian accidents
and safety. Regardless of the number of levels or definition of labels, however, the
principle remains the same health and safety are products of the numerous factors
interacting at various levels. Social ecology is a flexible and fluid approach that can
provide a common framework for interdisciplinary dialog and for the linking of
theory, research, and practice.
35


Reciprocal determinism is a critical component of any ecological model. It is the idea
that the environment influences behavior just as behavior can influence the
environment (McLeroy 1988). Reciprocal determinism has two major implications
for behavioral and social change: 1) the environment largely controls or sets limits on
the behavior that occurs within it; and 2) changing environmental variables results in
modification of behavior (Green et al. 1996; Green & Kreuter 1999). Therefore,
...health promotion can achieve its best results by exercising whatever control or
influence it can over the environment. But the reciprocal side of this equation also
holds that the behavior of individuals, groups, and organizations also influences their
environments (Green & Kreuter 1999:23). Social ecology suggests that program
developers and researchers should consider the possibility of changing both sides of
the equation.
Reciprocal determinism can be observed in studies of pedestrian accidents and safety,
as well. For example, Jacobsen reports that increasing the number of pedestrians and
bicyclists can actually reduce the rate of pedestrian and bicycle collisions with motor
vehicles by changing driver behavior and thus creating a safer environment for
pedestrians and bicyclists (2003). Although the rationale of increasing pedestrians
and bicyclists to reduce the rate of collisions with motor vehicles initially may be
counter-intuitive, it serves as an important reminder of the principle of reciprocal
determinism. When considering interventions to improve pedestrian safety, one
should not just consider how changing the environment may affect behavior but also
how changing behavior may affect the environment.
36


Strengths and Weaknesses of Social Ecology
Many of the strengths of social ecology have been mentioned above in the overall
description of the approach. Social ecology provides a framework that is flexible and
can be adapted for use with many issues in a variety of situations. As a multi-level
model, it forces researchers as well as program design and implementation personnel
using social ecology to consider the role various factors may be playing at different
levels even if all factors are not to be addressed. It encourages the acknowledgement
of the interaction of various levels of influence on human behavior. Finally, it
provides a useful organizational tool that can serve as a starting point for dialog
across disciplines in relation to a specific topic. It provides a heuristic device that
keeps one from focusing purely on the individual or on limited factors in isolation.
Many topics in health and safety are extremely complex and can benefit from the
skills and insight provided by a variety of disciplines. A social ecological framework
can serve to help integrate this knowledge and provide a more comprehensive
understanding of the topic.
Like any perspective, however, social ecology also has its weaknesses. Many
theories in health behavior and health promotion outline specific constructs and
variables that provide guidance to researchers and program developers. For example,
the Health Belief Model includes four major constructs (Strecher & Rosenstock
1997). The Theories of Reasoned Action and Planned Behavior emphasize the
constructs of intentions, attitudes, norms, and perceived behavioral control (Montano
et al. 1997). Banduras Social Cognitive Theory stresses the role of observational
learning and self-efficacy among other constructs (Baranowski et al. 1997). Having
such theoretical guidance about the importance and relationship of specific variables
and constructs can help focus research and interventions. Social ecology does not
provide this type of theoretical guidance, however. Not only does social ecology not
37


provide guidance about specific variables or constructs of importance, it also does not
provide any guidance on the mechanisms of effect or the mechanisms of interaction
between levels of influence. While flexibility may be strength of social ecology, at
the same time its lack of structure is one of its greatest weaknesses.
Another potential weakness of social ecology is that with all the focus on multiple
levels and a multitude of factors the individual at the center of the framework
becomes an afterthought. Individual people make up the populations that are at the
heart of public health. The disciplines goal is to create the safest, healthiest lives for
as many people as possible. Researchers and program developers must not lose sight
of that goal even as they attempt to be comprehensive in their approach and
understanding.
A final weakness, related to the issue of not losing sight of the individual, is that
social ecology may encourage too complex an approach. While it is good to think
about issues in a holistic manner, a single research project or program intervention
cannot address everything at once. Limits must be set about which piece will be the
focal point.
Social Ecology as a Framework for Pedestrian Accidents
Despite its weaknesses, social ecology provides a useful framework upon which to
organize and integrate findings about pedestrian accidents and safety. A social
ecological perspective encourages consideration of how a variety of factors at
different levels play a role in affecting pedestrian accidents and safety. During initial
design of this study, social ecology was useful in framing the problem because it
38


draws attention to a range of variables at multiple levels that may play a role in
pedestrian accidents and safety and a consideration of how these variables interact.
Another reason a social ecological perspective is well suited for framing research on
pedestrian accidents and safety is that the perspective is inherently interdisciplinary in
its approach. It is not just interdisciplinary, however; it is a framework that can serve
as common ground for many disciplines, including those not traditionally included
in health (Best et al. 2003:172). Much of the existing work regarding pedestrian
safety has been done in the urban planning and transportation disciplines (e.g., NCSA
2003; STPP 2002; Retting et al. 2003; Frank et al. 2003). These studies focus on the
role of the physical and built environments as they affect pedestrians. In the medical
literature, trauma research concentrates on the individual factors involved in
pedestrian accidents (e.g., Anderson et al. 1997; Tester et al. 2004; Hameed et al.
2004). Utilizing a social ecological framework enables one to integrate insights from
all these disciplines as they relate to pedestrian accidents and safety. It also affords
the opportunity to consider how issues of pedestrian accidents and safety fit into
broader models of physical activity, obesity, and unintentional injury prevention.
A third reason that social ecology provides a good framework for considering
pedestrian accidents and safety is that it highlights the need to intervene at multiple
levels in order to have the greatest impact (Emmons 2000; Green et al. 1996; Green &
Kreuter 1999). In order to intervene at multiple levels, one must have an
understanding of the influences at the various levels, how they interact, and which
ones may be more amenable to change. Therefore, research guided by a social
ecological perspective should seek to investigate variables at multiple levels.
Pedestrian accidents and safety not only involve factors at various levels, they also
need to be considered across various geographic units. Minkler and Wallerstein state
39


that an ecological theoretical perspective is ...particularly useful in the study of
autonomous geographic communities, focusing as it does on population
characteristics such as size, density, and heterogeneity; the physical environment; the
social organization or structure of the community; and the technological forces
affecting it (2002:33). These are the types of issues relevant to pedestrian accidents
and safety.
This study specifically concentrates on two levels within a social ecological
framework the social level and the built environment level and across two
geographic units of analysis the census block group and the census tract.
Unfortunately, information at the personal level characteristics of the driver and/ or
pedestrian involved in accidents was not available from the data sources available
for the study. Therefore, in this study, the personal level can only be acknowledged
through presentation of descriptive information about characteristics of the individual
pedestrian/ motor vehicle accidents in the source database. Likewise, many of the
potential explanatory variables of interest are not available at the point location. As a
result, the outcomes are analyzed mainly at the aggregated units of the census block
group and census tract for which the potential explanatory variables are available.
Figure 2.1 provides a representation of how a social ecological framework helps
organize the variables utilized in this study. The lines surrounding the built
environment and the social environment are dotted to indicate the fluid nature of
these levels. Population density, for example, is mainly a characteristic of the social
environment, but one could argue that it is a characteristic of the built environment as
well. The type of residential development and land use within a neighborhood often
determines the population density within the location. Therefore, some
characteristics might not fit cleanly and obviously into one level or another.
40


Although social ecology is a useful theoretical perspective from which to consider
pedestrian accidents and safety, some cautions and limitations should be noted. First,
tremendously important, the individual is still important. As Emmons notes, .. .in
our efforts to address the broader social context we should not totally abandon efforts
to intervene at the individual level... the most effective intervention strategies are
likely to incorporate both the individual whose health behavior is in question and the
larger community and governmental forces that influence the life of that individual
(2000: 249). The literature often discusses the need for increased education of
pedestrians and school children about pedestrian safety (e.g., Joly et al. 1991;
Staunton et al. 2003). While these efforts are important, they may achieve the
greatest success in reducing injuries and fatalities from pedestrian accidents if they
are combined with other efforts to improve the social and environmental
characteristics of pedestrian safety such as sidewalks, lighting, cross walks, off-street
play areas, law enforcement, policy changes, or public transportation.
while the focus on the broader social and environmental context of the health issue is
Figure 2.1: Social Ecological Framework for Pedestrian Safety
Built Environment
Variables: Street structure of the area, density of liquor
license establishments within the area, land use within the
area
Social Environment
Variables: area population density, area educational
attainment, area economic indicators, area ethnic
composition, area age distribution, area mode of
transportation to work
Variables: accident location, day of week
month of year, weather condition, road con
condition, street location, street class
Individual Accident Character
41


