This document outlines a framework to identify factors associated with high pedestrian and bicycle crash locations in Tennessee using geographic information systems and statistical analysis. The methodology involves collecting crash, roadway, demographic, and socioeconomic data; geocoding the crash data in GIS; performing cluster and hot spot analysis to identify crash clusters; and developing statistical models. Specific high crash zones are identified in Davidson County for further analysis. The goal is to prioritize funding to reduce severe crashes.
1. Daniel Emaasit
Master of Civil Engineering Candidate
Advisor: Dr. Deo Chimba
Civil and Architectural Engineering
FRAMEWORK TO IDENTIFY FACTORS
ASSOCIATED WITH HIGH PEDESTRIAN
AND BICYCLE CRASH LOCATIONS
USING GEOGRAPHIC INFORMATION
SYSTEM AND STATISTICAL ANALYSIS
By
2. OUTLINE
Introduction
Statement of the Problem
Objective
Literature Review
Study Data
Methodology
GIS Geo-Coding
Cluster Analysis
Hot Spot Analysis
Statistical Modeling
Results
Conclusions & Recommendations
3. INTRODUCTION
• Bicyclists and pedestrians are a class of
vulnerable road users that are often over-
represented in fatal or incapacitating injury
crash statistics.
• While passenger car fatalities have shown
sharp declines in the last decade in
Tennessee, pedestrian and bike fatalities
have remained relatively constant.
• A robust methodology is not currently
available to identify bicycle and pedestrian
high-crash locations in Tennessee.
4. STATEMENT OF THE PROBLEM
TDOT has an extensive road safety audit program
which uses criteria based on the ratio of crashes to
average daily traffic.
That program does not target locations with a high
number of bike/pedestrians crashes since there are no
bicycle and pedestrian counts.
A robust methodology is not currently available to
identify bicycle and pedestrian high-crash locations in
Tennessee.
The challenge is allocating funds, from TDOT’s
Highway Safety Improvement Program (HSIP),
equitably among rural and urban areas in a way that is
most effective at reducing bicycle and pedestrian
fatalities and incapacitating injuries.
5. OBJECTIVE
• To develop a framework to identify factors
associated with bicycle and pedestrian
high crash locations for investment
prioritization of TDOT funds to maximize
the reduction in state-wide severe bicycle
and pedestrian crashes.
7. LITERATURE REVIEW Application of GIS in Pedestrian and Bicyclist Safety Analysis
GIS has been used by several researchers for display and cluster
analysis of these type of safety studies
GIS can improve data collection, integrate crash data with
roadway, traffic, demographic data and display results graphically
to enhance decision-making capabilities.
Statistical Modeling of Ped/Bike Crashes
Count models are recommended for modeling ped/bike crash
frequencies.
Logistic regression, ordered-response, Multinomial models have
been applied for modeling ped/bike injury severities.
One of the key limitations that hinder ped/bike safety analysis is
the lack of travel exposure information.
Cluster analysis
Spatial statistical tools in GIS are recommended for cluster
analysis.
8. STUDY DATA Categories of data used:-
Crash data
Roadway Data
Geospatial data
Demographic and Socioeconomic data
CRASH DATA
Obtained from three different sources:-
TRIMS:- A total of 7,503 pedestrian crash incidents and 2,558
bicyclist crash incidents were downloaded from TRIMS.
TITAN:- Provided additional micro-level information about the
crashes from TRIMS such as crash city, urban-rural designation,
location highway street, location estimate, location distance type,
location direction, location from intersection, etc.
Study Period:- 7-Year Crash data (2003 to 2009)
9. CRASH DATA-Descriptive statistics
As shown, we used 7-years of crash data which is
more than the minimum of 3-years recommended in
the literature and previous studies
10. CRASH DATA-Descriptive statistics
County % of Total Ped % of Total Bike
Shelby 33.1 26.3
Davidson 20.4 16.3
Hamilton 6.9 8.3
Knox 6.9 8.2
Montgomery 2.4 3.2
Rutherford 2.2 5.8
Sullivan 1.9 2.6
Madison 1.6 1.2
Crash City % of Total Ped % of Total Bike
MEMPHIS 32.5 23.0
NASHVILLE 19.4 15.1
CHATTANOOGA 5.9 6.1
KNOXVILLE 5.4 7.0
CLARKSVILLE 1.9 3.0
JACKSON 1.4 1.0
MURFREESBORO 1.2 4.7
JOHNSON 1.1 1.3
KINGSPORT 1.0 1.5
13. ROADWAY GEOMETRY GEO-SPATIAL DATA
TDOT provided the following geospatial data files in the
form of shapefiles:-
TDOT road geometrics:- consisted of spatial data of the
entire roadway network in Tennessee which contains
information such as route number, begin and end log miles,
codes for land use, posted speed limit, number of lanes,
terrain, illumination etc.
