Content uploaded by Chia-Yuan Yu
Author content
All content in this area was uploaded by Chia-Yuan Yu on Feb 05, 2018
Content may be subject to copyright.
https://doi.org/10.1177/0739456X17696035
Journal of Planning Education and Research
1 –15
© The Author(s) 2017
Reprints and permissions:
sagepub.com/journalsPermissions.nav
DOI: 10.1177/0739456X17696035
jpe.sagepub.com
Research-Based Article
Introduction
Traffic-related fatalities and injuries have become a major
concern in recent years. In the US, the estimated cost of
motor vehicle crashes was around $230 billion in 2010
(Teigen and Shinkle 2012). Also in 2010, the fatality rate per
100 million vehicle miles traveled (VMT) fell to an historic
low (1.10); only 32,885 people died in motor vehicle crashes,
the lowest number since 1949 (National Highway Traffic
Safety Administration 2012). However, there were still an
average of 90 persons killed by traffic crashes every day, and
one person died every 16 minutes (National Highway Traffic
Safety Administration 2012). Recently, researchers investi-
gating traffic safety have examined not only individual char-
acteristics (e.g., age, gender, and education level) and traffic
engineering measures (e.g., speed-reducing devices, and stop
signs), but also the spatial-scale attributes of built environ-
ments (e.g., street connectivity and land use patterns), in an
effort to develop regional crash models for crash prediction.
In the long term, these regional crash models could help to
better incorporate safety concerns into transportation plan-
ning (Washington et al. 2006).
Since crashes are often unevenly spatially distributed
(Loukaitou-Sideris, Liggett, and Sung 2007), the relation-
ship between built environments and crashes may vary by
each area’s particular characteristics. Most studies on traffic
safety have applied Generalized Linear Modeling (GLM) to
develop models for guiding traffic safety planning
(Hadayeghi, Shalaby, and Persaud 2010). However, this
approach uses fixed coefficients to represent the average
relationships between built environments and crashes, and
cannot investigate potentially variant associations across dif-
ferent areas. Examining the spatial variations of crashes and
their relationship to the built environment could provide
important information that would be useful in identifying
safety issues, especially in terms of the environmental
designs of specific areas. This information could then be
used to implement environmental interventions and improve
traffic safety. Furthermore, because collisions with different
levels of injury severity tend to be related to different charac-
teristics of the built environment (Clifton, Burnier, and Akar
2009; Ewing, Hamidi, and Grace 2016; Yu 2015a), it is
imperative to explore the various associations among the
types of built environments and collisions resulting in differ-
ent levels of injury severity.
This study explored built environment correlates with
collisions resulting in three different injury severities (fatal,
injury, and no injury) in Austin, Texas, by using local models
(Geographically Weighted Negative Binomial Regressions
[GWNBRs]). This research also compared the performances
of global models (negative binomial models) with those of
local models (GWNBRs) to examine whether the local mod-
els had better predictive power. This study contributes to the
existing body of the literature on the relationship between the
696035JPEXXX10.1177/0739456X17696035Journal of Planning Education and ResearchYu and Xu
research-article2017
Initial submission May 2015; Revisions February 2016, December 2016,
January 2017; Final decision February 2017
1School of Public Administration, University of Central Florida, Orlando,
FL, USA
2Department of Landscape Architecture and Urban Planning, Texas A&M
University, College Station, TX, USA
Corresponding Author:
Chia-Yuan Yu, School of Public Administration, University of Central
Florida, 4364 Scorpius Street, Orlando, FL 32816, USA.
Email: ychiayuan@gmail.com
Local Variations in the Impacts of
Built Environments on Traffic Safety
Chia-Yuan Yu1 and Minjie Xu2
Abstract
This study examined the influence of built environments on crashes with different levels of injury severity, and employed
Geographically Weighted Negative Binomial Regressions to test whether these relationships varied across different
neighborhoods in Austin, Texas. The results showed that high-speed roads produced more total and fatal crashes, and their
influence was stronger in downtown areas than in peripheral regions. Commercial and office areas experienced more injury
crashes, especially in downtown locations. It is crucial to implement programs to reduce vehicle travel demand, retrofit high-
speed roads, and design land areas that generate fewer vehicle trips, especially for downtown spaces.
Keywords
traffic safety, built environment, GWNBRs, local variations, neighborhood planning
2 Journal of Planning Education and Research
built environment and traffic safety by considering collisions
with different levels of injury severity, and adopting a local
approach to investigating nonstationary associations among
built environment factors and crash frequency. Moreover,
this research considered the influence of various dimensions
of built environments—density, diversity, design, and
regional accessibility—and explored the potential interac-
tion effects among land uses and road types on traffic safety.
The results provide direction for improving traffic safety in
different areas with a variety of characteristics, especially for
governments with limited financial and other resources.
Literature Review
Researchers have utilized several novel methodologies to
identify the factors related to traffic safety and, therefore, to
develop crash prediction models. The most common method
of calibrating this type of model is using GLMs such as nega-
tive binomial (NB) models (Hadayeghi et al. 2006; Noland
2003; Noland and Quddus 2004) and Poisson regressions
(Cheng, Geedipally, and Lord 2013; Ye et al. 2009). GLMs
examine the stationary associations among crashes and
related factors by estimating the global and fixed coefficients
within a given study area. A GLM’s stationary assumption
limits its ability to consider spatial influence, a factor that
affects the distribution of crashes (Brunsdon, Fotheringham,
and Charlton 1998). Another assumption of GLMs is that
each observation is independent, but this is violated because
of the spatial dependence (a correlation between what hap-
pens at one location and what happens in other places)
(Anselin 1988; Cliff and Ord 1973). Therefore, a GLM’s
main drawbacks are that it cannot consider probable spatial
nonstationary issues and the spatial dependence of the data.
Several studies have identified spatial dependence in traf-
fic safety (Aguero-Valverde and Jovanis 2008; Cottrill and
Thakuriah 2010; Flahaut 2004; Huang, Abdel-Aty, and
Darwiche 2010; Levine, Kim, and Nitz 1995; Siddiqui,
Abdel-Aty, and Choi 2012). For instance, Cottrill and
Thakuriah (2010) detected spatial clusters of pedestrian
crashes, showing that crashes tended to be located in certain
suburban locations in Chicago. A study in Florida found a
significant spatial dependence in crash occurrences across
adjacent counties (Huang, Abdel-Aty, and Darwiche 2010).
In response to this issue, researchers have used a methodol-
ogy called “spatial regression (spatial error and lag models)”
to address this spatial dependence issue (Aguero-Valverde
and Jovanis 2008; Eksler and Lassarre 2008; Flahaut 2004;
LaScala, Johnson, and Gruenewald 2001; Levine, Kim, and
Nitz 1995; Miaou 2003). For example, Aguero-Valverde and
Jovanis (2008) emphasized the importance of incorporating
spatial dependence into crash models and found improved
performance in models considering this aspect. LaScala,
Johnson, and Gruenewald (2001) used spatial error models
to explore factors associated with alcohol-related pedestrian
injury collisions in four California communities. The model
fit improved after accounting for spatial dependence. Flahaut
(2004) used spatial lag logistic models to examine the effects
of the road environment on road safety in Belgium. The
results showed a reduction in correlations of error terms after
accounting for spatial dependence. However, such spatial
models are typically considered semilocal, in that they only
address spatial dependence and cannot consider issues that
are spatially nonstationary. It is possible that some indepen-
dent variables may have strong predictive power regarding
crashes at certain locations, but may be weak predictors or
insignificant at other sites (Brunsdon, Fotheringham, and
Charlton 1998).
Local spatial models—geographically weighted regres-
sions (GWRs)—were used to consider various coefficients
for different subareas within the entire study location, in
order to calibrate the spatially nonstationary process.
Although GWRs have primarily been used in the health
(Gilbert and Chakraborty 2010; Nakaya et al. 2005) and eco-
nomic fields (Deller 2010; Harris et al. 2010; Zhang et al.