Second, even taking a holistic approach, it would be impossible to analyze and
understand every component at every level of an issue. Therefore, parameters have to
be defined. Setting such parameters, however, always leaves one open the criticism
... that [one has] not gone deeply enough to the root of the problem (Green &
Kreuter 1999:25). Yet, parameters have to be set about how much is manageable in
any given study. Perceptions of pedestrian safety and danger by neighborhood
residents and enforcement of pedestrian and motor vehicle laws in an area are two
factors that may play an important role in patterns of pedestrian/ motor vehicle
accidents within the City and County of Denver, but this study could not address
those issues.
Conclusion
While a social ecological framework was useful in framing the research questions and
organizing previous research, it is less useful in interpreting results and for
considering future research directions. Therefore, additional theories and theoretical
constructs will be considered in the Discussion Chapter to help interpret the findings
of this study. These additional theoretical constructs will help shed light on the
relationships between social and built environment factors and levels of pedestrian/
motor vehicle accidents as well as to suggest directions for future research that would
further enhance our understanding of pedestrian accidents and safety.
42


CHAPTER 3
METHODS
This study aimed to answer two major research questions. First, what are the spatial
patterns of pedestrian/ motor vehicle accidents within the City and County of Denver?
To answer this question, traffic accident report (TAR) data from the Department of
Public Works of the City and County of Denver were mapped and analyzed. Second,
what is the relationship of characteristics of the built environment and social
environment to pedestrian/ motor vehicle accidents at the census block group and
census tract levels within the City and County of Denver? To answer these research
questions, the TAR data were used to create the outcome variables describing
pedestrian/ motor vehicle accidents per block group and per tract. Land use data from
the GIS Office at the City and County of Denver, liquor license establishment data
from the Colorado Department of Revenue, and street structure data from the City
and County of Denver were used to create variables representative of the built
environment as explanatory variables. Additional explanatory variables were derived
from the 2000 U.S. Census data creating variables representative of the social
environment. Characteristics of the social environment investigated at the census
block group and census tract units included population density, population age
distribution, economic indicators, ethnic composition, education attainment, and main
mode of transportation to work.
43


Research Design
This exploratory, ecological study employed a retrospective, cross-sectional design
utilizing publicly available, secondary data to answer the research questions. A
similar design was used to explore the relationships between infrastructure neglect in
a neighborhood and gonorrhea rates (Cohen et al. 2000) and between alcohol
availability and gonorrhea rates (Scribner et al. 1998). While such designs do not
provide concrete evidence of causality, they do allow for exploration of potentially
new relationships and for hypothesis generation for future studies.
The study design integrated geographic and quantitative analysis of secondary data.
First, descriptive analyses were conducted about the characteristics of pedestrian/
motor vehicle accidents. These analyses identified, described and explored
characteristics of motor vehicle accidents involving pedestrians within the City and
County of Denver at the point location. Next, geographic analyses of the patterns of
pedestrian/ motor vehicle accident locations were conducted at both the census block
group and census tract geographic units. Spatial statistics testing for significant
global clustering of accidents were employed. Finally, statistical modeling of socio-
demographic and built environment characteristics associated with pedestrian
accident rates was conducted at the census block group and census tract geographic
units. These models were then verified using weighted regression models that
account for spatial relationships within the data.
The purpose of conducting both the epidemiological and geographic analyses was to
consider the possible role of spatial relationships within the data. Spatial dependence
is .. .the extent to which the value of an attribute in one location depends on the
values of the attribute in nearby locations (Fotheringham et al. 2002). Similarly,
spatial autocorrelation posits that ... observations are correlated strictly due to their
44


relative locational positions, resulting in spillover of information from on location to
another (Griffith & Layne 1999:3). Because events closer together in space are
more likely to be related than those farther apart therefore violating any assumption
of independence, potential spatial relationships in the data must be checked and
accounted for with statistical techniques. Yet, to date, most analyses in public health
considering geographic units have treated the geographic units as independent and
have mainly used linear models. Taking spatial autocorrelation within the data into
account should help reduce standard errors of measurement and provide more
reasonable estimates of spatially-based ecology (Raudenbush 2005).
This study utilized a Geographic Information System (GIS) to help analyze, integrate
and interpret the findings. GIS allowed for the visual display and analysis of data
within a spatial context. The integration of geographic and quantitative data through
GIS can provide comparative data over space and time as well as a more in-depth
understanding about the why of peoples behaviors and the ability to portray the
realities experienced by the subjects in a vivid way (Moran & Butler 2001:65).
Geographic data has been described as the hook around which various kinds of
information can be organized (Moran & Butler 2001:68). Furthermore, a GIS model
offers the opportunity to represent different perceptions of reality in an accessible
fashion (Moran & Butler 2001). GIS can help ease the visualization, analysis, and
integration of spatially based data (Ghose 2001; Ricketts 2003). In this study, a GIS
model facilitated the analysis, integration, interpretation, and presentation of the
epidemiological data through a common geographic link.
Such a model also facilitated a link back to the theoretical perspective. Since the
study involved analyses at multiple geographic units of analysis and across various
characteristics of the built and social environments, GIS afforded the opportunity to
visualize recurring patterns and compare similarities in spatial relationships.
45


Therefore, GIS provided an important tool for understanding and relating the data at
multiple scales.
Study Location
The City and County of Denver was selected as the focus of this study for several
important reasons. First, the county experienced the highest number of fatal
pedestrian accidents among Colorado counties in 2002 according to FARS (see Table
1.1 in Chapter 1). Margai (2001) used a similar rationale for selecting two counties in
the state of New York within which to conduct a geographic analysis of accidental
releases of hazardous materials. Second, the City and County of Denver is a single
administrative unit meaning that traffic and pedestrian laws as well accident-reporting
policies should be relatively uniform within the study location. It is also the central
core of the larger metropolitan area.
Within the City and County of Denver, pedestrian/ motor vehicle accidents were
considered at three geographic units the point location of the accident, the census
block group, and the census tract. The analysis of the accident point locations is
merely descriptive and is intended to provide a foundation for understanding the
distribution of accidents as the units of geographic aggregation increase. Census
block groups and census tracts were selected as the most appropriate units of
geographic aggregation for a number of reasons. First, they provide an opportunity to
analyze pedestrian/ motor vehicle accidents on a very local level. Second, block
groups are the smallest and tracts are the second smallest geographic units for which
complete data from the 2000 U.S. Census is publicly available. Third, if similar
patterns are observed at multiple geographic units of aggregation it reduces the risk
that the pattern is merely an arbitrary result of the aggregation and therefore
46


strengthens confidence in the results. As Waller and Gotway note, .statistically
significant clustering at the census tract level does not necessarily imply significant
clustering at the block group level, and vice versa (2004:201).
Study Period
This study encompassed a 4-year period from January 1, 2000 through December 31,
2003 inclusive. This period was selected because it extended from the 2000 U.S.
Census through the most recent year for which complete TAR data were available at
time the project began. Using such a multi-year period should also help ensure that
findings are not simply the result of a single fluke year in terms of pedestrian/
motor vehicle accidents.
Data Sources
Because no single data source was available to provide a comprehensive picture of
pedestrian accidents and safety in Denver, various data sources contributed to
answering the research questions. MacIntyre and colleagues expound the importance
of utilizing multiple data sources, especially those that go beyond just socio-
demographic data, when conducting place-based analyses (1993). Data were
obtained from the Department of Public Works of the City and County of Denver, the
GIS office of the City and County of Denver, the United States Census Bureau, and
the Colorado Department of Revenue. Each type of data and its source is described in
more detail in the sections below. Table 3.1 provides an overview of the data types
and their sources.
47