Tennessee Road TIPS:- consisted of spatial data of the
entire roadway network in Tennessee included in the
geometry data but with more detailed information including
the zip code, road name and the distance from the reference
point such as from the intersection or known node.
14. DEMOGRAPHIC & SOCIOECONOMIC DATA
A Tennessee Census tract shapefile was downloaded from the TIGER
webpage of the US census website:
A Census tracts are the smallest geographic area for which the
Census Bureau collects and tabulates decennial census data
2010 US decennial census demographic and socioeconomic data was
downloaded at census tract level from the US census website.
Demographic data consists of:-
counts of population,
housing,
race, and
age distribution
Socio-economic data consists of:-
income,
vehicle availability,
employment,
commuting to work,
occupations,
poverty status data.
15. INTEGRATING ALL STUDY DATA INTO GIS
CRASH DATA
A key component in identifying high crash zones
involves accurately coding the location of crashes on
digital maps.
This was done in a GIS environment using the
“addressmatch” feature for address-type crash data and
the “linear referencing” feature for the highway-type crash
data.
Address-type crash data:- consists of location information such
as street name, intersection name, distance from a reference
point etc. Commonly used in urban areas.
Highway-type crash data:- consists of mile-post or log-mile
location information used to geocode crash points along
highways. Commonly used in rural areas.
17. CODING CRASH DATA INTO GIS
CRASH DATA
5584 out of 7500 ped crashes (approx. 75%) were accurately mapped.
1890 out of 2558 bike crashes (approx. 74%) were accurately mapped
Note that some crashes had un-recognizable route numbers such as “M0000”
and “C0000” for GIS geocoding
Distribution of
Pedestrian
Crashes
Distribution of
Bicyclist
Crashes
High concentrations in Shelby, Davidson, Hamilton and Knox Counties.
18. CLUSTER ANALYSIS
It involves finding patterns of observations within a data set.
The combination of neighborhood attributes, social-
economic and demographic data are used to uncover
correlated factors associated with bicycle and pedestrian
crashes.
The objective
To identify locations that experience a significantly higher
percentage of crashes through pattern detection
technique.
To identify attributes (crash, geometrics, demographic
and socio-economic attributes) associated with crash
clusters for further analysis/investigation.
19. CLUSTER ANALYSIS
ANSELIN LOCAL MORAN'S I STATISTIC
To quantify the spatial correlation, the ANSELIN LOCAL
MORAN'S I STATISTIC was used.
This tool identifies spatial clusters of features with high or low
values.
To do this, the tool calculates
Local Moran's I index,
Z-score,
p-value, and
Cluster type.
The z-scores and p-values represent the statistical significance
of the computed index values.
21. CLUSTER ANALYSIS
ANSELIN LOCAL MORAN'S I STATISTIC
The outputs of this statistic:-
The I Index value
Sign of I Value Intepretation Conclusion
Positive (+)
This feature has neighboring
features with similarly high
or low attribute values
This feature is
part of a cluster
Negative (-)
This feature has neighboring
features with dissimilar
values
This feature is an
outlier
Z Value Intepretation
Z>1.96
This feature has neighboring features with
similarly high or low attribute values
Z<-1.96
This feature has neighboring features with
dissimilar values
The Z score Value
22. CLUSTER ANALYSIS-SHELBY
• SHELBY COUNTY
As shown, pedestrian
crash clusters are
associated with
areas with high
population density
of African American
Distribution of Crash Clusters per Population
density of African American population
23. CLUSTER ANALYSIS-SHELBY
• SHELBY COUNTY Distribution of Crash
Clusters per Population
density of Whites population
1. As shown, predominantly
whites populated areas are
associated with low
pedestrian crash clusters
2. However, there are some
few areas shown to have
pockets of pedestrian
clusters which will also be
further investigated for the
possibility of being hot
spots for TDOT
considerations
25. CLUSTER ANALYSIS--SHELBY
• SHELBY COUNTY
Distribution of Crash Clusters with Households that have No
Vehicle and Proportion of workers who Walk to Work
27. HOTSPOT ANALYSIS
Gi* Spatial Statistic
The Gi* index was used to locate unsafe road segments and
intersections and discern cluster structures of high- or low-value
concentration among local observations
A simple form of the Gi* statistic as defined by Getis and Ord(1995)
Where
Gi* = statistic that describes the spatial dependency of incident J over all
n events,
xj = magnitude of variable X at incident location j
wij = weight value between event i and j that represents their spatial
interrelationship.