2011), some studies have employed this approach to explore
the traffic safety issue (Delmelle and Thill 2008; Erdogan
2009; Hadayeghi, Shalaby, and Persaud 2003, 2010;
Pirdavani et al. 2014; Shariat-Mohaymany et al. 2015). A
study in Iran employed GWRs with a Poisson distribution to
examine the influence of traffic volume, street intersection
densities, and street network density on the total number of
crashes in traffic analysis zones (Shariat-Mohaymany et al.
2015). Zhang et al. (2012) used GWRs with a Poisson distri-
bution to test the relationships among the number of crashes
involving pedestrians and bicyclists and several other factors
such as street intersection densities, commercial and residen-
tial areas, travel behaviors, and sociodemographic character-
istics in census tracts in Alameda County, California.
Hadayeghi, Shalaby, and Persaud (2003) explored local
associations among the number of deaths and sociodemo-
graphic characteristics in traffic analysis zones in Toronto,
Canada, by using GWRs with a Poisson distribution. Another
study also used GWRs with a Poisson distribution to exam-
ine the relationships among spatial factors and the number of
collisions in traffic analysis zones in Toronto, Canada
(Hadayeghi, Shalaby, and Persaud 2010). All results con-
firmed the existence of spatially nonstationary issues and
reported improved model performance by substituting local
models.
Although these studies demonstrated the effectiveness of
local models in the traffic safety field, they had certain limi-
tations. Most considered the total number of collisions within
the specified area and did not examine crashes with different
levels of injury severity (e.g., fatal, injury, and no injury).
Collisions with different levels of injury severity are likely to
be related to different built environmental factors. Dense
urban areas with short links to destinations and low vehicle
speeds have lower miles driven per capita, and are associated
with a decreased number of fatal crashes and increased vol-
ume of lower-severity collisions; conversely, sprawl areas
Yu and Xu 3
with low street connectivity and high vehicle speeds tend to
generate more fatal crashes (Clifton, Burnier, and Akar 2009;
Ewing, Hamidi, and Grace 2016). Therefore, it is necessary
to consider collisions with different levels of injury severity
(fatal, injury, and no injury).
Furthermore, these studies did not comprehensively con-
sider the influence of the built environment. Some only
included street intersection and road densities and ignored
other aspects such as transit service, road type, and certain
land areas (e.g., office, industrial, school, and park). To
address the identified research gaps, this study (1) examined
the effects of various attributes of the built environment on
crashes with different levels of injury severity and (2) tested
whether these influences varied for different levels of injury
severity and across neighborhoods.
Methods
Study Area and Boundary
The city of Austin was chosen as the study area, because of
the (1) variety of built environments, (2) diversity of sociode-
mographic characteristics, and (3) availability of compre-
hensive and updated data sets. The wide range of variation in
the study area offered advantageous conditions for examin-
ing the relationship between built environments and traffic
safety. For the study boundary, the city border was first con-
sidered. However, crash data were available only within the
jurisdiction of the Austin Police Department (APD), which
was smaller than the city boundary. Therefore, the APD’s
limit was selected in order to ensure data availability.
Units of Analysis
This research focused on the influence of neighborhood-
level built environments on traffic safety. Three types of
units have commonly been used in traffic safety studies,
including census block groups (Delmelle, Thill, and Ha
2012; Ha and Thill 2011; Ukkusuri et al. 2012; Wier et al.
2009), census tracts (Dumbaugh and Li 2011; Dumbaugh
and Rae 2009), and traffic analysis zones (de Guevara,
Washington, and Oh 2004; Hadayeghi, Shalaby, and Persaud
2010). However, there is no consistent definition of “neigh-
borhood scale.” In addition, results derived from only one
spatial unit may be subject to the Modifiable Areal Unit
Problem (MAUP); statistical estimations differ when the
boundaries of the zones used in the analysis change
(Openshaw 1984; Yu and Zhu 2016). Therefore, it is neces-
sary to consider the potential effect of the MAUP in this
study, since it may cause unreliable estimations and lead to
inaccurate implications.
A sensitivity analysis that explores the statistical estima-
tions and uses different spatial configurations would be a
feasible approach to accounting for the MAUP issue
(Fotheringham, Brunsdon, and Charlton 2000). Abdel-Aty
et al. (2013) investigated the influence of roadway and socio-
economic characteristics on total, severe, and pedestrian
crashes by using three different spatial units (e.g., census
block groups, census tracts, and traffic analysis zones).
Therefore, it was possible to explore the potential effects of
spatial configurations in this research through a sensitivity
analysis.
This study used census tracts, census block groups, and
traffic analysis zones to examine the relationships among
built environments and crashes with different levels of injury
severity. The signs and significances of the variables were
virtually the same among these three spatial units. This
research only reported the results from the census tracts.
Variables and Measurements
Dependent variables. The dependent variables in this study
were collisions with different levels of injury severity (fatal,
injury, and no injury). Because there were a limited number
of fatal crashes each year within each spatial unit, this
research aggregated three years (2010–2012) of crash data to
calculate the total number of all types of crashes for each
level of injury severity (fatal, injury, and no injury) as depen-
dent variables. A fatal crash was a collision with an injury
resulting in death. An injury crash indicated an accident pro-
ducing any injury except a fatality, and a no-injury crash was
defined as an accident with no injuries or only property dam-
age. Although different crash types (e.g., vehicle–vehicle,
pedestrian–vehicle, cyclist–vehicle, and pedestrian–cyclist)
may be related to different built environmental characteris-
tics (Clifton, Burnier, and Akar 2009; Dumbaugh and Rae
2009), this study focused on crashes with different levels of
injury severity. The associations among crash types and built
environments were beyond its scope.
Collision data were collected from the APD for three
years (2010–2012). These data provided the levels of injury
severity (fatal, injury, or no-injury) and geographic locations
of crashes (X and Y coordinates). Each collision location was
geocoded for spatial analysis in ArcGIS 10.0. As shown in
Table 1, a total of 117,828 crashes occurred between 2010
Table 1. The Numbers and Percentages of Crashes with
Different Levels of Injury Severity between 2010 and 2012 within
the Austin Police Department Boundary, Texas.a
Fatal CrashbInjury CrashcNo-Injury Crashd
2010 133 19,882 18,227
2011 146 19,804 17,315
2012 220 22,814 19,287
Total 499 62,500 54,829
Percentage (%) 0.43 53.04 46.53
aData source: 2010–2012 Austin Police Department (APD).
bFatal crash: an injury that results in death.
cInjury crash: any injury, other than a fatal injury.
dNo-injury crash: no injury—property damage only.
4 Journal of Planning Education and Research
and 2012 within the study area. In terms of injury severity,
0.43 percent were fatal crashes, 53.04 percent were injury
crashes, and 46.53 percent resulted in no injuries.
The crash count measure was used in this study because it
is an efficient approach to dealing with nonnormalized data
and has been applied extensively in traffic safety research
(Dumbaugh and Rae 2009; Marshall and Garrick 2011;
Ukkusuri et al. 2012). This study also tested the negative
binomial distribution of the crash count variables and showed
that all crash counts were overdispersed; if uncorrected, this
could distort the estimated standard errors and test statistics.
Consequently, a corresponding adjustment for the overdis-
persed data was considered in the model selection (Rabe-
Hesketh and Skrondal 2012).
Independent variables. This study systematically reviewed
the current research exploring the correlates of traffic safety,
in order to guide the researcher’s selection of the study vari-
ables. Several conceptual models were developed to discover
potential factors related to traffic safety (Ewing and
Dumbaugh 2009; Ukkusuri et al. 2012; Wier et al. 2009).