Table 3.1: Data Type by Source
Data Type___________Source________________________Year_______Variables Created
Traffic Accident Report (TAR) data Department of Public Works (DPW), City & County of Denver 2000-2003 Outcome variables
Socio-demographic data 2000 U.S. Census, United States Census Bureau 2000 Explanatory variables
Liquor license data Colorado Department of Revenue 2001 Explanatory variables
Land use data GIS Office, City & County of Denver 2005 Explanatory variables
Street classification data DPW, City & County of Denver 2005 Explanatory variables
Traffic Accident Report Data
The Traffic Accident Report (TAR) data were procured through the Department of
Public Works of the City and County of Denver. TAR forms are completed by law
enforcement officers responding to reports of motor vehicle accidents within the State
of Colorado. The TAR form contains over 100 pieces of information about the
accident itself, the vehicles involved, and the drivers of those vehicles. In cases of
motor vehicle accidents involving pedestrians, very little information is collected
about the pedestrian, however.
Various state and local agencies in Colorado receive a copy of each completed TAR
form for motor vehicle accidents that occur within their jurisdictions. Each agency
maintains its own TAR database for accidents within their jurisdiction, entering and
48


tracking the variables of interest to that agency. Most agencies do not track every
variable collected on the TAR form. The Department of Public Works of the City
and County of Denver was selected as the source of TAR data for this study due to
the completeness of the accident location information in its database.
The Department of Public Works TAR database was queried based on two criteria to
generate the pedestrian/ motor vehicle accident cases for this study. First, date of
event (accident) had to occur between January 1, 2000 and December 31, 2003
inclusive. Second, the first harmful event on the TAR form, defined as the event
when injury or damage first occurred (State of Colorado, Motor Vehicle Division
1997), had to list a collision with a pedestrian either as a child traveling to or from
school or as all other pedestrians. While this second selection criterion likely
excluded a few motor vehicle accidents involving pedestrians where the first harmful
event was not a collision with a pedestrian (such as accidents where two motor
vehicles collide and then one ricochets to hit a pedestrian), first harmful event is the
variable within DPWs TAR database that most reliably identifies motor vehicle
accidents involving pedestrians. Because DPW only receives and tracks completed
TAR forms for motor vehicle accidents that occur within the City and County of
Denver, it was not necessary to query the database by county or location of event.
The Department of Public Works of the City and County of Denver recorded 1,892
pedestrian/motor vehicle accidents in the county between January 1, 2000 and
December 31, 2003 inclusive. All analyses presented here excluded 27 of those
accidents that were listed as being in alley and 35 accidents that occurred at a
highway interchange. The in alley accidents were excluded because there was no
separate map available to locate those accidents in alleys instead of on public
roadways and because the location information in the dataset did not provide the
necessary detail to determine in which alley of the corresponding streets the accident
49


occurred. The highway interchange accidents were excluded because it is most
likely those pedestrians were walking on the highway interchange as a result of
automobile trouble or hitchhiking, not to get exercise or as transportation for routine,
daily activities. In addition, 19 accidents were excluded based on preliminary
mapping when these accidents mapped to jurisdictions outside the City and County of
Denver. Eight of these 19 mapped to Aurora, six to Glendale, two each to Adams
County and Jefferson County, and one to Arapahoe County. These above exclusions
left 1,811 pedestrian/motor vehicle accidents, representing 95.7% of the original
dataset, for inclusion in the exploratory analysis by point location. Additional
accidents had to be excluded for the geographical analyses at the census block group
and census tract units of analysis either because they could not be assigned to a tract
or block group or because the entire block group or tract had to be excluded from the
analyses.
Once the pedestrian/ motor vehicle cases were selected and exclusions made, the
remaining accident locations were assigned zip codes. Then, the accident location
information had to be formatted in order to increase the likelihood of successful geo-
coding. For example, accidents listed as occurring at Colfax and Colorado had to be
formatted to read E. Colfax Ave. and Colorado Blvd. The 1,811 pedestrian/ motor
vehicle accidents at the point location were geocoded in ArcGIS using the
TIGER/line file (Topographically Integrated Geographic Encoding and Referencing)
for the City and County of Denver from the 2000 U.S. Census (Map 3.1).
Pedestrian/ motor vehicle accident locations were then assigned to the appropriate
census block group and tract based on the block group or tract assignment provided
by the U.S. Census Bureau website for that location (USCB 2005). While many
accidents occurred on streets that serve as census block group and/or census tract
boundaries, the USCB assigns a single tract or block group to the location depending
50


on how the location is entered into its database. For example, an accident listed as
occurring at East Colfax Avenue and Colorado Boulevard may be assigned a different
census block group or tract than one listed at Colorado Boulevard and East Colfax
Avenue because these streets serve as census block group and tract boundaries.
The DPWs TAR dataset provides a primary street and closest cross street for every
accident. The primary street was listed first when entered in the USCB database
followed by the secondary or closest cross street. The corresponding census block
group or tract assignment provided by the U.S. Census Bureau was utilized. Map 3.2
shows the all census block groups within the City and County of Denver while Map
3.3 presents the all census tracts. As will be described in the section on the census
data, block groups 0041051 (Stapleton), 0041052 (Stapleton), and 0083131 (DIA) as
well as tracts 41.05 (Stapleton) and 83.13 (DIA) were excluded from the spatial
analyses, so these block groups and tracts appear in Map 3.2 and Map 3.3 but not in
future maps.
A separate variable was also created indicating if the accident fell on the boundary of
a census block group and/or tract. Over half of all accidents (52.0%) fell on both a
tract and block group border, while another 15.0% fell on just a block group border.
Since most pedestrian/ motor vehicle accidents occur along the same arterial street
that serve to sub-divide a city, any analysis of pedestrian/ motor vehicle accidents at
the local level has to decide how to handle this dilemma. Because the explanatory
variables at the census block group and census tract units were derived from 2000
U.S. Census data, it was deemed important to utilize the USCBs block group and
tract assignments of a location instead of an alternative method.
51


Map 3.1: Streets within the
City and County of Denver
7T
52


Map 3.2: Census Block Groups within
the City and County of Denver

I
m
_r



i


--

§
A
0 1 2
4
6
8


Map 3.3 Census Tracts within
the City and County of Denver
N
0 t.25 2.5
7.5
10
iMiles
54


After accidents were assigned to census block groups and census tracts, summary
statistics were calculated for each block group and tract describing the number of
accidents that occurred within that area. Variables were calculated in terms of counts
and rates the absolute number of accidents, accidents per population in the
geographic unit, accidents per roadway mile in the geographic unit, and accidents per
square foot in the geographic unit. Table 3.2 summarizes the pedestrian/ motor
vehicle accident variables created by block group and by tract. The main outcome
variable, accidents per 1,000,000 square feet in the geographic unit, was then selected
based on results of the exploratory analysis including frequency distributions,
geographic distributions, correlations, and scatterplots.
Table 3.2: Calculated Variables Based on Traffic Accident Report Data
Variable Description Type
Pedestrian Accidents Total number of pedestrian accidents in area Interval
Accidents per 1,000 Population (Number of accidents/ Total population)* 1,000 Continuous
Accidents per 10 Roadway Miles (Number of accidents/ Total roadway mileage)* 10 Continuous
Accidents per 1,000,000 Square Feet in Area (Number of accidents/ Total square feet)* 1,000,000 Continuous
From DPWs TAR data, variables were created describing characteristics of the
accidents. These variables included variables indicating the day of week, month, and
year of the accident based on the event date listed in the database for each case.
Variables in the dataset included road, weather and lighting conditions at the time of
the accidents, number of vehicles involved, and number of pedestrian injuries and
fatalities incurred.
55