n = the number of incidents
Getis-Ord Hot spot Analysis
32. STATISTICAL MODELING
A comparative crash pattern and trend was performed
Development of statistical crash models.
The models examine relationships between pedestrian/bicycle
crashes with respect to:-
Demographic characteristics,
Population,
Socio-economic characteristics,
Age groups,
Neighborhood and land use characteristics,
Roadway geometry and features,
Traffic flow,
Speed characteristics.
34. STATISTICAL MODELING
Criteria for Modeling Crash Frequency
Poisson and negative binomial distributions are often more appropriate
for modeling discrete counts of events
Poisson Regression model
The probability of section i having yi crashes per year is (Cameroon and
Trivedi, 1998)
– yi = 0,1,2....
– μ = the expected (mean) number of crashes
Negative Binomial Regression Model
The p.m.f. of the Negative Binomial (NB) model is (Cameroon and
Trivedi, 1998) :
– mean μ = E( y) = v exp(Xβ ).
– variance is Var( y) = μ +αμ2 .
36. Fatal Pedestrian Crashes
PDO Pedestrian Crashes
Incap Bicycle Crashes
Non Incap Bicycle Crashes
PDO Bicycle Crashes
Injury Bicycle Crashes
Selecting Modeling Distribution
Negative Binomial Vs Poisson
STATISTICAL MODELING
37. MODEL ESTIMATION RESULTS
Negative Binomial Regression
Number of observations = 152
Fatal Pedestrian Crashes Coefficient Std. Err. Z-Value
Traffic Volume (AADT) 0.00002 9.05E-06 1.96
Households with Income from $25000 to $49999 (%) 0.0040 0.020 0.2
Households with Income from $50000 to $74999 (%) -0.0279 0.034 -0.81
Households with Income from $75000 to $99999 (%) -0.0437 0.071 -0.61
Occupied housing units with no vehicle (%) 0.0348 0.015 2.27
Occupied housing units with 2 vehicles (%) -0.0173 0.028 -0.63
Occupied housing units with 3 or more vehicles (%) -0.0036 0.040 -0.09
POPN of 16 years and over in Civilian labor force (%) -0.0089 0.015 -0.6
Households with Food Stamp benefits (%) 0.0141 0.014 1.04
Economic Factors-Pedestrian
Negative Coefficient Positive Coefficient
38. MODEL ESTIMATION RESULTS
Economic Factors-Bicycle
Poisson Regression
Number of observations = 42
Non-Incapacitating Crashes Coefficient Std. Err. Z-Value
Traffic Volume (AADT) 1.46E-06 1.48E-06 0.99
Households with Income below $25000 (%) 0.0035 0.0156 0.22
Households with Income from $25000 to $49999 (%) 0.0051 0.0211 0.24
Households with Income from $50000 to $74999 (%) 4.80E-02 0.0315 1.53
Households with Income from $75000 to $99999 (%) -0.0033 3.55E-02 -0.09
Mean Household Income ($) -4.99E-07 0.00001 -0.05
Occupied housing units with No vehicle (%) 0.0099 0.0199 0.5
Occupied housing units with 1 vehicles (%) -0.0190 0.0160 -1.19
POPN of 16 years and over in Civilian labor force (%) -0.0011 0.0148 -0.07
Households with Food Stamp benefits (%) 0.0107 0.0180 0.6
39. MODEL ESTIMATION RESULTS
Negative Binomial Regression
Number of observations = 152
Fatal Pedestrian Crashes Coefficient Std. Err. Z-Value