The framework for these models generally considered the
impacts of risk exposure, sociodemographic characteristics,
travel behavior, and built environments on crash frequency
and severity. Traffic volume and conflict were the two main
factors related to crash frequency, while traffic speed was the
primary determinant of crash severity. In terms of risk expo-
sure, higher traffic volume areas experienced more crashes
than locations with less traffic (Ewing and Dumbaugh 2009;
LaScala, Gerber, and Gruenewald 2000; Loukaitou-Sideris,
Liggett, and Sung 2007). This study accounted for vehicle
miles traveled in each spatial unit over a three-year period.
The average daily traffic (ADT) for each segment was
obtained from both the Texas Department of Transportation
(TxDOT) and the city of Austin.
The following briefly describes the procedures used in
calculating the vehicle miles traveled for each spatial unit.
First, the street layer was clipped by the boundary of the spa-
tial unit to obtain the street segments in each spatial unit.
Second, the ADT for each segment was multiplied by the
length of the street segment, in order to obtain the daily vehi-
cle miles traveled (VMT for each street segment within the
particular spatial unit). Third, the daily VMT was multiplied
by 365 days and three years, in order to be consistent with the
three-year crash data. Finally, the VMTs of all of the street
segments were summed and divided by the total number of
miles of street in the particular spatial unit.
For sociodemographic characteristics, this study included
the population under eighteen years of age and sixty-five
years and older. Numerous studies have found that accidents
involving people under age eighteen or over sixty-five were
more likely to result in severe injuries (de Guevara,
Washington, and Oh 2004; Sze and Wong 2007; Ukkusuri
et al. 2012; Zajac and Ivan 2003). For education level, loca-
tions with more people with high school educations (or
higher) experienced fewer collisions (LaScala, Gerber, and
Gruenewald 2000). This research also accounted for the por-
tion of the population who did not complete high school.
Because several studies have recognized that socioeconomi-
cally deprived areas (e.g., areas with lower incomes or con-
centrated minority populations) experienced more crashes
(Loukaitou-Sideris, Liggett and Sung 2007; Lu 2013;
Morency et al. 2012; Noland, Klein, and Tulach 2013;
Sengoelge et al. 2013), this analysis also considered the non-
white population and populations living below the poverty
line. In the category of travel behavior, the number of people
walking, biking, and taking public transit were important
measurements with regard to possible exposure to traffic
conflict. Because the data for all travel purposes were not
available, this study used proxy measures available from the
2010 American Community Survey, including the percent-
ages of workers commuting to work by walking, biking, and
using public transit.
In terms of the built environment, this study used the 3Ds
(density, design, and diversity) to select the relevant vari-
ables (Cervero and Kockelman 1997). For density, areas with
larger populations were associated with an increased number
of collisions (Clifton and Kreamer-Fults 2007; de Guevara,
Washington, and Oh 2004; Graham and Glaister 2003; Wier
et al. 2009). This study also considered the population den-
sity for each spatial unit in the statistical analysis.
Furthermore, different intersection types (e.g., three-leg and
four-or-more-leg intersections) had different effects on crash
frequency, since four-or-more-leg intersections produce
more conflict-prone traffic movement; more conflict is likely
to lead to more crashes (Dumbaugh and Rae 2009; Ukkusuri
et al. 2012). Thus, three- and four-or-more-leg intersection
densities needed to be considered separately. Regarding the
impact of transit services, some studies have reported that a
greater number of transit stops are associated with a reduced
number of collisions (Clifton and Kreamer-Fults 2007; Yu
2015b). This study considered the transit stop density of each
spatial unit to be a representation of the availability of transit
service. Different road types with different functions and
designs (e.g., road, shoulder, and median widths) have been
found to have different effects on crashes (Dumbaugh and Li
2011; Ukkusuri et al. 2012; Wier et al. 2009). The percent-
ages of highways/freeways, arterial roads, and local streets in
each spatial unit were used. Moreover, higher posted speed
limits tend to increase the probability of serious injury
(Desapriya et al. 2011; Eluru, Bhat, and Hensher 2008).
Generally, the likelihood of a pedestrian being killed by a
vehicle traveling 40 MPH is 85 percent (Zegeer et al. 2002).
As a result, this study considered the percentage of roads
with 40 MPH or higher speed limits in each spatial unit.
In terms of design, nonmotorized infrastructure may pro-
vide a separation space for pedestrians/cyclists and vehicles,
which could improve safety (Yu 2015b). Thus, this study
included the level of sidewalk and bike lane completeness of
each spatial unit. For diversity, traffic-generating areas (e.g.,
Yu and Xu 5
commercial, office, industrial, school, and park areas) were
identified as trip attractors and considered to have an influ-
ence on traffic safety (Kim, Brunner, and Yamashita 2006;
LaScala, Gruenewald, and Johnson 2004; Ukkusuri et al.
2012). The percentages of commercial, office, industrial,
school, and park areas in each spatial unit were included in
the analysis. Moreover, this study also considered the effect
of land use mix. The corresponding measurement referred to
the Strategies for Metropolitan Atlanta’s Regional
Transportation and Air Quality Study (Frank et al. 2005); it
ranged from 0 to 1. A higher value meant an even distribution
of residential, commercial, and office land uses throughout
the spatial unit.
Regional accessibility was identified as how the regional-
scale built environment related to travel demand (Ewing and
Cervero 2010; Handy 1993). Distance to the CBD (central
business district) was negatively associated with vehicle
travel per person (Zegras 2007). This study measured the dis-
tance between the centroids of the spatial unit and the area’s
CBD (considered to be the spatial unit housing the City
Hall). To consider the potential moderating effects of com-
mercial and residential areas on crashes, four interaction
terms between the percentages of commercial/residential
areas and arterials/local roads were generated. Table 2 pres-
ents the selected dependent and independent variables, their
measurements, data sources, time periods, units of measure-
ment, and descriptive statistics.
Data analysis. All point, line, and parcel data were aggre-
gated into the corresponding spatial units. Both global and
local models were used to compare their performances in
various built environment–traffic safety relationships. Nega-
tive binomial (NB) regressions were selected for the global
models, due to the overdispersed crash count data (Long and
Freese 2006). All were created using Stata 12.0.
In terms of local models, this study used GWNBRs to test
spatial variations in the associations among crashes and
related factors. GWNBRs consider spatial heterogeneity by
multiplying the coordinates of each regression point with
each independent variable (Fotheringham, Brunsdon, and
Charlton 2002). More specifically, the relationships among
the dependent variables and each independent variable were
calculated for each spatial unit across the study area.
GWNBRs estimate spatially varying relationships by obtain-
ing a variety of local estimates over the entire geographic
space. Thus, the formula was as follows (da Silva and
Rodrigues 2014):
yNBt x
jj k
k
jj jk jj
exp,
,,
βµ αµ
∑
()
()
νν
(1)
where tj is an offset variable; µ
jj
,
ν
()
denotes the coordi-
nates of the j point in space; βk is the parameter related to
the independent variable xk; βµ
kjj
,
ν
()
is a function
indicating the coordinates of the j point, which allows the
measure to be a continuous surface and accounts for the spa-
tial variability of the surface; and
α
is the parameter of over-
dispersion. Using this approach, this study was able to obtain
parameter estimates, standard errors, and certain diagnostic
statistics for every regression point.
In terms of the kernel type, the bandwidth was expressed
as the number of data points to be considered in the kernel
(Fotheringham, Brunsdon, and Charlton 2002). Because the
size of the spatial unit was irregular, “Adaptive” was used to
specify the bandwidth; this allowed for the observation of the
same number of data points in the local sample, which made
comparable the standard errors from each model. Also, the
bandwidth was determined by AIC minimization, which
considered the different number of degrees of freedom in the
different models in order to make more accurate compari-
sons of model performance (Fotheringham, Brunsdon, and
Charlton 2002). The local regressions were estimated using
SAS 9.2.