Liquor License Establishment Data
Data on establishments within the City and County of Denver holding liquor licenses
as of January 2001 were obtained through the Colorado Department of Revenue.
Liquor license information is publicly available data. This dataset provided the type
of liquor license awarded as well as the address for each licensed establishment.
Liquor license establishments were considered an important factor of the built
environment to investigate because previous research indicates that the number and
type of establishments in an environment influence both the number of drinking
drivers as well as the number of drinking pedestrians (Gruenewald et al. 1999;
Scribner et al. 1994).
The liquor license database contained a total of 1,368 establishments. Of these, 89
licenses were excluded because the type of license, such as importer or wholesale
manufacturer, was not relevant to this study. Ten more licenses were excluded
because they were the license enforcement agency headquarters. An additional 24
licenses were excluded because they mapped to jurisdictions outside of the City and
County of Denver despite being listed as Denver County in the state database.
These exclusions left a total of 1,245 establishments, or 91.01% of the original
dataset, as part of the analysis. These 1,245 licensed establishments were then
categorized based on the type of license issued for the establishment.
After categorization, the addresses for the liquor license establishments were assigned
zip codes. As with the pedestrian/motor vehicle accident locations, addresses then
had to be standardized to the format used by the U.S. Postal Service in order increase
the likelihood of successful geocoding. For example, a liquor license establishment
listed at 1505 Colfax had to be re-formatted as 1505 E. Colfax Ave. These 1,245
56


liquor license establishments were then geocoded in ArcGIS using the TIGER/line
file for Denver County from the 2000 U.S. Census.
Next, establishments were aggregated to the census block group and tract units based
on assignment by the USCBs website (USCB 2005), just as the pedestrian/ motor
vehicle accident locations were. Because the liquor license establishments are
physically located on one side of the road or the other, they do not fall on boundaries
between census block groups and tracts. Once the census block group and tract
assignments were made, variables were calculated indicating the type of licenses in
the geographic unit as well as counts of total numbers and rates per population, per
roadway mile, and per square foot. Table 3.3 lists the variables created from the
liquor license outlet data.
All liquor license variables were included in the exploratory analysis involving
frequency distributions, geographic distributions, correlations, and scatterplots. Then,
based on the results of the exploratory analysis, certain liquor license variables were
selected for inclusion as explanatory variables in the multivariate analyses. Inclusion
criteria for liquor license variables as explanatory variables comprised of the variable
having 1) a relatively Gaussian (non-skewed) distribution or the ability to be
transformed to a Gaussian distribution, 2) a relatively high correlation with one or
more outcome variable, 3) a relatively low correlation with other potential
explanatory variables, and 4) linearity with the outcome variable.
57


Table 3.3: Calculated Variables Based on Liquor License Establishment Data
Variable Description
Total Liquor Hotel/Restaurant Liquor Store Bar Beer Proportion Hotel/Restaurant Proportion Liquor Store Proportion Bar Proportion Beer Liquor per Population Hotel/Restaurant per Population Liquor Store per Population Bar per Population Beer per Population Liquor per Area Hotel/Restaurant per Area Liquor Store per Area Bar per Area Beer per Area Liquor per Roadway Hotel/Restaurant per Roadway Liquor Store per Roadway Bar per Roadway Beer per Roadway Total number of liquor licenses (all types) in area Total number of hotel and restaurant licenses in area Total number of liquor store licenses in area Total number of bar licenses in area Total number of 3.2 beer only licenses in area Number of hotel & restaurant licenses/ Total licenses Number of liquor store licenses/ Total licenses Number of bar licenses/ Total licenses Number of 3.2 beer only licenses/ Total licenses Number of liquor licenses (all types)/ Total population Number of hotel & restaurant licenses/ Total population Number of liquor store licenses/ Total population Number of bar licenses/ Total population Number of 3.2 beer only licenses/ Total population Number of liquor licenses (all types)/ Total area Number of hotel and restaurant licenses/ Total area Number of liquor store licenses/ Total area Number of bar licenses/ Total area Number of 3.2 beer only licenses/ Total area Number of liquor licenses (all types)/ Total road mileage Number of hotel & restaurant licenses/ Total road mileage Number of liquor store licenses/ Total roadway mileage Number of bar licenses/ Total roadway mileage Number of 3.2 beer only licenses/ Total roadway mileage
Land Use Data
Current land use data for the City and County of Denver was obtained through the
GIS Office of the City and County of Denver with the assistance of the Department of
Public Works. This dataset contained the location and descriptive information for
over 166,000 land parcels within the City and County of Denver in a GIS-compatible
format. The land use categories provided in the database were extremely detailed
(e.g., apartment with 8 units). Descriptions of land parcels first had to be grouped
into a few broader categories such as residential use, business activity, industrial use,
58


or institutional use. Land parcels with more than one use were assigned to the
primary use based on the City and County of Denvers description of the parcel. For
example, retail with residential was classified as business activity while apartment
unit with commercial usage was categorized as residential.
The land use parcels file was overlaid onto a census block group and then a census
tract file in ArcGIS. An intersection of the two layers was performed. Then, based
on the intersection result, summary scores were calculated for each census block
group and tract indicating the proportion of land within the geographic unit dedicated
to residential use, business activity, industrial use, and institutional use. Table 3.4
provides a summary of the variables created using the land use data. These variables
are similar to some of the land use variables utilized by Handy and colleagues in their
study of the built environment and physical activity (2002).
Table 3.4: Land Use Variables
Variable Calculation
Proportion business activity Total land dedicated to business activity/ Total land
Proportion residential use Total land dedicated to residential use/ Total land
Proportion industrial use Total land dedicated to industrial use/ Total land
Proportion institutional use Total land dedicated to institutional use/ Total land
59


Street Structure Data
Because a single source was not available to provide street structure information at
both the point and the geographic aggregate units, street structure data were gathered
and classified using separate mechanisms for the point and the census tract units.
Street structure data was not available at the census block group level. Therefore, no
variables indicating street structure were included in the analyses conducted at the
census block group level.
For use at the point level, a hard-copy map of the all streets within the City and
County of Denver indicating their classification as arterial, collector, or local streets
was obtained through the Department of Public Works of the City and County of
Denver. Arterial streets are ... designed to provide a high degree of mobility and
generally serve longer vehicle trips to, from, and within urban areas (City & County
of Denver 2006). Arterials usually have a speed limit between 30-45 miles per hour
(MPH) and accommodate 10,000 to 75,000 vehicles per day (City & County of
Denver 2006). These are the most heavily traveled thoroughfares in the city.
Collector streets are ... designed to provide a greater balance between mobility and
land access within residential, commercial, and industrial areas (City & County of
Denver 2006). Collectors usually have a posted speed limit between 25-35 MPH and
carry between 5,000 and 20,000 vehicles a day (City & County of Denver 2006).
Local streets are ... tailored more to providing local access and community
livability (City & County of Denver 2006). Posted speed limits typically range
between 25-30 MPH and traffic volumes are generally less than 5,000 vehicles per
day (City & County of Denver 2006).
60


Based on this map, each pedestrian/ motor vehicle accident location was assigned to
its corresponding street classification. For accidents occurring at intersections with
streets carrying two differing classifications, priority was given to the more heavily
traveled of the street classifications. For example, at the intersection of Colorado
Boulevard and 7th Avenue, Colorado Boulevard is classified as an arterial street while
7th Ave is classified as a collector street. A pedestrian/ motor vehicle accident
occurring at this intersection therefore would be assigned an arterial street
classification since that is the classification of the more heavily traveled street
involved in the accident.
In order to provide an indication of the street structure within a census tract, total
miles of each type of street within the census tract were obtained from calculations
using data from the City and County of Denver. Then, the proportions of total street
miles that were arterial streets, collector streets, and local streets were calculated.
Table 3.5 provides a summary of street structure variables calculated at the census
tract unit.
Table 3.5: Street Structure Variables
Variable____________________________Calculation_______________________________
Proportion local streets Total local street miles/ Total street miles
Proportion collector streets Total collector street miles/ Total street miles
Proportion arterial streets Total arterial street miles/ Total street miles
61


Census Data
Socio-demographic variables of interest from the 2000 U.S. Census were downloaded
from the U.S. Census Bureaus website (http://factfinder.census.gov~) at the census
block group and tract geographic units. As block groups are the smallest unit for
which complete census data is publicly available and tracts are one unit of
aggregation above block groups but are still sub-divisions of a city, block groups and
tracts were chosen as the geographic units for the place-based analyses in this study.
The USCB provides tables with raw data by category. Proportions and rates per
population can then be calculated for variables based on these tables. Variable
categories (e.g., age groups) can be combined, as necessary. After the data tables
were downloaded, variables of interest were calculated from the absolute values.
Potential explanatory variables of interest for this study were selected based on
previous research. Table 3.6 lists variables of interest calculated from the 2000
Census data. Population density was selected because previous research has shown it
to be associated with levels of walking or pedestrian injuries at the individual level
(LaScala et al. 2000; LaScala et al. 2001; Craig et al. 2002; Frank & Pivo 1995).
Economic status and educational attainment variables were chosen because there
were utilized in a Canadian study of neighborhood levels factors associated with
physical activity (Craig et al. 2002). The age distribution variables were included
because several studies have shown that children and older adults are
disproportionately affected by pedestrian/ motor vehicle accidents relative to their
proportion of the population (LaScala et al. 2001; LaScala et al. 2000). The ethnic
composition variables were selected based on the various studies that have shown
certain ethnic groups bear a disproportionate amount of the burden of pedestrian
accidents in the United States (STPP 2002; Hanzlick et al. 1999; Marosi 1999;
62