Area of Zone -64.3004 29.461 -2.18
Land Use Type
Fringe 0.1538 0.5531 0.28
Residential & Public parks -0.9638 0.6875 -1.4
Speed limit
30mph to 40mph 13.7433 1150 0.01
45mph 13.9741 1150 0.01
Presence of School speed limit -13.6720 1005 -0.01
Number of lanes 0.2609 0.1486 1.76
Traffic (AADT) 0.00003 0.00001 2.52
Constant -24.0749 1150 -0.02
Length Exposure
Roadway Factors-Pedestrian
40. MODEL ESTIMATION RESULTS
Roadway Factors-Bicycle
Negative Binomial Regression
Number of observations = 40
Injury Bicycle Crashes Only Coefficient Std. Err. Z-Value
Right of Way -0.0433 0.0254 -1.7
Rolling terrain 2.5925 1.2541 2.07
Land Use Type
Fringe 2.8718 1.3849 2.07
Residential & Public parks -0.1440 0.8032 -0.18
Presence of School Speed Limit -22.034 17402 0
Number of Lanes -0.0129 0.3903 -0.03
Traffic (AADT) 1.13E-05 7.32E-06 1.54
41. MODEL ESTIMATION RESULTS
Age Factors-Pedestrian
Negative Binomial Regression
Number of observations = 152
Fatal Pedestrian Crashes Coefficient Std. Err.Z-Value
Population under 10yrs (%) 0.0009 0.0225 0.04
Population from 10 to 19yrs (%) -0.0329 0.0208 -1.58
Population from 20 to 29yrs (%) -0.0205 0.0155 -1.32
Population from 30 to 64yrs (%) -0.0119 0.0089 -1.34
Where;
PCF=Fatal Pedestrian Crashes
P1 = Population under 10yrs (%),
P2 = Population from 10 to 19yrs (%),
P3 = Population from 20 to 29yrs (%),
P4 = Population from 30 to 64yrs (%).
42. MODEL ESTIMATION RESULTS
Age Factors-Bicycle
Negative Binomial Regression
Number of observations = 42
Injury Bicycle Crashes Only Coefficient Std. Err. Z-Value
Traffic Volume (AADT) 3.01E-06 3.94E-06 0.76
Population under 10yrs (%) 0.0721 0.0729 0.99
Population from 10 to 19yrs (%) 0.0185 0.0500 0.37
Population from 20 to 29yrs (%) -0.0331 0.0342 -0.97
Population from 30 to 64yrs (%) -0.0470 0.0419 -1.12
Population above 64yrs (%) 0.0018 0.0883 0.02
Where;
BCInj= injury bicycle crashes only,
AADT = Traffic Volume,
P1 = Population under 10yrs (%),
P2 = Population from 10 to 19yrs (%),
P3 = Population from 20 to 29yrs (%),
P4 = Population from 30 to 64yrs (%),
43. MODEL ESTIMATION RESULTS
Race Factors-Pedestrian
Negative Binomial Regression
Number of observations = 152
Fatal Pedestrian Crashes Coefficient Std. Err. Z-Value
White Population (%) -0.0042 0.0629 -0.07
Black Population (%) 0.0058 0.0629 0.09
American-Indian Population (%) 0.8412 0.8326 1.01
Asian Population (%) -0.0171 0.0907 -0.19
Traffic volume (AADT) 0.00002 7.41E-06 2.55
Constant -9.8592 6.1567 -1.6
Length of Crash Zone Exposure
44. MODEL ESTIMATION RESULTS
Race Factors-Bicycle
Negative Binomial Regression
Number of observations = 42
Injury Bicycle Crashes Only Coefficient Std. Err. Z-Value
White population (%) 0.0425 0.0126 3.37
Black population (%) 0.0452 0.0062 7.23
Asian population (%) -0.2131 0.2918 -0.73
Hispanic population (%) 0.0437 0.0256 1.7
Traffic Volume (AADT) 1.79E-06 3.43E-06 0.52
Area of Zone Exposure