To compare the performances of the global and local
models, this study used several indicators (e.g., corrected
AIC, Mean Absolute Deviance, and Mean Square Prediction
Error) (Xu et al. 2014). The corrected AIC (AICC) was used
to consider bias generated by the small sample size (Hurvich
and Tsai 1989). In general, models with lower AICC values
tend to be better fits. Mean absolute deviance (MAD) and
mean squared predictive error (MSPE) were also used to test
the model fit (Xu et al. 2014). The MAD and MSPE values
were calculated as follows:
MADn
yy
i
pred
i
obs
i
n
=−
=
∑
1
1
(2)
MSPE n
yy
i
pred
i
obs
i
n
=−
()
=
∑
12
1
(3)
where yi
pred is the expected number of crashes in each spa-
tial unit, and yi
obs is the observed number of crashes in each
spatial unit. Models with lower MAD and MSPE values rep-
resented a better level of fit.
Correlation tests and a variance inflation factor (VIF)
were also used to detect any multicollinearity issues. Because
this study excluded independent variables with a VIF greater
than 4 (Belsley, Kuh, and Welsch 1980), streets with speed
limits of 40 MPH or higher were not included in the model.
Also, this study applied a “mean center” approach to address
potential multicollinearity issues with the interaction terms.
Results
Global Models
Table 3 presents the results from the negative binomial mod-
els. Areas with higher traffic volumes were associated with
more total, fatal, injury, and no-injury collisions. Areas with
6
Table 2. Study Variables and Their Measurements, Data Sources, Time Periods, and Units of Measurement.
Variable
Raw Data Descriptive Statistics
Variable Measurement in This Study Data Source Time of Data
Spatial Unit of
Measurement Mean (SD)
Dependent variables
Total crash Number of total collisions (2010–2012) Austin Police
Department
2010–2012 Point 1017.74 (1052.26)
Fatal crash Number of collisions with fatality (2010–2012) 3.93 (5.53)
Injury crash Number of collisions with injury (2010–2012) 565.04 (572.91)
No-injury crash Number of collisions with no injury (2010–
2012)
255.78 (305.88)
Independent variables
Risk exposure
Traffic volume Vehicle miles traveled during three years / total
miles of streets in the spatial unit
TxDOT,
City of Austin
2010 Line 387.65 (276.47)
Sociodemographic characteristics
Percentage of population aged under 18 Population aged under 18 / total population US Census
Bureau
2010 Spatial unit (census tract,
census block group, and
traffic analysis zone)
0.19 (0.15)
Percentage of population aged 65 and older Population aged 65 and older / total population 0.09 (0.06)
Percentage of nonwhite population Nonwhite population / total population 0.53 (0.24)
Percentage of population less than high school Population with the education level less than
high school / total population
0.12 (0.04)
Percentage of population below the poverty line Population below the poverty line / total
population
0.18 (0.06)
Travel behaviors
Percentage of workers commuting by walking Number of workers commuting to work by
walking / number of workers
American
Community
Survey
2010 Spatial unit (census tract,
census block group, and
traffic analysis zone)
0.05 (0.07)
Percentage of workers commuting by public transit Number of workers commuting to work by
public transit / number of workers
0.03 (0.04)
Percentage of workers commuting by biking Number of workers commuting to work by
biking / number of workers
0.01 (0.02)
Built environments
Density
Population density Total population / area of the spatial unit
(acres)
US Census
Bureau
2010 Spatial unit (census tract,
census block group, and
traffic analysis zone)
6.52 (3.08)
Street connectivity
Three-leg intersection density Number of three-leg intersections / area of the
spatial unit (acres)
City of Austin 2011 Point 0.16 (0.10)
Four-or-more-leg intersection density Number of four-or-more-leg intersections /
area of the spatial unit (acres)
0.20 (0.13)
Transit service
Transit stop density Number of transit stops / area of the spatial
unit (acres)
Capital Metro
– Austin Public
Transit
2010 Point 0.04 (0.04)
(continued)
7
Variable
Raw Data Descriptive Statistics
Variable Measurement in This Study Data Source Time of Data
Spatial Unit of
Measurement Mean (SD)
Road type
Highway/freeway The miles of highway/freeway / total miles of
streets in the spatial unit
City of Austin 2011 Line 0.04 (0.08)
Arterial road The miles of arterial roads / total miles of
streets in the spatial unit
0.12 (0.10)
Local road The miles of local roads / total miles of streets
in the spatial unit
0.42 (0.28)
Street with 40 MPH or higher The miles of streets with 40 miles per hour or
higher / total miles of streets in the spatial
unit
0.21 (0.15)
Design
Nonmotorized infrastructure
Sidewalk completeness (Sidewalk length) / (street length × 2) in the
spatial unit
City of Austin 2010 Line 0.63 (0.16)
Bike lane completeness (Bike lane length) / (street length × 2) in the
spatial unit
0.21 (0.16)
Diversity
Land use type
Commercial area Commercial area / area of the census tract
(acres)
City of Austin 2010 Parcel 0.07 (0.08)
Office area Office area / area of the spatial unit (acres) 0.04 (0.08)
Industrial area Industrial area / area of the spatial unit (acres) 0.04 (0.08)
School School area / area of the spatial unit (acres) 0.06 (0.07)
Park Park area / area of the spatial unit (acres) 0.12 (0.14)
Land use mix The evenness of residential, commercial, and
office areasa
0.17 (0.09)
Regional accessibility
Distance to CBD Distance (miles) between the centroid of the
spatial unit and the centroid of area’s CBD
(central business district)
US Census
Bureau
2010 Spatial unit (census tract,
census block group, and
traffic analysis zone)
10.85 (9.52)
Interaction between land areas and road type
Commercial uses × Arterial roads Percentage of commercial areas × percentage
of arterial roads
City of Austin 2010 − 0.01 (0.01)
Commercial uses × Local roads Percentage of commercial areas × percentage
of local roads
0.01 (0.01)
Residential uses × Arterial roads Percentage of residential areas × percentage of
arterial roads
0.02 (0.01)
Residential uses × Local roads Percentage of residential areas × percentage of
local roads
0.01 (0.01)
Note: SD = standard deviation; TxDOT = Texas Department of Transportation.
a(−1) × [(area of R / total area of R, C, and O) × ln(area of R / total area of R, C, and O) + (area of C / total area of R, C, and O) × ln(area of C / total area of R, C, and O) + (area of O / total area of R, C, and O) × ln(area of O /
total area of R, C, and O)] / ln(number of land uses present), where R = residential use; C = commercial use; O = office use.
Table 2. (continued)
8 Journal of Planning Education and Research
more of the population living below the poverty line had
more total and injury collisions. In terms of the built environ-
ment, areas with more highways/freeways and arterial roads
were associated with more total and fatal collisions, while
local roads were related to fewer fatal and injury crashes.
Higher percentages of commercial areas were related to
more total and injury crashes. Office areas were also associ-
ated with more injury crashes. Furthermore, the interaction
terms between commercial uses and arterial roads were asso-
ciated with more total, fatal, and injury crashes. In terms of
regional accessibility, areas far away from the CBD were
associated with fewer total, injury, and no-injury crashes.
Local Models
The GWNBR models generated local coefficients, t values,
and p values among all of the independent variables and each
dependent variable for each spatial unit. Figures 1 through 5
showed the coefficients and p values of the significant inde-
pendent and dependent variables for each census tract. The
percentages of highways/freeways and arterial roads were
significantly related to more total and fatal crashes in all cen-
sus tracts in the study area (Figures 1 and 2). Downtown
areas typically had larger coefficients between highways/
freeways and total and fatal crashes, and between arterial
roads and total and fatal crashes, than did peripheral areas.
The model indicated that a one percentage increase in high-
ways/freeways and arterial roads led to a more substantial
increase in total and fatal crashes in downtown areas than in
peripheral areas.
Local roads were associated with fewer fatal and injury
crashes in all census tracts in the study area (Figure 3). A one
percentage increase in local roads in downtown areas yielded
a more substantial decrease in fatal and injury crashes than
the same decrease in peripheral areas.