Moreno & Sipress 1999). Transportation to work was selected because it is the only
variable collected by the USCB that provides an indication of potential pedestrian
exposure to motor vehicles. A review of the peer-reviewed literature did not reveal
any previous studies that have used this census variable as an explanatory variable in
studies of pedestrian accidents or safety.
Table 3.6: Variables Calculated from 2000 U.S. Census Data
Construct Variable
Population Density Total population per 1,000,000 square feet in area Total population per 10 roadway miles in area
Age Distribution Proportion of population under 18 Proportion of population over 64
Economic Status Median household income of area in 1999 U.S. dollars
Proportion of households living below poverty in 1999 Proportion of labor force unemployed
Ethnic Composition Proportion of population identifying as Hispanic Proportion of population identifying as non-Hispanic Other
(non-White)
Education Attainment Proportion of population 25 or older with a high school degree as highest level of education Proportion of population 25 or older with a college degree
or higher
Transportation to Work Proportion of labor force 16 or older who walk as main mode of transportation to work Proportion of labor force 16 or older who walk or take
public transport as main mode of transportation to work
As with the liquor license variables, all census variables listed in Table 3.6 were
included in exploratory analyses which involved descriptive statistics, frequency
distributions, geographic distributions, correlations, and scatterplots. Based on the
63


results of the exploratory analysis, certain variables were then selected for inclusion
as explanatory variables in the multivariate analyses. Similar to the inclusion criteria
for liquor license variables, multivariate analyses included socio-demographic
variables as explanatory variables if the variable had 1) a relatively Gaussian (non-
skewed) distribution or could be transformed to a Gaussian distribution, 2) a
relatively high correlation with one or more outcome variable, 3) a relatively low
correlation with other potential explanatory variables, and 4) good linearity with the
outcome variable.
Based on 2000 U.S. Census data, two census tracts within the City and County of
Denver and the three corresponding block groups had to be excluded from all
aggregate-level analyses. Census tract 83.13, the census tract incorporating Denver
International Airport, had a total population of only 4 in 2000. Tract 83.13
encompasses a single block group. Likewise, the population of census tract 41.05,
incorporating the Stapleton redevelopment neighborhood, was an entirely
institutionalized population in 2000, as this tract houses the jail for the City and
County of Denver and had no other residential population at the time of the census.
Tract 41.05 encompasses two block groups. These exclusions left 467 census block
groups and 134 census tracts for analysis.
Exploratory Data Analysis
Both non-spatial and spatial exploratory data analyses were conducted for all data.
Non-spatial exploratory data analysis involved running distribution characteristics
such as frequencies, ranges, means, medians, skewness values, standard deviations,
and quartiles. Then, correlations between and among explanatory and outcome
variables were run. Histograms and scatterplots were produced to visually represent
64


the distributions and relationships in the data. Non-spatial exploratory data analysis
was conducted in SPSS 13.0 (SPSS, Inc. Chicago).
Variables found to be positively skewed during exploratory data analysis were
transformed. Initially, numerous transformations appropriate for correcting positive
skewness were attempted. The transformation that best corrected the positive
skewness of each variable was selected. If any values for the variables requiring a
natural log transformation were zero, then a one was added to all values prior to
transformation. The natural log and square root transformations brought skewness
values for all positively skewed variables to within acceptable limits and converted all
the distributions to relatively Gaussian distributions. Table 3.7 illustrates the final
transformations applied to variables at the census block group and census tract units.
After transformation, exploratory data analysis was re-run on all transformed
variables.
Spatial exploratory data analysis involved creating point and choropleth maps for all
potential explanatory and outcomes variables. This was done to look for visual
clusters or patterns at the point, census block group, and census tract geographic units
and to compare visual patterns across levels. In addition, global Morans I analyses
were run using the potential outcome variables to check for statistically significant
spatial clustering. Spatial exploratory data analysis mapping characteristics and
calculating Global Morans I values was conducted using ArcGIS 9.1 (ESRI, 2004).
The most appropriate explanatory and outcome variables for inclusion in the
multivariate analysis were selected based on a review of the non-spatial and spatial
exploratory data analysis. All non-spatial and spatial exploratory analyses were run
at both the census block group and census tract units of analysis. Select exploratory
analyses were run by point location, as appropriate.
65


Table 3.7: Variable Transformations at the
Census Block Group and Census Tract Levels
Final Transformation Selected
Type_________Variable_________________Census Block Group Census Tract
Outcome
Total pedestrian/ motor vehicle accidents Natural log Natural log
Accidents per 1,000,000 square feet Natural log Natural log
Accidents per 10 roadway miles Natural log Natural log
Accidents per 1,000 people Explanatory Natural log Natural log
Population per 1,000,000 square feet Square root Natural log
Proportion Hispanic Natural log
Proportion of labor force unemployed Natural log
Proportion of labor force that walks or takes public transportation to work Square root Natural log
Liquor licenses per 1,000,000 square feet Natural log Natural log
Proportion of land Natural log
dedicated to business
activity
66


Multivariate Data Analysis
An iterative process was utilized to select variables for inclusion in the multivariate
analyses. Inclusion was determined based on the results of both the spatial and non-
spatial exploratory data analysis. At both the census block group and census tract
geographic units of analysis, an initial selected set of explanatory variables was run in
exploratory linear regression models with each of the four potential outcome
variables using the simultaneous regression method. Linear regression models were
run using SPSS 13.0 (SPSS, Inc. Chicago).
Assumptions for linear regression were checked for each regression model using the
studentized residuals for the models, where necessary. Assumptions of linear
regression include a linear relationship between each explanatory variable and the
outcome variable, normally distributed error that is uncorrelated with the explanatory
variables, and a lack of multicollinearity among explanatory variables (Leech et al.
2005). At both the census block group and census tract units of analysis, the selection
of explanatory variables was refined and a single outcome variable selected after
checking the model assumptions.
During multivariate analyses, global Morans I was also calculated in ArcGIS using
the studentized residuals from the exploratory linear regression models at both the
census block group and census tract units of analysis. This analysis checks for
statistically significant spatial clustering in the residual of the regression model.
Because the results of the global Morans I calculations using the studentized
residuals indicated potential for residual spatial autocorrelation in the regression
models at both the census block group and census tract units of analysis, additional
conditional autoregressive (CAR) models were run using the statistics package R
and its associated modules (Pebesma et al. 2006; R-core members 2005; Lewis-Koh
67


et al. 2005). These weighted regression models account for the spatial relationships
within the data and provide an estimate of any remaining residual spatial dependence
in the model. The CAR models utilize a geographic connectivity matrix. Such a
matrix uses binary indicator values (Os and Is) to note whether each census block
group or census tract neighbors each other block group or tract in the matrix (Griffith
& Layne 1999).
Human Subjects
This study involved only de-identified, publicly available, secondary data. Exempt
status was requested and granted through the Human Subjects Research Committee -
Institutional Review Board of the University of Colorado at Denver. Protocol #2005-
094 received approval as exempt research on February 23, 2005 and a one-year
extension on February 8, 2006. Copies of the approval letters from the Human
Subjects Research Committee can be found in the Appendix.
68


CHAPTER 4
POINT-LEVEL RESULTS
This chapter presents descriptive information about the characteristics of pedestrian/
motor vehicle accidents and liquor license establishments by point location.
Individual pedestrian/ motor vehicle accidents and liquor license establishments serve
as the units of analysis in this chapter. No multivariate analyses are presented at the
point level because no other data were available at the point level. The descriptive
information in this chapter is only intended to provide a foundation for understanding
the information as it is aggregated to the broader geographic units of census block
groups and then census tracts.
Pedestrian/motor vehicle accidents
The 1,811 pedestrian/ motor vehicle accidents that occurred on non-highway, public
roadways within the City and County of Denver from 2000 to 2003 were geo-coded.
The initial match rate for geo-coding through automatic matching was 1610, or 89%,
matched with a score of 80-100; 59 (3%) matched with a score of less than 80; and
142 (8%) remained unmatched. Interactive matching of the unmatched accident
locations increased the match rate to 1641 (91%) with a score of 80-100; 97 (5%)
with a score of less than 80; and 73 (4%) still remained unmatched. Map 4.1
illustrates the geo-coded locations of the pedestrian/motor vehicle accidents. Each
dot represents a single accident location; more than one accident may have occurred
at that particular location. Some locations, such as the intersection of Colorado
Boulevard and East Colfax Avenue experienced a multitude of accidents over the
four-year study period.
69