45. Injury Crashes Only Coefficient Std. Err. Z-Value P-Value
Right of Way -0.0433 0.0254 -1.7 0.089 -0.0931 0.0065
Rolling terrain 2.5925 1.2541 2.07 0.039 0.1345 5.0506
Landuse
Fringe 2.8718 1.3849 2.07 0.038 0.1575 5.5862
Residential & Public parks -0.1440 0.8032 -0.18 0.858 -1.7183 1.4302
Presence of School Speed Limit -22.034 17402 0 0.999 -34129 34085
Number of Lanes -0.0129 0.3903 -0.03 0.974 -0.7779 0.7520
Traffic (AADT) 0.0000 7.32E-06 1.54 0.123 -3.05E-06 2.56E-05
Alpha 1.2046 1.0796 0.2080 6.9774
95% Conf. Interval
Log likelihood = -29.966986
Likelihood-ratio test of alpha=0: chibar2(01) = 2.80 Prob>=chibar2 = 0.047
Negative Binomial Regression
Number of obs = 40
Wald chi2(7) = 9.47
Prob > chi2 = 0.2207
MODEL ESTIMATION RESULTS
All Crashes Combined Coefficient Std. Err. Z-Value P-Value
County
Hamilton & Knox -0.3891 0.4643 -0.84 0.402 -1.2991 0.5209
Davidson 0.4717 0.2991 1.58 0.115 -0.1146 1.0580
Shelby 0.1392 0.5122 0.27 0.786 -0.8646 1.1430
Right of Way 0.0041 0.0070 0.58 0.563 -0.0097 0.0178
Rolling terrain 0.1674 0.3465 0.48 0.629 -0.5118 0.8465
Landuse
Fringe 0.4939 0.5351 0.92 0.356 -0.5549 1.5427
Residential & Public parks 0.0624 0.2946 0.21 0.832 -0.5149 0.6397
Speed Limit
35mph to 40mph 0.4565 0.2968 1.54 0.124 -0.1252 1.0382
45mph to 55mph 0.5311 0.4953 1.07 0.284 -0.4397 1.5020
Presence of School Speed Limit -0.7365 0.6649 -1.11 0.268 -2.0396 0.5667
Number of Lanes -0.0400 0.1651 -0.24 0.808 -0.3637 0.2836
Traffic Volume (AADT) 8.78E-07 1.59E-06 0.55 0.581 -2.24E-06 3.99E-06
Poisson Regression
Number of obs = 40
Wald chi2(12) = 113.41
Prob > chi2 = 0
95% Conf. Interval
Log likelihood = -61.348043
Injury Crashes Only Coefficient Std. Err. Z-Value P-Value
Right of Way -0.0433 0.0254 -1.7 0.089 -0.0931 0.0065
Rolling terrain 2.5925 1.2541 2.07 0.039 0.1345 5.0506
Landuse
Fringe 2.8718 1.3849 2.07 0.038 0.1575 5.5862
Residential & Public parks -0.1440 0.8032 -0.18 0.858 -1.7183 1.4302
Presence of School Speed Limit -22.034 17402 0 0.999 -34129 34085
Number of Lanes -0.0129 0.3903 -0.03 0.974 -0.7779 0.7520
Traffic (AADT) 0.0000 7.32E-06 1.54 0.123 -3.05E-06 2.56E-05
Alpha 1.2046 1.0796 0.2080 6.9774
95% Conf. Interval
Log likelihood = -29.966986
Likelihood-ratio test of alpha=0: chibar2(01) = 2.80 Prob>=chibar2 = 0.047
Negative Binomial Regression
Number of obs = 40
Wald chi2(7) = 9.47
Prob > chi2 = 0.2207
Property Damage Only Coefficient Std. Err. Z-Value P-Value
Right of Way -0.0036 0.0146 -0.24 0.807 -0.0323 0.0251
Rolling terrain 0.9013 0.6131 1.47 0.142 -0.3003 2.1029
Landuse
Fringe 0.3098 1.1676 0.27 0.791 -1.9785 2.5982
Residential & Public parks 0.0903 0.5629 0.16 0.873 -1.0129 1.1935
Presence of School Speed Limit -15.478 2207 -0.01 0.994 -4341 4310
Number of Lanes 0.3008 0.2836 1.06 0.289 -0.2550 0.8566
Traffic Volume (AADT) 2.35E-06 3.98E-06 0.59 0.554 -5.44E-06 0.00001
Constant -2.4546 0.9424 -2.6 0.009 -4.3016 -0.6076
Alpha 7.57E-23 . . .