Census tracts with more commercial areas had signifi-
cantly more total and injury crashes (Figure 4). In terms of
the spatial variation of coefficients, a one percentage increase
in commercial areas was associated with more total and
injury crashes in downtown areas than in peripheral areas.
Also, commercial areas along arterial roads were associated
with more total, fatal, and injury crashes. However, this
Table 3. Global Model Results: Negative Binomial Model Predicting Crashes with Different Levels of Injury Severity (Unit of Analysis:
Census Tract).
Variables Total Crash Fatal Crash Injury Crash No-Injury Crash
Coefficient (p value)
Traffic volume 0.1135***
(<0.001)
0.1148*
(0.015)
0.1002***
(<0.001)
0.1058***
(<0.001)
Percentage of population below the poverty line 0.038*
(0.014)
0.035*
(0.031)
Highway/freeway 0.2317*
(0.012)
0.2812*
(0.018)
Arterial road 0.1811*
(0.028)
0.2108**
(0.009)
Local road −0.2433*
(0.013)
−0.1865*
(0.022)
Commercial area 0.2498***
(<0.001)
0.2685***
(<0.001)
Office area 0.2112*
(0.021)
Distance to CBD −0.0342**
(0.008)
−0.0126*
(0.012)
−0.0112*
(0.024)
Commercial uses × arterial roads 0.0160*
(0.014)
0.0187**
(0.008)
0.0204**
(0.007)
Global model (negative binomial model)
Corrected AIC 6305.01 1785.22 5528.16 5385.96
MAD 13.11 6.25 10.36 9.54
MSPE 34.35 15.96 27.53 26.74
Local model (GWNBR)
Corrected AIC 6041.26 1652.06 5312.08 5274.31
MAD 10.25 5.58 9.24 8.12
MSPE 30.25 13.86 23.99 21.58
N 144 144 144 144
Note: AIC = Akaike information criterion; MAD = mean absolute deviance; MSPE = mean squared predictive error.
*p < 0.05, **p < 0.01, ***p < 0.001.
Yu and Xu 9
effect did not show the general pattern between downtown
and peripheral areas.
Office areas were only significantly related to increased
injury crashes in the local model (Figure 5). The same pat-
tern was apparent; a one percentage increase in office areas
was associated with more injury crashes in downtown loca-
tions than in peripheral spaces.
To compare the performances of the global and local
models, AICC, MAD, and MSPE values were calculated
(Table 3). The GWNBRs outperformed the NB models with
Figure 1. Coefficients and p values between the percentage of highways/freeways and total and fatal crashes in census tracts.
Figure 2. Coefficients and p values between the percentage of arterial roads and total and fatal crashes in census tracts.
Figure 3. Coefficients and p values between the percentage of local roads and fatal and injury crashes in census tracts.
10 Journal of Planning Education and Research
lower AICC, MAD, and MSPE values when the two models
had the same independent variables.
Discussion
Areas with high traffic volumes consistently had more total,
fatal, injury, and no-injury crashes. This finding was con-
firmed by the results of previous studies (LaScala,
Gruenewald, and Johnson 2004; Loukaitou-Sideris, Liggett,
and Sung 2007; Miranda-Moreno, Morency, and El-Geneidy
2011; Wier et al. 2009), indicating that the greater the amount
of traffic, the higher the crash risk. Moreover, areas with
larger nonwhite populations and populations below the pov-
erty line had more injury crashes. This potential disparity
issue was also found in previous studies (Graham and
Glaister 2003; Loukaitou-Sideris, Liggett, and Sung 2007;
Noland, Klein, and Tulach 2013; Yu 2014). The various local
coefficients for each spatial unit in the local models implied
that the degree of impact of the built environment on traffic
safety could differ for each spatial unit in the study area. The
associations were generally strong (with large absolute coef-
ficient values) in downtown areas and weak (with small
absolute coefficient values) in peripheral areas.
High-speed and high-volume facilities (e.g., highways/
freeways and arterial roads) were associated with more total
and fatal crashes; the influence was stronger in downtown
areas than in peripheral areas. Although these roads were
designed for high travel speeds with wide and straight lanes,
this high-speed design can lead to traffic safety issues, espe-
cially in downtown areas. Since such locations often feature
a mix of traffic types, high-speed designs can cause traffic
conflicts among vehicles and between vehicles and nonmo-
torized users (often with a high level of injury severity). It is
also possible that fast-moving vehicles on highways will
need to decelerate before they enter other types of roads
(e.g., local outlets). This speed difference can cause traffic
conflicts between vehicles traveling on highways and those
on adjacent low-speed roads, leading to more crashes. In
terms of local variation, this study examined whether high-
ways were more likely to be designed to connect with low-
speed roads (i.e., local roads) in downtown areas with larger
local coefficients than in peripheral areas with smaller local
coefficients. The authors categorized census tracts into sub-
groups based on the median values of local coefficients
between highways/freeways and total crashes and highways/
freeways and fatal crashes. The results showed that the per-
centages of highways/freeways that connected to local roads
were all significantly higher in areas with larger local coef-
ficients than in areas with smaller local coefficients. This
indicated that areas with larger local coefficients may have
more traffic conflicts between vehicles on highways than on
low-speed roads. Furthermore, local roads were related to
fewer fatal and injury crashes. Narrow local roads are usu-
ally designed with low traffic speeds and reduced stopping-
sight distances, providing drivers with more time to respond
to unexpected road hazards. The benefits of these safety
measures were more obvious in downtown areas than in
peripheral locations.
Figure 4. Coefficients and p values between the percentage of commercial areas and total and injury crashes in census tracts.
Figure 5. Coefficients and p values between the percentage of
office areas and injury crashes in census tracts.
Yu and Xu 11
Areas with a higher percentage of commercial areas expe-
rienced more total and injury crashes, especially in down-
town areas. This result was consistent with previous studies
in Baltimore, Maryland (Clifton and Kreamer-Fults 2007),
New York (Ukkusuri et al. 2012), and Hawaii (Kim, Brunner,
and Yamashita 2006). One possible reason is that commer-
cial areas lead to more traffic, as implied by the significant
correlation coefficient (0.259) between commercial areas
and travel volume in this study; more traffic increases the
overall crash risk (Ukkusuri et al. 2012). Downtown loca-
tions also experienced higher traffic volumes than did the
peripheral areas examined in this study. Offices act as trip
attractors during peak hours, producing more traffic and, in
turn, increasing the number of injury crashes, especially in
downtown areas. Moreover, commercial areas adjacent to
arterial roads led to safety problems. One possible explana-
tion of this finding is linked to auto-oriented access to com-
mercial land uses. Most stores are located behind parking
lots, a design that increases conflicts between pedestrians
and vehicles.
What strategies should planners and policymakers use to
respond to these traffic safety issues? This research identi-
fied three crucial directions for planning practice: (1) reduc-
ing travel demand, especially in downtown locations; (2)
retrofitting high-speed and high-volume roads; and (3)
designing land areas that generate fewer lower vehicle trips.
Reducing Travel Demand
Downtown areas, because of their higher traffic volume,
were associated with an increased number of crashes.
Consequently, it is necessary to find ways of strategically
reducing travel demand to downtown locations. Commute
Trip Reduction (CTR) programs are one feasible approach
for planners seeking to reduce workers’ reliance on automo-
biles. Such programs include several Travel Demand
Management (TDM) strategies (e.g., carpooling, telework-
ing, and flexible work times) that have been demonstrated as
effective in reducing automobile dependence (Boarnet and
Handy 2010; Sammer and Saleh 2012). Commuter financial
incentives such as parking cash-out programs that provide
the equivalent cost of subsidized parking can also shift auto-
mobile use to alternative commuting modes (Salon et al.
2012; Shoup 1997).