Map 4.1: Pedestrian/ Motor Vehicle
Accident Locations within the
City & County of Denver

N
A
0 1.25 2.5 5 7.5 10
1 m Miles
70


As Map 4.1 demonstrates, while the 1,811 pedestrian/motor vehicle accidents
occurred throughout Denver, certain intersections and areas of town experienced
more accidents than other locations. The map shows a solid concentration of
accidents in the central (in and around downtown and Lower Downtown) part of the
city as well as accidents all along the main arterial streets of the city Colorado
Boulevard, Colfax Avenue, Federal Boulevard, Broadway Street, and Alameda
Avenue. The greatest number of accidents at a single location over the four-year
study period occurred at the intersection of Colorado Boulevard and East Colfax
Avenue where 13 accidents happened. Eleven accidents occurred at the intersection
of South Federal Boulevard and West Alameda Avenue. Another ten accidents
occurred at 8400 Pena Boulevard, the site of Denver International Airport. Visually,
the map illustrates that pedestrian/ motor vehicle accidents are not distributed evenly
across town.
DPWs TAR database provides both a primary street and a secondary street, or
nearest cross-street, location for all accidents. Table 4.1 lists the primary streets that
experienced twenty or more pedestrian/motor vehicle accidents along their entire
length in the county as well as the total number of accidents along that street. The
table also presents the streets (along with total number of accidents) that were the
nearest cross-street to an accident that experienced twenty or more pedestrian/motor
vehicle accidents along their entire length in the county. As seen in Table 4.1., there
is a substantial drop-off between the number of accidents on the top two primary
streets, Colfax Avenue and Federal Boulevard, and the other primary streets with
large numbers of accidents. Likewise, Colfax Avenue experienced more than twice
the number of pedestrian/motor vehicle accidents than any other nearest cross-street.
All of the streets listed are major traffic thoroughfares that are classified by DPW as
arterial streets for most or all of their distance through the City and County of Denver.
71


Table 4.1: Streets with Twenty or More Pedestrian/Motor Vehicle Accidents
______Primary Street of Accident_______Secondary, or Cross-street, of Accident
Street Number Street Number
Colfax Avenue 144 Colfax Avenue 115
Federal Boulevard 130 Federal Boulevard 53
Broadway Street 79 Alameda Avenue 46
Colorado Boulevard 70 Evans Avenue 43
Alameda Avenue 49 Colorado Boulevard 28
16th Street 32 13 th Avenue 27
Sheridan Boulevard 32 Broadway Street 25
Monaco Parkway 32 38th Avenue 22
Evans Avenue 31
13 th Avenue 26
14th Avenue 23
Downing Street 20
Over the 4-year study period, an average of 1.27 pedestrian/ motor vehicle accidents
on non-highway, public roadways a day were reported to law enforcement in Denver.
In addition, some of these accidents involved more than one pedestrian. This statistic
indicates that being hit by a motor vehicle is a daily risk for pedestrians walking in
the City and County of Denver.
Pedestrian/motor vehicle accidents occurred across all years, months, days, and hours.
Table 4.2 illustrates the distribution of pedestrian/motor vehicle accidents by year,
month, season, day of week, and time of day. The lowest percentage of accidents,
23.5%, occurred in 2003 while the highest, or 27.3%, occurred in 2001. A chi-square
test indicates this is not a significant difference in pedestrian/ motor vehicle accidents
across years (x,2 = 6.43, df = 3, p = .09). Pedestrian/ motor vehicle accidents appear to
occur relatively consistently across years.
72


The distribution of accidents across seasons did show a significant difference,
however. The proportion of accidents ranged from 22.3% during the spring months
of March, April, and May to 28.2% during the fall months of September, October,
and November (x2 = 15.28, df = 3, pc.Ol). This difference by season might be a
result of low numbers in April as described in the paragraph below.
Across months, accidents ranged from a low of 107 (5.9%) in April to a high of 194
(10.7%) in December. This difference is also a significant difference (x2 = 42.33, df
= 11, p C.01). This significant difference in pedestrian accidents across months might
be driven by data reporting errors, however. Every month of the four-year study
period (48 months) saw at least 20 accidents, except for two months February 2000
and April 2003. Only 9 pedestrian/ motor vehicle accidents were recorded in
February 2000, seven of which happened during the first week of the month. April
2003 saw only 15 accidents for the entire month, and only 4 of these 15 accidents
occurred during the initial 15 days of the month. Given the average number of
accidents of more than one accident per day and the relatively consistent occurrence
of accidents during the other 46 months of the study, these few weeks with only a
very small number of accidents suggest that some accidents may have occurred
during this time but not been reported or recorded. If underreporting occurred in
April 2003, it would have affected the calculations about the seasons, as well.
Fridays, the day of the week with the highest number of accidents, saw more than
twice as many accidents as Sundays, the day with the lowest number of accidents. A
chi-square test indicates a significant difference across day of the week (x2= 61.25, df
= 6, pc.Ol). This difference in frequency of accidents on Friday versus Sunday is not
entirely unexpected since Friday likely combines both the pedestrian and motor
vehicle traffic of a weekday with that of a weekend while Sunday may be a less busy
day for both pedestrian and motor vehicle traffic.
73


Table 4.2: Description of When Pedestrian / Motor Vehicle Accidents Occur
___________VARIABLE_______________N (1811)_________%______________
Year 2000 2001 2002 2003 433 495 457 426 23.9 27.3 25.2 23.5
Month
January 151 8.3
February 126 7.0
March 135 7.5
April 107 5.9
May 161 8.9
June 146 8.1
July 130 7.2
August 150 8.3
September 175 9.7
October 167 9.2
November 169 9.3
December 194 10.7
Season
Spring 403 22.3
Summer 426 23.5
Fall 511 28.2
Winter 471 26.0
Day
Sunday 167 9.2
Monday 249 13.7
Tuesday 279 15.4
Wednesday 250 13.8
Thursday 285 15.7
Friday 336 18.6
Saturday 245 13.5
Time
Midnight-5:59am 193 10.7
6:00am-l 1:59am 360 19.9
Noon-5:59pm 786 43.4


Again, not entirely unexpectedly, the frequency of pedestrian/ motor vehicle
accidents varies significantly in terms of time of day in which they occur. By far the
greatest number of accidents occurred between noon and 5:59 p.m.; the fewest
number of accidents occurred between midnight and 5:59 a.m. (y2 = 414.13, df =3,
p<01). The four oclock p.m. (4:00-4:59 p.m.) and the five oclock p.m. (5:00-5:59
p.m.) hours saw the greatest proportion of accidents, each hour with 8.23% of all
accidents. On the other hand, the four oclock a.m. hour (4:00-4:59 a.m.) saw only
0.2% of all the accidents (data not shown). These differences by hour of the day are
significant as well (x = 572.87, df = 23, pc.Ol). This discrepancy in the time of day
of pedestrian/motor vehicle accidents mirrors what one might expect based on
assumptions of when more pedestrians would be walking and when more motor
vehicles would be on the roads.
Likewise, the conditions under which the pedestrian/motor vehicle accidents occurred
reflect what one might expect based on the general conditions of the Denver
metropolitan area. Table 4.3 shows the distribution of pedestrian/motor vehicle
accidents by weather, lighting, and road conditions, as well as by location on street,
street classification and number of vehicles involved. Over 90% of accidents for
which weather condition information was available occurred under no adverse
weather condition; nearly 90% of accidents occurred on dry roads. Correspondingly,
the weather in Denver is sunny and dry most days. Nearly two-thirds of accidents
occurred under daylight lighting conditions while approximately another one-third
happened under dark-lighted conditions. This substantial proportion of accidents
occurring under daylight lighting conditions seems to reflect the previously
mentioned fact that over 40% of all accidents happen between noon and 5:59 p.m.,
daylight hours most of the year.
75