95% Conf. Interval
Likelihood-ratio test of alpha=0: chibar2(01) = 0.00 Prob>=chibar2 = 1.000
Pseudo R2 = 0.1318
Prob > chi2 = 0.2341
LR chi2(10) =9.27
Number of obs = 40
Negative Binomial Regression
Log likelihood = -30.51258
Non-Incapacitating Crashes Coefficient Std. Err. Z-Value P-Value
County
Hamilton & Knox -0.8152 0.6328 -1.29 0.198 -2.0554 0.4250
Davidson 0.6370 0.3489 1.83 0.068 -0.0469 1.3209
Shelby 0.0517 0.6219 0.08 0.934 -1.1671 1.2706
Area of Zone 16.593 25.657 0.65 0.518 -33.694 66.881
Right of Way 0.0109 0.0086 1.26 0.207 -0.0060 0.0277
Rolling terrain -0.4144 0.4706 -0.88 0.379 -1.3367 0.5079
Landuse
Fringe 0.1090 0.6532 0.17 0.867 -1.1711 1.3892
Residential & Public parks 0.0087 0.3606 0.02 0.981 -0.6981 0.7155
Speed Limit
35mph to 40mph 0.6719 0.3557 1.89 0.059 -0.0253 1.3690
45mph to 55mph 0.7332 0.6810 1.08 0.282 -0.6015 2.0679
Presence of School Speed Limit 0.1938 0.6988 0.28 0.782 -1.1758 1.5634
Number of Lanes -2.82E-01 2.01E-01 -1.4 0.161 -6.77E-01 0.1126
Traffic Volume(AADT) 2.06E-07 1.95E-06 0.11 0.916 -3.62E-06 4.03E-06
Log likelihood = -54.6222
95% Conf. Interval
Poisson Regression
Number of obs = 40
Wald chi2(13) = 39.63
Prob > chi2 = 0.0002
Injury Crashes Only Coefficient Std. Err. Z-Value P-Value
White population (%) 0.0425 0.0126 3.37 0.001 0.0178 0.0672
Black population (%) 0.0452 0.0062 7.23 0 0.0329 0.0574
Asian population (%) -0.2131 0.2918 -0.73 0.465 -0.7850 0.3588
Hispanic population (%) 0.0437 0.0256 1.7 0.089 -0.0066 0.0939
Traffic Volume (AADT) 1.79E-06 3.43E-06 0.52 0.602 -4.94E-06 8.52E-06
Area of Zone
Alpha 0.5840 0.9122 0.0274 12.4718
Prob > chi2 = 0
Exposure
95% Conf. Interval
Likelihood-ratio test of alpha=0: chibar2(01) = 0.83 Prob>=chibar2 = 0.181
Log likelihood = -32.055339
Negative Binomial Regression
Number of obs = 42
Wald chi2(5) = 192.53
Injury Crashes Only Coefficient Std. Err. Z-Value P-Value
Traffic Volume (AADT) 3.01E-06 3.94E-06 0.76 0.446 -4.72E-06 1E-05
Population under 10yrs (%) 0.0721 0.0729 0.99 0.322 -0.0707 0.2149
Population from 10 to 19yrs (%) 0.0185 0.0500 0.37 0.711 -0.0794 0.1165
Population from 20 to 29yrs (%) -0.0331 0.0342 -0.97 0.334 -0.1002 0.0340
Population from 30 to 64yrs (%) -0.0470 0.0419 -1.12 0.262 -0.1290 0.0351
Population above 65yrs (%) 0.0018 0.0883 0.02 0.983 -0.1712 0.1749
Alpha 1.9899 1.4888 0.4592 8.6236
Likelihood-ratio test of alpha=0: chibar2(01) = 5.58 Prob>=chibar2 = 0.009
95% Conf. Interval
Negative Binomial Regression
Number of obs = 42
Wald chi2(6) = 9.04
Prob > chi2 = 0.1716
Log likelihood = -33.027192
Property Damage Only Coefficient Std. Err. Z-Value P-Value
Population under 10yrs (%) -0.0155 0.0468183 -0.33 0.74 -0.1073 0.0762
Population from 10 to 19yrs (%) 0.0261 0.0393262 0.66 0.508 -0.0510 0.1031
Population from 20 to 29yrs (%) 0.0654 0.0142064 4.61 0 0.0376 0.0933
Population from 30 to 64yrs (%) 0.0791 0.0182419 4.34 0 0.0433 0.1148
Population above 65yrs (%) -0.0609 0.0706128 -0.86 0.389 -0.1993 0.0775
Area of Zone
Alpha 9.38E-07 0.0017372 0 .
