a presentation on his work

A Framework to Predict High Risk
Roadways for Pedestrians in
Tennessee
Chris Cherry
Associate Professor-Civil and Env. Engineering
University of Tennessee
February 11, 2014
Collaborators: Zane Pannell (UTK), Deo Chimba and
Daniel Emaasit (TSU)
Brian Hurst and Jessica Wilson (TDOT)
Background
• “Statewide Pedestrian and Bicyclist Safety
Analysis and Safety Investment Policy” for
TDOT
– Cluster Analysis
– Hotspot Analysis
– Ranking of Hot Spot Zones
– Crash Frequency
– Injury Severity
– Total Relative Risk
Why?
• Bicycle and pedestrian crash fatality rate ≈
100 fatalities/per year
• Constant over past decade
• Goal: To create a framework for TDOT to
invest Highway Safety Improvement Program
funding towards bicycle and pedestrian
projects
• Other States: No robust methodology to
invest in pedestrian and bicycle safety
projects.
Data
• TRIMS crash data for pedestrian crashes in
Tennessee from 2003 to 2009
• TDOT geospatial data for road geometrics
• Demographic and Socioeconomic from 2010
US census
• Annual Average Daily Traffic (AADT) count
data from TDOT
• All joined together using linear referencing
(highway) and address match (local) and
spatial joining in ArcGIS
Pedestrian Crash Data in Tennessee
Total = 5,587 crashes
Years: 2003-2009
Pedestrian Crash Locations
Note: 1 dot = 1 crash
Statistical Modeling
• Crash Frequency
– Count Data Models
• Negative Binomial Regression Model
• Injury Severity
– Discrete Outcome Models
• Multinomial Logistic Regression Model
– With and without AADT
STATA statistical software was used.
Statistical Modeling
Crash Frequency
• Negative Binomial Regression Model
λi=
Where:
λ is excepted number of crashes per length of roadway (crash
frequency rate)
β is a vector of estimated coefficients
Xi is the value of the independent variable related to the
occurrence of crash
i is a potential outcome
•
•
168,920 roadways segments with an average of 0.5 miles per
roadway segment and 0.034 crashes per roadway segment
Estimated with 90% level of significance level (P-value=0.10)
Statistical Modeling
Injury Severity
• Multinomial Logistics
Regression Model
Where:
β is a vector of estimated coefficients
Xi is the vector of independent variables
i is a potential outcome
Estimated with 90% significance level
Total Relative Risk Data
• All census data and attributes
• All roadway geometrics and attributes
• 2010 AADT Data (excludes local roadways)
Total Roadway Segments
= 193,574
Total Relative Risk
• Combined
– Crash Frequency Model
– Injury Severity Models
• With AADT
• Without AADT
– 2010 Highway Safety Manual Crash Cost
Guidelines
• Final Outcome: 2 total relative risk
models in GIS and excel
Crash Severity Rate
Total Risk=
E(λ) pdo & non-incapacitating injury * PPDO & non-incapacitating injury * CF
+ E(λ) incapacitating injury * Pincapacitating injury * CF
+ E(λ) fatal * P fatal* CF
Where;
•
•
•
E(λ) = Crash Severity Rate
P = Probability of injury type
CF = Crash Frequency Rate
Collision Type
Fatal Injury
Incapacitating
Injury
NonIncapacitating
Injury
PDO
2012
Comprehensi
ve Crash
Costs
$5,392,545
Number of
Crashes
443
Weight
Factor
5,392,545
2012
Comprehensive
Crash Costs
Ratio
98.48
$287,008
1,083
287,008
5.24
$59,241
$9,626
3,694
367
54,757
1.00
Total Relative Risk Results without AADT
• 10 Classifications for greater variation
Summary of Results
Total relative risk categories grouped
Attributes
High (2.147267-10.286418)
(601 roadway segments)
68% of road segments are in Shelby County; 2-8