Table 4.3: Conditions under which Pedestrian / Motor Vehicle Accidents VARIABLE N % Occur
Weather Condition No adverse (n=1805) 1669 92.5
Rain 80 4.4
Fog 3 0.2
Wind 5 0.3
Snow-Sleet-Hail 48 2.7
Lighting Condition Dawn or dusk (n=1806) 65 3.6
Dark, unlighted 51 2.8
Dark, lighted 554 30.7
Daylight 1136 62.9
Road Condition Non-dry (n=1794) 185 10.3
Dry 1609 89.7
Street Location At intersection (n=1766) 898 50.8
At driveway access 107 6.1
Intersection-related 98 5.5
Non-intersection (rural) 4 0.2
Mid-block (city) 659 37.3
Street Classification Local street (n=1781) 212 11.9
Collector street 195 10.9
Arterial street 1374 77.1
Vehicles Involved One vehicle involved (n=1799) 1618 89.9
More than one vehicle involved 181 10.1
76


Intersections are the street location that account for more than half of all pedestrian/
motor vehicle accident locations. Over 55% of accidents occurred at an intersection
or were intersection related. Another 37% occurred at a mid-city block. This
distribution of accidents mostly at intersections or mid-city blocks likely reflects the
grid-like pattern of most streets in Denver.
Of the 1,781 accidents for which a street classification was assignable, approximately
12% occurred on a local street or at the intersection of two local streets. Nearly 11%
occurred on a collector street, at the intersection of two collector streets, or at the
intersection of a collector and a local street. Just over 77% of accidents occurred on
an arterial street, at the intersection of two arterial streets, or at the intersection of an
arterial and a collector or an arterial and a local street. Approximately 90% of
accidents involved a single motor vehicle. Again, these frequencies of accident
conditions reflect what one might expect given that arterial streets are major
thoroughfares with higher volumes of traffic and often vehicles traveling at greater
speeds than on collector or local streets.
Because the TAR form only requests a small amount of information about the
individuals involved as pedestrians in traffic accidents and because DPW abstracts
and tracks only information from the TAR form most vital to its mission, only a very
limited amount of information is available about those pedestrians hit by motor
vehicles. Table 4.4 provides an overview of fatalities and injuries suffered by
pedestrians as well as the number of children going to or from school who are
pedestrians hit by motor vehicles. Although only 4.4% of the 1,793 accidents for
which information was available resulted in a pedestrian fatality, this represents an
average of nearly 20 pedestrians killed by motor vehicles per year in Denver.
77


This number of pedestrian fatalities does not include, however, those
pedestrian/motor vehicle accidents that occurred on highway interchanges as these
accidents where excluded prior to analysis. Previous research has shown that
pedestrian/motor vehicles accidents fatal to the pedestrian tend to involve accidents
with specific characteristics such as higher velocity and head-on nature (Anderson et
al. 1997) which might be present more frequently among accidents occurring on
highway interchanges. Therefore, the 79 pedestrian fatalities noted here most likely
under-represent the true number of pedestrian fatalities in Denver during the study
period.
In addition to the 4.4% of accidents that result in a pedestrian fatality, another 83%
result in injuries to one or more pedestrians. While there is no indication on the TAR
form of the type or severity of the injury sustained by the pedestrian, the physics of an
unprotected pedestrian being hit by a large, moving metal machine supports this high
percentage of accidents resulting in injury to the pedestrian. Nearly 9 out of 10
pedestrian/ motor vehicle accidents result in injury or fatality to the pedestrian.
Statistically, any pedestrian who is hit by a motor vehicle and not injured or killed is
very fortunate.
Ten percent of all pedestrian/motor vehicle accidents in Denver during the study
period involved students walking to or from school, or 182 students. This translates
to an average of 46 students a year hit by motor vehicles while walking to or from
school. The other pedestrian category includes all adults as well as children hit as
pedestrians at times other than when walking to or from school. This 10% of
pedestrian accidents involving students traveling to or from school is important as
public health activities increasingly promote children walking or biking to and from
school as part of the efforts to combat the rising rates of childhood obesity and
diabetes (U.S. DHHS 2001). This 10% in Denver also does not include kids riding
78


bicycles to or from school who were hit by motor vehicles, so the number of students
hit while traveling to or from school is likely even greater. National surveys of
parents indicate concern about traffic safety as a major obstacle to children biking or
walking to school (Dellinger & Staunton 2002). The 46 students per year on average
who are hit by motor vehicles going to or from school indicate that traffic safety may
be a real concern for parents in Denver as well.
Table 4.4: Pedestrians Involved in Pedestrian / Motor Vehicle Accidents
VARIABLE N %
Pedestrian Fatality No pedestrian fatality (n=1793) 1714 95.6
One pedestrian fatality 79 4.4
Pedestrian Injury No pedestrian injured (n=1794) 292 16.3
One or more pedestrians injured 1502 83.7
Pedestrian Type Student going to or from school (n= 1811) 182 10.0
Other pedestrian 1629 90.0
In general, the descriptive statistics presented here reflect what one might expect
about pedestrian/ motor vehicle accidents given the general conditions of the Denver
metropolitan area. The large percentage of pedestrians injured or killed in accidents
with motor vehicles, however, highlights the importance of pedestrian safety as a
public health issue. The percentage of pedestrians who are students walking to or
from school also speaks to the importance of pedestrian safety as a public health issue
as the nation promotes walking and bicycling to school as a means to combat the
rising childhood obesity and diabetes epidemics.
79


Liquor License Establishments
Just as the pedestrian/motor vehicle accident locations were, the liquor license
establishments were geo-coded to a street map of Denver. For the 1,245 liquor
license establishments, the initial match rate through automatic matching was 1048,
or 84%, matched with a score of 80-100; 7 (1%) matched with a score of less than 80;
and 190 (15%) were unmatched. Interactive matching of the unmatched liquor
license establishments improved the match rate to 1063 (85%) with a score of 80-100;
7 (1%) with a score of less than 80; meanwhile, 175 (14%) remained unmatched.
Map 4.2 presents the geocoded liquor license establishment locations. As with the
pedestrian/ motor vehicle accidents, a dot represents a single location even though
some locations such as the airport or a mall may have more than one liquor license
establishment at a single location.
In comparing the map of liquor license establishments to Map 4.1 of pedestrian/motor
vehicle accidents, one notes the visual similarity of distribution throughout Denver of
the two sets of locations. One noticeable difference is that nearly all liquor license
establishments are located on arterial and collector streets compared to pedestrian/
motor vehicle accidents. While most pedestrian/ motor vehicle accidents also occur
on arterial and collector streets, there are some accidents on local streets throughout
the county.
80


Map 4.2: Liquor License
Establishment Locations
within the City & County
of Denver
N
A
0 1.25 2.5 5 7.5 10
m Miles
81


As with pedestrian/ motor vehicle accidents, maps indicate that liquor license
establishments are not evenly dispersed throughout the city. Liquor license
establishments are heavily concentrated along arterial and collector streets as well as
in certain geographic areas of the city. Some geographic areas even appear to have
no liquor license establishments.
Using the codes assigned by the Colorado Department of Revenue indicating type of
liquor license, the 1,245 liquor license establishments were initially categorized into
four groups: hotel or restaurant, liquor store, bar, and 3.2 beer sold only. Just over
half (52.4%) of all liquor license establishments in the City and County of Denver are
hotel or restaurant establishments. Due to the relatively small number of
establishments in some of these categories when aggregated to the census block group
and census tract units, the categories were then combined into two larger groups
based on where the alcohol sold in the establishment is intended to be consumed. On-
premise establishments sell alcohol to be consumed by patrons on the premise of the
business, while off-premise establishments sell alcohol intended to be consumed
elsewhere. Nearly three-fourths of all liquor license establishments in Denver
(72.5%) sell alcohol for on-premise consumption by patrons. Table 4.5 illustrates the
categorization of liquor license establishments based on type of liquor license issued
as well as the total number of establishments in each category.
82