Likelihood-ratio test of alpha=0: chibar2(01) = 0.0e+00 Prob>=chibar2 = 0.500
Exposure
Negative Binomial Regression
Number of obs = 42
Wald chi2(5) = 422.19
Prob > chi2 = 0
Log likelihood = -36.13978
95% Conf. Interval
Nonincapacitating Crashes Coefficient Std. Err. Z-Value P-Value
Population under 10yrs (%) 0.0119 0.0350 0.34 0.735 -0.0568 0.0805
Population from 10 to 19yrs (%) -0.0076 0.0359 -0.21 0.833 -0.0780 0.0628
Population from 20 to 29yrs (%) -0.0041 0.0276 -0.15 0.882 -0.0581 0.0500
Population from 30 to 64yrs (%) -0.0100 0.0356 -0.28 0.779 -0.0798 0.0598
Constant 1.0244 2.8298 0.36 0.717 -4.5219 6.5707
95% Conf. Interval
Poisson Regression
Number of obs = 42
LR chi2(4) = 0.73
Prob > chi2 = 0.9471
Pseudo R2 = 0.0055
Log likelihood = -66.025984
Incapacitating Crashes Coefficient Std. Err. Z-Value P-Value
Population under 10yrs (%) 0.0566 0.0705 0.8 0.422 -0.0817 0.1949
Population from 10 to 19yrs (%) 0.0162 0.0481 0.34 0.737 -0.0781 0.1104
Population from 20 to 29yrs (%) -0.0196 0.0265 -0.74 0.459 -0.0715 0.0323
Population from 30 to 64yrs (%) -0.0326 0.0385 -0.85 0.397 -0.1082 0.0429
Population above 65yrs (%) -0.0218 0.0870 -0.25 0.802 -0.1923 0.1487
Alpha 1.8868 1.4909 0.4010 8.8780
Likelihood-ratio test of alpha=0: chibar2(01) = 4.85 Prob>=chibar2 = 0.014
95% Conf. Interval
Negative Binomial Regression
Number of obs = 42
Wald chi2(5) = 9.63
Prob > chi2 = 0.0865
Log likelihood = -32.522491
Injury Crashes Only Coefficient Std. Err. Z-Value P-Value
Traffic Volume (Average AADT) 1.77E-06 3.07E-06 0.58 0.565 -4.25E-06 7.78E-06
Households with Income & Benefits below $25000 (%) 0.0243 0.0395 0.6200 0.5380 -0.0531 0.1016
Households with Income & Benefits from $25000 to $49999 (%) 0.0671 0.0404 1.6600 0.0970 -0.0121 0.1464
Households with Income & Benefits from $50000 to $74999 (%) 0.1833 0.0646 2.8400 0.0050 0.0567 0.3099
Households with Income & Benefits from $75000 to $99999 (%) -0.1185 0.0865 -1.3700 0.1710 -0.2881 0.0510
POP of 16 years and over in Civilian labor force (%) -0.0246 0.0322 -0.7600 0.4460 -0.0877 0.0386
Households with Food Stamp benefits (%) 0.0017 0.0411 0.0400 0.9660 -0.0788 0.0823
Families below poverty level (%) 0.0207 0.0483 0.4300 0.6690 -0.0741 0.1154
Area of Zone
Alpha 0.0508 0.6009 4.26E-12 6.05E+08
Likelihood-ratio test of alpha=0: chibar2(01) = 0.01 Prob>=chibar2 = 0.465
Exposure
95% Conf. Interval
Log likelihood = -29.321777
Negative Binomial Regression
Number of obs = 42
Wald chi2(8) = 338.34
Prob > chi2 = 0
46. Pedestrian Crashes
Fatal
Low Income, No
Vehicle, Food
Stamps, Young Age,
Fringe, Speed,
AADT, Black POPN,
Non-Incap
Rolling terrain,
Fringe & Residential,
Speed, School Zone,
AADT,
Injury
Only
AADT, Lanes,
PDO
Rolling terrain,
Speed, School Zone,
AADT,
MODEL ESTIMATION RESULTS
Summary of Factors with +ve Correlation-Pedestrian
47. Bicycle Crashes
Incap
Low to Middle
Income, Young &
Teens, White & Black
& Hispanic, Rolling
terrain, Fringe &
Residential, Speed,
AADT
Non-
Incap
Low to Middle
Income, No Vehicle,
Food Stamps, AADT,
Low Employment
rate, Young, Fringe &
Residential, Speed,
AADT
Injury
Only
Low to Middle
Income, Poverty
Level, AADT, Low
Employment rate,
Food Stamp, Young
& Teens & Seniors,
White & Black &
Hispanic
PDO
Low Income, Low
Employment rate,
Fringe &
Residential, Speed,
AADT
MODEL ESTIMATION RESULTS
Summary of Factors with +ve Correlation-Bicycle
48. CONCLUSIONS
The objective of the research was:-
To develop a framework to identify factors associated with bicycle and
pedestrian high crash locations.