Retrofitting High-Speed and High-Volume Roads
Since highways/freeways and arterial roads were associated
with more total and fatal crashes and local roads were related
to fewer fatal and injury crashes (especially in downtown
locations), retrofitting these high-speed and high-volume
roads to encourage lower traffic speeds may provide safety
benefits. Complete Streets programs that retrofit downtown
thoroughfares for diverse users and activities have been
found to lower vehicle travel speeds in urban downtown
areas (Ranahan, Lenker, and Maisel 2014). A report from
Smart Growth America 2015 compared the before and after
levels of traffic safety for thirty-seven Complete Streets proj-
ects and found that 70 percent of the roads saw significant
reductions in collisions after project implementation.
Designing Land Areas That Generate Fewer
Vehicle Trips
The results of this research showed that a combination of
arterial roads and commercial land uses led to safety prob-
lems. Therefore, it is important to design trip-attracting land
uses into the other way to lower vehicle trip generations.
Car-free planning involves designing particular areas to
accommodate minimal automobile use. Commercial centers
and districts that feature a concentration of businesses and
convenient accessibility are an alternative to automobile-
oriented suburban strip designs. Because people travel short
distances and use alternative transportation modes in these
districts, the total number of automobile trips is reduced
(Holian and Kahn 2012; Voigt and Polenske 2006).
Efforts aimed at reducing automobile travel through pro-
grams and retrofitting (e.g., CTR programs, TDM strategies,
Complete Streets programs, and car-free planning) require
planners to work in partnership with multilevel agencies and
stakeholders such as transportation planners and engineers,
neighborhood and business associations, real estate develop-
ers, local governments, etc. It is also crucial to note that shift-
ing the travel demand away from downtown areas may
disperse development (sprawl), potentially leading to a lower
number of low-injury crashes but a higher number of fatal
crashes (Ewing, Hamidi, and Grace 2016).
Conclusion
The associations among built environments and traffic safety
vary by each area’s characteristics. However, studies apply-
ing GLMs have not yet focused on the nonstationary issue in
traffic safety analysis. A local approach would be an efficient
way of investigating local variations in each spatial unit. This
study examined spatial variation in the relationships among
built environments and traffic safety in order to inform plan-
ning practices, with the intent of improving safety in high-
risk areas.
Several limitations of this study need to be addressed.
First, the GIS data sets had slight variations in their time
frames, because of limited availability. However, these GIS
data provided objective measurements for built environ-
ments and made it possible to directly translate the study
results into intervention strategies. Second, some informa-
tion or data were not available for the analysis. For example,
because the data for the number of people walking, biking,
and taking public transit were not available, this study used
proxy measures: the number of workers commuting to work
by walking, biking, and taking public transit, as obtained
12 Journal of Planning Education and Research
from the 2010 American Community Survey. Moreover, the
results showed that built environments such as commercial
and office areas caused traffic safety issues. It is possible that
more traffic accidents occurred because of the significant
population flowing around these environments. However,
information regarding such population flows was not avail-
able. Third, caution is needed when generalizing study
results. The GWNBRs served as an effective tool for estimat-
ing nonstationary relationships among crashes and the inde-
pendent variables because they produced local coefficients
for each spatial unit in Austin. However, the results cannot be
spatially transferred and generalized to other areas. Different
jurisdictions should develop their own models to guide them
in their safety planning. Finally, a related problem for global
model selection was the spatial autocorrelation issue. The
individual spatial unit could not to be considered indepen-
dent because the characteristics of the particular spatial units
could have been influenced by adjacent areas (Fotheringham,
Brunsdon, and Charlton 2002).
Despite the above limitations of this study, the results of
the nonstationary relationships among crashes and indepen-
dent variables in each spatial unit indicated that a uniform
policy may not be appropriate for the city of Austin. This
research showed that highways/freeways, arterial roads,
commercial areas, and office areas saw more crashes in
downtown areas than in the periphery. The findings have
several implications for planners and policy makers. It is
important for multilevel stakeholders and agencies to col-
laborate when implementing strategies designed to reduce
travel demand, retrofit high-speed and high-volume roads,
and design land areas to generate fewer vehicle trips. The
application of local models may be used to adjust the outputs
of global models insensitive to the nonstationary issue and
guide the allocation of limited resources to improve traffic
safety. From the perspective of spatial planning, governmen-
tal agencies should put more planning effort in downtown
spaces, since one unit change of certain built environmental
characteristics was associated with more of an increase in
crashes than the same unit change in peripheral areas. Since
understanding urban patterns and their relationship to traffic
safety on a local scale is important in guiding urban develop-
ment, this base study provides a potential approach to guid-
ing the allocation of limited resources to improve traffic
safety, as well as formulating different policies to fit a variety
of locations. Uniform policies and programs are unlikely to
result in expected outcomes for all areas, and therefore need
to be carefully considered. The repeated application of local
models over time will also highlight the dynamic associa-
tions among built environments and traffic safety, by allow-
ing stakeholders to consider the issue of spatial nonstationary
processes on a local scale.
Acknowledgments
We would like to thank Drs. Xuemei Zhu and Chanam Lee, who
provided invaluable suggestion on the manuscript. We would also
like to thank Dr. Chih-Hao Wang for the suggestion on the model-
ing. We would also like to thank JPER’s editors and reviewers for
their invaluable comments and recommendations on an earlier ver-
sion of this article.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect
to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, author-
ship, and/or publication of this article.
References
Abdel-Aty, M., J. Lee, C. Siddiqui, and K. Choi. 2013. “Geographical
Unit Based Analysis in the Context of Transportation Safety
Planning.” Transportation Research Part A: Policy and
Practice 49:62–75.
Aguero-Valverde, J., and P. P. Jovanis. 2008. “Analysis of Road
Crash Frequency With Spatial Models.” Transportation
Research Record 2061:55–63.
Anselin, L. 1988. Spatial Econometrics: Methods and Models.
Boston: Kluwer Academic.
Belsley, D. A., E. Kuh, and R. E. Welsch. 1980. Regression
Diagnostics: Identifying Influential Data and Sources of
Collinearity. New York: John Wiley.
Boarnet, M., and S. Handy. 2010. “Draft Policy Brief on the Impacts
of Residential Density Based on a Review of the Empirical
Literature.” In California Air Resources Board. Sacramento,
CA.
Brunsdon, C., A. S. Fotheringham, and M. Charlton. 1998.
“Geographically Weighted Regression: Modelling Spatial
Non-stationarity.” The Statistician 47:431–43.
Cervero, R., and K. Kockelman. 1997. “Travel Demand and the
3Ds: Density, Diversity, and Design.” Transportation Research
Part D: Transport and Environment 2 (3): 199–219.
Cheng, L., S. R. Geedipally, and D. Lord. 2013. “The Poisson–
Weibull Generalized Linear Model for Analyzing Motor
Vehicle Crash Data.” Safety Science 54:38–42.
Cliff, A., and J. Ord. 1973. Spatial Autocorrelation. London: Pion.
Clifton, K. J., C. V. Burnier, and G. Akar. 2009. “Severity of Injury
Resulting from Pedestrian-Vehicle Crashes: What Can We
Learn from Examining the Built Environment?” Transportation
Research Part D: Transport and Environment 14 (6): 425–36.
Clifton, K. J., and K. Kreamer-Fults. 2007. “An Examination of
the Environmental Attributes Associated with Pedestrian–
Vehicular Crashes Near Public Schools.” Accident Analysis
and Prevention 39 (4): 708–15.
Cottrill, C. D., and P. Thakuriah. 2010. “Evaluating Pedestrian
Crashes in Areas with High Low-Income or Minority
Populations.” Accident Analysis and Prevention 42:1718–28.
da Silva, A. R., and T. C. V. Rodrigues. 2014. “Geographically
Weighted Negative Binomial Regression-Incorporating
Overdispersion.” Statistics and Computing 24 (5): 769–83.
de Guevara, F. L., S. P. Washington, and J. Oh. 2004. “Forecasting
Crashes at the Planning Level—Simultaneous Negative
Binomial Crash Model Applied in Tucson, Arizona.” Statistical
Methods and Safety Data Analysis and Evaluation 1897:191–99.