lanes; speed limit 30-55 mph; CBD, commercial,
fringe, industrial, residential or public land use
Medium (0.445827-2.147266)
(9,958 roadway segments)
State routes; 1-8 lanes (bi-directional); speed
limit 30-65 mph; CBD, commercial, fringe,
industrial, residential or public land use
Low (0.032703-2.558273)
(59,719 roadway segments)
State routes and local roadways; 1-7 lanes (bidirectional); speed limit 5-70 mph; all land uses
(including rural land use) roadways on census
tracts with low population density; school zones
Very Low (0.000000-0.032702)
(123,296 roadway segments)
Freeways, state routes, and local roadways; all
speed limits; all land uses (includes rural land
use)
Total Relative Risk Results without
GIS Tool
State Route 177
GIS Tool
Conclusion
• Final Deliverable: Framework for TDOT to develop a
methodological approach to target roadways that are
at risk for incapacitating injuries and fatalities for
pedestrians
• Roadways with 6 or more lanes (minor arterials); high
speed limits ranging from 40 to 55 mph; CBD,
commercial, fringe, or industrial land use are most at
risk for pedestrians
• Shelby County is the most at risk county for
pedestrians (highest crash count, highest population,
largest area)
Thanks
Christopher Cherry
Associate Professor
Civil and Environmental Engineering
University of Tennessee-Knoxville
321 JD Tickle Bldg.
Knoxville, TN 37996-2313
phone: 865-974-7710
mobile: 865-684-8106
email: [email protected]
http://web.utk.edu/~cherry
http://www.cycleushare.com
http://TRBemergingtech.com
Crash Frequency
Variable
Coef.
Std.
Error
PZ-Stat. Value
Number of Lanes (continuous)
0.585
0.024
24.51
0.000
Speed Limit: 30 to 35mph (1/0)
2.063
0.062
33.14
0.000
Speed Limit: 40 to 55mph (1/0)
2.052
0.070
29.48
0.000
Presence of a School Zone (1/0)
0.365
0.103
3.53
0.000
Central Business District (CBD), Commercial,
Fringe & Industrial Land Use (1/0)
2.057
0.076
26.99
0.000
Residential, Public Use & Parks Land Use (1/0)
1.368
0.077
17.86
0.000
White Population (0-100)
-2.6E-04
0.006
-0.05
0.963
Black Population (0-100)
0.010
0.006
1.86
0.062
Hispanic Population (0-100)
0.023
0.006
4.13
0.000
Crash Frequency Cont…
Variable
Population from 20 to 64years of age (0-100)
Mean Household Income (continuous)
Households with Income below $25000 (0-100)
Population in Labor Force (0-100)
Population below Poverty Level (0-100)
Housing units with 0 vehicles (0-100)
Housing units with 1 vehicle (0-100)
Housing units with 2 or more vehicles (0-100)
Constant
Segment Length
Number of observation = 168,920
P-value = 0.0000
Pseudo R2 = 0.279
Coef.
0.014
5.3E-06
-0.025
0.006
0.024
0.015
0.014
-0.029
-8.620
Offset
Std.
Error
0.004
1.4E-06
0.005
0.003
0.005
0.006
0.004
0.004
0.435
ZStat.
3.67
3.81
-5.52
2.09
5.10
2.60
3.14
-6.54
-19.82
PValue
0.000
0.000
0.000
0.037
0.000
0.009
0.002
0.000
0.000
Injury Severity with AADT
Std. Err.
ZStat.
PValue
-0.094
0.087132
-0.15547
0.040755
0.292167
0.135917
0.230338
0.116805
-0.32
0.64
-0.67
0.35
0.748
0.521
0.500
0.727
Speed Limit 65+ mph (1/0)
3 or More Vehicles Available per Household
(0-100)
Household Income and Benefits $150,000+
(0-100)
Number of Lanes (continuous)
0.495186
0.26474
1.87
0.061
2.174999
0.583621
3.73
0.000
-1.77822
0.09185
0.798182
0.03934
-2.23
2.33
0.026
0.020
White Population Density (continuous)
Black Population Density (continuous)
-5.8E-05
-0.00015
0.000038
4.37E-05
-1.53
-3.33
0.126
0.001
Constant
-1.65926
0.186235
-8.91
0.000
Variable
Coef.