Two methods were proposed to examine these factors:-
GIS-Cluster Analysis
Statistical Analysis
Major findings include:-
Low Income,
Poverty Level,
Food Stamp Benefits,
No vehicle ownership,
Young & Senior Population,
Black Populated areas,
Traffic Volume,
Fringe neighborhoods
Narrow ROW
Increase Crash Frequency
49. RECOMMENDATIONS
Injury severity Modeling should be performed.
To identify design mitigation issues, such as design of
crosswalks and intersections that influence the outcomes
of pedestrian/Bike crashes.
To provide additional insight into pedestrian behavior (e.g.
impairment by alcohol or drugs) that contributes to the
likelihood of a fatality in a crash.
Other factors should be considered:-
Education level,
Intersection studies,
Time of day, e.t.c.
50. REFERENCES
1. Harkey, D. (1999). Development of a GIS-Based Crash Referencing and Analysis System, Proc., Enhancing
Transportation Safety in the 21st Century, ITE International Conference.
2. Bicycle and Pedestrian Data: Sources, Needs, and Gaps. BTS00-02. (2000). U.S. Department of Transportation,
Bureau of Transportation and Statistics.
3. Levine, N., K. Kim, and L. Nitz. (1995). Spatial Analysis of Honolulu Motor Vehicle Crashes: I. Spatial Patterns,
Accident Analysis and Prevention, Vol. 27, No. 5, pp. 663-674.
4. Kim. K., D. Takeyama, and L. Nitz. (1994). Moped Safety in Honolulu Hawaii. Journal of Safety Research, Vol. 26,
No. 3, 1195, pp. 177-185.
5. Hanks Mohle and Associates. (1996). GIS for small Municipalities. Presentation Material. OTS Summit.
6. Pele. A., Hja-Yehia, and A.S. Hakkert. (1996). Arch Info-Based Geographical Information System for Road safety
Analysis and Improvement.
7. Pulugurtha, S.S., Krishnakumar, K.V., and Nambisan, S.S. (2007). New methods to identify and rank high
pedestrian crash zones: An illustration. Accident Analysis and Prevention 39, 800–811.
8. Chu. Y., M. Azer, F. Catalonotto, H. Ungar, and L. Goodnman. (1999). Safety/GIS Models reviewed and Related to
Long Island Arterial Needs Study. Proc., Enhancing Transportation Safety in the 21st Century. ITE International
Conference.
9. Miller. J. S. (2000).The Unique Analytical Capabilities Geographic Information Systems Can Offer the Traffic Safety
Community. Presented at the 79th Annual Meeting of the Transportation Research Board, Washington. D.C.
10. Braddock. M., G. Lapidus, E. Comley, R. Cromley, G. Burke, and L. Banco. (1994). Using a Geographic Information
System to Understand Child Pedestrian Injury. American Journal of Public Health, Vol. 84, No. 7, pp. 1158-1161.
11. McMahon. P. A. (1999). Quantitative and Qualitative Analysis of the Factors Contributing to Collisions between
Pedestrians and Vehicles along Roadway Segments. Master’s project. University of North Carolina at Chapel Hill.
12. Pedestrian and Bicycle Safety Analysis Tools. (2000). North Carolina Center for Geographic Information and Analysis
(NC CGIA).
13. Cameron, A.C. And Trivedi, P.K. Regression Analysis of Count Data. Cambridge University Press, 1998.
14. Ord, J. K. and Getis, A. Local Spatial Autocorrelation Statistics: Distributional Issues and an Application.
Geographical Analysis, Vol. 27, 1995, 286-306.
51. CONFERENCE PRESENTATIONS
Emaasit, D., Chimba, D., Cherry, C., Kutela, B., Wilson, J.
“Methodology to Identify Factors Associated with Pedestrian
High-Crash Clusters Using GIS-Based Local Spatial
Autocorrelation”. Accepted for presentation at the
Transportation Research Board 92nd Annual Meeting, (TRB),
Washington, DC, January 15th, 2013.
52. Emaasit, D., Chimba, D. “Methodology to Identify Factors Associated with
Pedestrian High-Crash Clusters Using GIS-Based Local Spatial
Autocorrelation”. Presented at the 35th Tennessee State University-Wide
Research Symposium, Nashville, April 4th, 2013.
CONFERENCE PRESENTATIONS