Code Table 4.5: Categorization of Liquor License Establishments Description Category Consumption N
1940 Retail liquor store license Liquor store Off-premise 169
1950 Liquor licensed drugstore Liquor store Off-premise 11
1960 Beer & wine license Hotel/ Restaurant On-premise 40
1970 Hotel & restaurant license Hotel/ Restaurant On-premise 585
1975 Brew-pub license Hotel/ Restaurant On-premise 11
1990 Club license Bar On-premise 32
2010 Tavern license Bar On-premise 219
2121 3.2 beer retail license on premises Hotel/ Restaurant On-premise 9
2122 3.2 beer retail license off premises 3.2 Beer Only Off-premise 162
2123 3.2 retail license on/off premises Hotel/ Restaurant On-premise 4
2340 Bed & breakfast Hotel/ Restaurant On-premise 3
TOTAL 1245
Liquor Store 180
Hotel/Restaurant 652
Bar 251
3.2 Beer Only 162
TOTAL 1245
On-premise 903
Off-premise 342
TOTAL 1245
Discussion
These analyses at the point level are purely descriptive and exploratory in nature.
The available data do not permit any further analysis by point location. Therefore, no
explanatory or causal conclusions can be drawn and no multivariate analyses were
conducted. Regardless, the descriptive analyses by point location reveal a number of
interesting observations about pedestrian/ motor vehicles accidents and the location of
liquor license establishments within the City and County of Denver.
83


The sheer number of pedestrian/ motor vehicle accidents that occurred during the
study period, more than an accident per day on average, indicates that pedestrian
safety should be a real concern for Denver, its residents, and anyone who must walk
in the city. The total number of accidents should also be of concern to the public
health community as it continues to promote walking for exercise and daily activities
in an effort to combat obesity and chronic disease and to improve health.
Furthermore, the relatively consistent nature of accidents over the four-year study
period indicates that pedestrian safety is an issue that needs constant attention. It
cannot be dismissed as one particularly bad year or due to another temporary
explanation. The results of the current analysis are similar to those of Braddock and
colleagues who found that child pedestrian accidents occurred relatively consistently
over their 3-year study period but were non-uniform in their geographic distribution
(1999).
The large number of accidents is especially concerning given the risk of injury or
death to any pedestrian hit by a motor vehicle. Over 87% of all accidents resulted in
injury or fatality to one or more pedestrians. Although the data did not provide
information on the proportion of accidents that resulted in injury or fatality to the
driver, the percentage is likely much lower as a metal-framed machine designed with
a number of safety features surrounds the driver while the pedestrian is essentially
naked. Pedestrians are at a distinct disadvantage during collisions with motor
vehicles. Therefore, it is important to prioritize safety for pedestrians. Just as the
public health community has done for years in advocating safety features such as seat
belts, airbags, and child-restraint seats for automobiles in an effort to reduce motor
vehicle injuries and fatalities, emphasis should now be placed on advocating for and
implementing structural changes, traffic enforcement, and education for pedestrians
and drivers to enhance pedestrian safety and decrease pedestrian injury and fatality
rates.
84


Ten percent of all pedestrian/ motor vehicle accidents in Denver involve a student
traveling to or from school. As a result, traffic safety may be a real concern for
parents in the Denver metropolitan area. The public health community must be
actively involved not only in promoting walking and biking as transportation to
school, but also in advocating for pedestrian safety to ensure that the environment is
as safe as possible for students traveling to and from school. Both students and
parents need to be educated about pedestrian safety. Increased traffic enforcement
around schools may be another way to create a safer environment for students. In
addition, public health professionals should work with Parent Teacher Associations
(PTAs) at schools to increase awareness of pedestrian safety and to advocate for
enhanced safety measures. For both the health and safety of students, it is important
to advocate for pedestrian safety.
Not surprisingly, most pedestrian/ motor vehicle accidents occurred along major
roadways, as illustrated both by Map 4.1 and by the fact that over 77% of accidents
occurred on streets designated as arterial streets by the City and County of Denver.
Of particular interest, two arterial streets Colfax Avenue and Federal Boulevard -
experience drastically more accidents than even other arterial streets in the city. This
observation suggests that concentrating safety improvements, traffic enforcement, and
educational efforts along these routes might result in the greatest impact for the
prevention efforts expended. Safety improvement and prevention efforts along the
Colfax Avenue corridor in particular may improve pedestrian safety, as Colfax
Avenue experienced the greatest number of accidents both as a primary street and as a
nearest cross-street.
The high percentage of pedestrian/ motor vehicle accidents occurring along arterials
streets also suggests a need to revisit the role of pedestrian activity in the design and
purpose of arterial streets. The Land Use Transportation Plan for the City and County
85


of Denver states that arterial streets are generally designed to accommodate
vehicle trips in the form of passenger cars, trucks, and buses ... [p]edestrian facilities
are always provided, but the width of these facilities varies depending on adjacent
land use and the level of pedestrian activity (City & County of Denver 2006).
Clearly, pedestrian activity along arterial streets in general and a couple of arterial
streets in particular is an important activity along the streets. While arterial streets
are designed for the mobility of vehicle traffic, the pedestrian use of these streets and
their safety should also be an important consideration.
Of additional interest are the ten accidents that occurred at 8400 Pena Boulevard, the
location of Denver International Airport. While pedestrians are not likely to be
walking around the airport for pleasurable exercise, it is a location with constant
exposure between pedestrians and motor vehicles. Increasing prevention and
awareness efforts of pedestrian safety around the airport may make the surrounding
area safer for everyone. Also, if they have not already done so, city officials and
others could bring the issue of pedestrian safety to the attention airport officials. The
airport is also a location where private funds might be available for infrastructure
improvements and educational efforts that would enhance pedestrian safety.
The timing of when pedestrian/ motor vehicle accidents occur reflects what one might
expect. The day of the week (Friday) and hours of the day (4:00-5:59 p.m.) when the
greatest percentage of pedestrian/ motor vehicle accidents occur are when one would
expect both a large number of cars being driven and pedestrians walking. Accidents
on Fridays might be a result of the motor vehicle and foot traffic from a weekday with
people going to and from work and school combined with weekend traffic with more
people venturing out for social events in the evening. The increased numbers of
accidents between the hours of 4:00-5:59 p.m. might reflect a combination of
increased car and foot traffic as a result of afternoon rush-hour as well as a popular
86


time to exercise. This combination may put more pedestrians on public roadways at
the same time as heavy motor vehicle traffic. Although these times probably reflect
hours of increased pedestrian/ motor vehicle exposure, it is interesting that it is not
the dark, nighttime hours when alcohol use among both the pedestrian and the driver
might be expected to be more prevalent combined with reduced visibility that
demonstrate the greatest percentage of accidents.
Similarly, the conditions under which accidents occur tend to reflect the general
conditions of the Denver metropolitan area. While one might expect increased
numbers of pedestrian/ motor vehicle accidents during adverse weather conditions, on
non-dry roads, or during non-daylight hours, this does not appear to be the case. Over
90% of all accidents occur under no adverse weather conditions and on dry roads.
Over 40% of accidents occur between 12:00-5:59 p.m., daylight hours nearly all year.
These observations about when and under what types of conditions pedestrian/ motor
vehicle accidents occur reiterate that pedestrian safety is a common, everyday issue.
It cannot be dismissed as something that happens mostly as a result of unfavorable or
infrequent conditions.
The similarity between the locations of pedestrian/motor vehicle accidents and liquor
establishments in Denver is striking when comparing Map 4.1 to Map 4.2. Arterial
streets appear to be key locations for both pedestrian/motor vehicle accidents and
liquor establishments. While nothing about the relationship of the two variables can
be said at the point level, both these variables are aggregated to broader geographic
areas in the next chapters and are included as part of the multivariate analyses by
census block group and census tract units of analysis.
87


CHAPTER 5
CENSUS BLOCK GROUP LEVEL RESULTS
This chapter presents exploratory and multivariate analyses of pedestrian/ motor
vehicle accidents within the City and County of Denver at the census block group
level. Census block groups serve as the unit of analysis in this chapter. The outcome
variables describing pedestrian/ motor vehicle accidents across census block groups
have been aggregated from the point location. The explanatory variables of the built
environment describing liquor license establishments within census block groups
have also been aggregated from the point location. The explanatory variables of the
social environment derived from census data are summary statistics describing the
composition of individuals residing within a census block group. On the other hand,
the explanatory variables of the built environment describing land use within block
groups provide an overview of the contextual characteristics of the census block
groups. No street structure data, another characteristic of the built environment, were
available at the census block group level. Therefore, those variables were not
calculated at the census block group level. In summary, the analyses at the census
block group level present a place-based analysis of a combination of contextual and
compositional characteristics of the built and social environments in relationship to
pedestrian/ motor vehicle accidents by census block group within the City and County
of Denver.
Exploratory Analysis: Pedestrian/Motor Vehicle Accidents
The City and County of Denver encompassed 470 census block groups at the time of
the 2000 U.S. Census. As discussed in the Methods chapter, three census block
88