Property Damage & Non-incapacitating (base
Injury
outcome)
Incapacitating Injury
AADT 60,000+ (1/0)
Rural Land Use (1/0)
School Zone Present (1/0)
Speed Limit 35-60 mph (1/0)
Injury Severity with AADT Cont…
Variable
Fatal
AADT 60,000+ (1/0)
Rural Land Use (1/0)
School Zone Present (1/0)
Speed Limit 35-60 mph (1/0)
Speed Limit 65+ mph (1/0)
3 or More Vehicles Available per Household
(0-100)
Household Income and Benefits $150,000+
(0-100)
Number of Lanes (continuous)
White Population Density (continuous)
Black Population Density (continuous)
Constant
Number of observation = 3261
P-value = 0.0000
Pseudo R2 = 0.05
Coef.
ZStd. Err. Stat.
PValue
0.712629
0.522875
-1.69982
0.413444
1.078388
0.320928
0.180172
0.720589
0.185134
0.31694
2.22
2.90
-2.36
2.23
3.40
0.026
0.004
0.018
0.026
0.001
3.495465
0.829313
4.21
0.000
-4.93152
0.137777
-0.00016
-0.00023
-3.08353
1.363292
0.057331
0.000071
0.00007
0.292611
-3.62
2.40
-2.25
-3.33
-10.54
0.000
0.016
0.024
0.001
0.000
Injury Severity without AADT
Variable
Property Damage & Non-incapacitating
Injury
Incapacitating Injury
Rural Land Use(1/0)
Industrial Land Use (1/0)
Speed Limit 35-60 mph (1/0)
Speed Limit 65+ mph (1/0)
0 Vehicles Available per Household (0-100)
1 Vehicle Available per Household (0-100)
2 Vehicles Available per Household (0-100)
Household Income and Benefits $0-$34,999 (0100)
Household Income and Benefits $35,000$149,999 (0-100)
Total Population Density (continuous)
Number of Lanes (continuous)
Constant
Coef.
(base
outcome)
Std. Err.
ZStat.
PValue
0.072736
-0.86804
-0.10341
0.022785
0.358384
-3.09649
-2.76908
0.107522
0.614569
0.179693
0.088147
0.223774
0.549141
0.445698
0.68
-1.41
-0.58
0.26
1.60
-5.64
-6.21
0.499
0.158
0.565
0.796
0.109
0.000
0.000
-1.25276
0.634086
-1.98
0.048
2.705852
1.592963
-0.00028
0.08859
0.453616
0.504684
0.000081
0.029947
5.97
3.16
-3.40
2.96
0.000
0.002
0.001
0.003
Injury Severity without AADT Cont…
Variable
Fatal
Coef.
ZStd. Err. Stat.
Rural Land Use(1/0)
0.588394
0.151573
3.88
0.000
Industrial Land Use (1/0)
1.062483
0.441109
2.41
0.016
Speed Limit 35-60 mph (1/0)
-1.14987
0.460343
-2.5
0.012
Speed Limit 65+ mph (1/0)
0.380659
0.142586
2.67
0.008
0 Vehicles Available per Household (0-100)
1.120074
0.266321
4.21
0.000
1 Vehicle Available per Household (0-100)
-3.46944
0.845476
-4.1
0.000
2 Vehicles Available per Household (0-100)
Household Income and Benefits $0-$34,999
(0-100)
Household Income and Benefits $35,000$149,999 (0-100)
Total Population Density (continuous)
-4.29873
0.697132
-6.17
0.000
-1.88967
0.952681
-1.98
0.047
4.477933
3.206062
0.737238
0.79738
6.07
4.02
0.000
0.000
Number of Lanes (continuous)
-0.00036
0.000136
-2.65
0.008
Constant
0.187014
0.043403
4.31
0.000
Number of observation = 5569
Pseudo R2 = 0.0466
P-value = 0.0000
PValue