Yu and Xu 13
Deller, S. 2010. “Rural Poverty, Tourism and Spatial Heterogeneity.”
Annals of Tourism Research 37:180–205.
Delmelle, E. C., and J. C. Thill. 2008. “Urban Bicyclists: Spatial
Analysis of Adult and Youth Traffic Hazard Intensity.”
Transportation Research Record 2074:31–39.
Delmelle, E. C., J. C. Thill, and H. H. Ha. 2012. “Spatial
Epidemiologic Analysis of Relative Collision Risk Factors
among Urban Bicyclists and Pedestrians.” Transportation 39
(2): 433–48.
Desapriya, E., M. Sones, T. Ramanzin, S. Weinstein, G. Scime,
and I. Pikel. 2011. “Injury Prevention in Child Death Review:
Child Pedestrian Fatalities.” Injury Prevention 17:4–9.
Dumbaugh, E., and W. H. Li. 2011. “Designing for the Safety of
Pedestrians, Cyclists, and Motorists in Urban Environments.”
Journal of the American Planning Association 77 (1): 69–88.
Dumbaugh, E., and R. Rae. 2009. “Safe Urban Form: Revisiting the
Relationship between Community Design and Traffic Safety.”
Journal of the American Planning Association 75 (3): 309–29.
Eksler, V., and S. Lassarre. 2008. “Evolution of Road Risk
Disparities at Small-Scale Level: Example of Belgium.” Safety
Research 39:417–27.
Eluru, N., C. R. Bhat, and D. A. Hensher. 2008. “A Mixed
Generalized Ordered Response Model for Examining
Pedestrian and Bicyclist Injury Severity Level in Traffic
Crashes.” Accident Analysis and Prevention 40 (3): 1033–54.
Erdogan, S. 2009. “Explorative Spatial Analysis of Traffic Accident
Statistics and Road Mortality among the Provinces of Turkey.”
Safety Research 40:341–51.
Ewing, R., and R. Cervero. 2010. “Travel and the Built Environment.”
Journal of the American Planning Association 76 (3): 265–94.
Ewing, R., and E. Dumbaugh. 2009. “The Built Environment and
Traffic Safety: A Review of Empirical Evidence.” Journal of
Planning Literature 23 (4): 347–67.
Ewing, R., S. Hamidi, and J. B. Grace. 2016. “Urban Sprawl as a
Risk Factor in Motor Vehicle Crashes.” Urban Studies 53 (2):
247–66.
Flahaut, B. 2004. “Impact of Infrastructure and Local Environment
on Road Unsafety—Logistic Modeling with Spatial
Autocorrelation.” Accident Analysis and Prevention 36 (6):
1055–66.
Fotheringham, A., C. Brunsdon, and M. E. Charlton. 2002.
Geographically Weighted Regression: The Analysis of
Spatially Varying Relationships. Chichester, UK: Wiley.
Fotheringham, A. S., C. Brunsdon, and M. E. Charlton. 2000.
Quantitative Geography: Perspectives on Spatial Data
Analysis. London: Sage.
Frank, L. D., T. L. Schmid, J. F. Sallis, J. Chapman, and B. E.
Saelens. 2005. “Linking Objectively Measured Physical
Activity with Objectively Measured Urban Form: Findings
from SMARTRAQ.” American Journal of Preventive Medicine
28 (2): 117–25.
Gilbert, A., and J. Chakraborty. 2010. “Using Geographically
Weighted Regression for Environmental Justice Analysis:
Cumulative Cancer Risks from Air Toxics in Florida.” Social
Science Research 40:273–86.
Graham, D. J., and S. Glaister. 2003. “Spatial Variation in Road
Pedestrian Casualties: The Role of Urban Scale, Density and
Land-Use Mix.” Urban Studies 40 (8): 1591–607.
Ha, H. H., and J. C. Thill. 2011. “Analysis of Traffic Hazard
Intensity: A Spatial Epidemiology Case Study of Urban
Pedestrians.” Computers Environment and Urban Systems 35
(3): 230–40.
Hadayeghi, A., A. S. Shalaby, and B. N. Persaud. 2010.
“Development of Planning Level Transportation Safety
Tools Using Geographically Weighted Poisson Regression.”
Accident Analysis and Prevention 42 (2): 676–88.
Hadayeghi, A., A. S. Shalaby, B. N. Persaud, and C. Cheung.
2006. “Temporal Transferability and Updating of Zonal
Level Accident Prediction Models.” Accident Analysis and
Prevention 38:579–89.
Hadayeghi, A., A. S. Shalaby, and H. N. Persaud. 2003. “Macrolevel
Accident Prediction Models for Evaluating Safety of Urban
Transportation Systems.” Statistical Methods and Modeling
and Safety Data, Analysis, and Evaluation 1840:87–95.
Handy, S. L. 1993. “Regional versus Local Accessibility:
Implications for Non-work Travel.” Transportation Research
Record 1400:58–66.
Harris, R., A. Singleton, D. Grose, C. Brunsdon, and P. Longley.
2010. “Grid-Enabling Geographically Weighted Regression: A
Case Study of Participation in Higher Education in England.”
Transactions in GIS 14:43–61.
Holian, M., and M. Kahn. 2012. “The Impact of Center City
Economic and Cultural Vibrancy on Greenhouse Gas Emissions
from Transportation.” in MTI Report. Mineta Transportation
Institute.
Huang, H., M. Abdel-Aty, and A. Darwiche. 2010. “County-Level
Crash Risk Analysis in Florida.” Transportation Research
Record 2148:27–37.
Hurvich, C. M., and C. Tsai. 1989. “Regression and Time Series
Model Selection in Small Samples.” Biometrika 76:297–307.
Kim, I., M. Brunner, and E. Y. Yamashita. 2006. “Influence of
Land Use, Population, Employment, and Economic Activity
on Accidents.” Transportation Research Record 1953:56–64.
LaScala, E. A., D. Gerber, and P. J. Gruenewald. 2000. “Demographic
and Environmental Correlates of Pedestrian Injury Collisions:
A Spatial Analysis.” Accident Analysis and Prevention 32 (5):
651–58.
LaScala, E. A., P. J. Gruenewald, and F. W. Johnson. 2004.
“An Ecological Study of the Locations of Schools and
Child Pedestrian Injury Collisions.” Accident Analysis and
Prevention 36 (4): 569–76.
LaScala, E. A., F. W. Johnson, and P. J. Gruenewald. 2001.
“Neighborhood Characteristics of Alcohol-Related Pedestrian
Injury Collisions: A Geostatistical Analysis.” Prevention
Science 2 (2): 123–34.
Levine, N., K. E. Kim, and L. H. Nitz. 1995. “Spatial Analysis of
Honolulu Motor Vehicle Crashes: Zonal Generators.” Accident
Analysis and Prevention 27:675–85.
Long, J., and J. Freese. 2006. Regression Models for Categorical
Outcomes Using Stata, 2nd ed. College Station, TX: Stata Press.
Loukaitou-Sideris, A., R. Liggett, and H. G. Sung. 2007. “Death on
the Crosswalk—A Study of Pedestrian-Automobile Collisions
in Los Angeles.” Journal of Planning Education and Research
26 (3): 338–51.
Lu, J. 2013. “The Geographic Distribution of Pedestrian Safety
Projects in New York City: What Social Equity Implications?”
MS thesis, Columbia University Press.
Marshall, W. E., and N. W. Garrick. 2011. “Does Street Network
Design Affect Traffic Safety? ”Accident Analysis and Prevention
43 (3): 769–81.
14 Journal of Planning Education and Research
Miaou, S. 2003. “Modeling Traffic Crash-flow Relationships for
Intersections: Dispersion Parameter, Functional Form, and
Bayes versus Empirical Bayes Methods.” Transportation
Research Record 1840:31–40.
Miranda-Moreno, L. F., P. Morency, and A. M. El-Geneidy. 2011.
“The Link between Built Environment, Pedestrian Activity
and Pedestrian-Vehicle Collision Occurrence at Signalized
Intersections.” Accident Analysis and Prevention 43 (5): 1624–
34.
Morency, P., L. Gauvin, C. Plante, M. Fournier, and C. Morency.
2012. “Neighborhood Social Inequalities in Road Traffic
Injuries: The Influence of Traffic Volume and Road Design.”
American Journal of Public Health 102 (6): 1112–19.
Nakaya, T., A. S. Fotheringham, C. Brundson, and M. Charlton.
2005. “Geographically Weighted Poisson Regression for
Disease Association Mapping.” Statistics in Medicine
24:2695–717.
National Highway Traffic Safety Administration. 2012. Traffic
Safety Facts 2010. Washington, DC: National Highway Traffic
Safety Administration.
Noland, R. 2003. “Traffic Fatalities and Injuries: The Effect of
Changes in Infrastructure and other Trends.” Accident Analysis
and Prevention 35:599–611.
Noland, R., N. Klein, and N. Tulach. 2013. “Do Lower Income
Areas Have More Pedestrian Casualties? ”Accident Analysis and
Prevention 59:337–45.
Noland, R., and M. A. Quddus. 2004. “A Spatially Disaggregate
Analysis of Road Casualties in England.” Accident Analysis
and Prevention 36:937–84.
Openshaw, S. 1984. The Modifiable Areal Unit Problem. Norwich,
England.
Pirdavani, A., T. Bellemans, T. Brijs, B. Kochan, and G. Wets.
2014. “Assessing the Road Safety Impacts of a Teleworking
Policy by Means of Geographically Weighted Regression
Method.” Journal of Transport Geography 39:96–110.
Rabe-Hesketh, S., and A. Skrondal. 2012. Multilevel and
Longitudinal Modeling Using Stata, 3rd ed. College Station,
TX: Stata Press.
Ranahan, M., J. Lenker, and J. Maisel. 2014. Evaluating the Impact
of Complete Streets Initiatives, edited by Center for Inclusive
Design & Environmental Access. University at Buffalo School
of Architecture and Planning, Buffalo, NY.
Salon, D., M. Boarnet, S. Handy, S. Spears, and G. Tal. 2012.
“How Do Local Actions Affect VMT? A Critical Review of
the Empirical Evidence.” Transportation Research Part D:
Transport and Environment 17 (7): 495–508.
Sammer, G., and W. Saleh. 2012. Travel Demand Management and
Road User Pricing: Success, Failure and Feasibility. Farnham,
UK: Ashgate.
Sengoelge, M., B. Elling, L. Laflamme, and M. Hasselberg. 2013.
“Country-Level Economic Disparity and Child Mortality
Related to Housing and Injuries: A Study in 26 European coun-
tries.” Injury Prevention 19 (5): 311–15.
Shariat-Mohaymany, A., M. Shahri, B. Mirbagheri, and A. A.
Matkan. 2015. “Exploring Spatial Non-stationarity and Varying
Relationships between Crash Data and Related Factors Using
Geographically Weighted Poisson Regression.” Transactions
in GIS 19 (2): 321–37.
Shoup, D. 1997. “Evaluating the Effects of Cashing Out Employer-
Paid Parking: Eight Case Studies.” Transport Policy 4:201–16.
Siddiqui, C., M. Abdel-Aty, and K. Choi. 2012. “Macroscopic
Spatial Analysis of Pedestrian and Bicycle Crashes.” Accident
Analysis and Prevention 45:382–91.
Smart Growth America. 2015. Safer Streets, Stronger Economies:
Complete Streets Project Outcomes from across the Country.
Washington, DC: Smart Growth America.
Sze, N. N., and S. C. Wong. 2007. “Diagnostic Analysis of the
Logistic Model for Pedestrian Injury Severity in Traffic
Crashes.” Accident Analysis and Prevention 39 (6): 1267–78.
Teigen, A., and D. Shinkle. 2012. “Traffic Safety and Public Health:
State Legislative Action 2011.” In Transportation Series,
February 2012, no. 36. Denver, Co: National Conference of
State Legislatures.
Ukkusuri, S., L. F. Miranda-Moreno, G. Ramadurai, and J. Isa-
Tavarez. 2012. “The Role of Built Environment on Pedestrian
Crash Frequency.” Safety Science 50 (4): 1141–51.
Voigt, K., and J. Polenske. 2006. “Applying New Urbanism Street
Principles in Downtown Milwaukee, WI, USA.” ITE Journal
76 (5): 26–30.
Washington, S. P., I. Van Schalkwyk, S. Mitra, M. Meyer,
E. Dumbaugh, and M. Zoll. 2006. NCHRP Report 546:
Incorporating Safety into Long-Range Transportation
Planning. Washington, DC: Transportation Research Board.
Wier, M., J. Weintraub, E. H. Humphreys, E. Seto, and R. Bhatia.
2009. “An Area-Level Model of Vehicle-Pedestrian Injury
Collisions with Implications for Land Use and Transportation
Planning.” Accident Analysis and Prevention 41 (1): 137–45.
Xu, P., H. Huang, N. Dong, and M. Abdel-Aty. 2014. “Sensitivity
Analysis in the Context of Regional Safety Modeling:
Identifying and Assessing the Modifiable Areal Unit Problem.”
Accident Analysis & Prevention 70:110–20.
Ye, X., R. M. Pendyala, S. Washington, K. Konduri, and J. Oh.
2009. “A Simultaneous Equations Model of Crash Frequency
by Collision Type for Rural Intersections.” Safety Science
47:443–52.
Yu, C. 2014. “Environmental Supports for Walking/Biking and
Traffic Safety: Income and Ethnicity Disparities.” Preventive
Medicine 67:12–16.
Yu, C. 2015a. “Built Environmental Designs in Promoting
Pedestrian Safety.” Sustainability 7 (7): 9444–60.
Yu, C. 2015b. “How Differences in Roadways Affect School
Travel Safety.” Journal of the American Planning Association
81 (3): 203–20.
Yu, C., and X. Zhu. 2016. “Planning for Safe Schools: Impacts
of School Siting and Surrounding Environments on Traffic
Safety.” Journal of Planning Education and Research 36 (4):
476–86.
Zajac, S. S., and J. N. Ivan. 2003. “Factors Influencing Injury
Severity of Motor Vehicle–Crossing Pedestrian Crashes in
Rural Connecticut.” Accident Analysis and Prevention 35 (3):
369–79.
Zegeer, C. V., C. Seiderman, P. Lagerwey, M. Cynecki, M.
Ronkin, and R. Schnieder. 2002. “Pedestrian Facilities Users
Guide—Providing Safety and Mobility.” In Federal Highway
Administration, edited by Report No. FHWa-RD-102-01.
Washington, DC.
Yu and Xu 15
Zegras, P. C. 2007. The Built Environment and Motor Vehicle
Ownership and Use: Evidence from Santiago de Chile.
Washington, DC: Transportation Research Board.
Zhang, H., J. Zhang, S. Lu, S. Cheng, and J. Zhang. 2011. “Modeling
Hotel Room price with Geographically Weighted Regression.”
International Journal of Hospitality Management 30:1036–43.
Zhang, Y., J. Bigham, Z. Li, D. Ragland, and X Chen. 2012.
“Associations between Road Network Connectivity and
Pedestrian-Bicyclist Accidents.” Paper presented at:
Transportation Research Board 91st Annual Meeting.
Washington, DC: Transportation Research Board.
Author Biographies
Chia-Yuan Yu is an Assistant Professor in the Urban & Regional
Planning program at the School of Public Administration at
University of Central Florida. His research interests include resi-
dential self-selection, children’s active commute to school, and traf-
fic safety.
Minjie Xu is pursuing a doctoral degree in Urban and Regional
Science at Texas A&M University. His research interests are built
environment and traffic safety, active transportation and active liv-
ing research, and economic benefits of walkable environments.
