CE463 Lecture 27 Traffic Safety

1
CE463 Lecture 27Traffic Safety Crash Data
Analysis

• Civil Engineering
• Purdue University

2
Safety Management at Locations for States,
Counties, and Cities

• Identifying hazardous locations
• Evaluating safety at selected locations
• Identifying effective countermeasures
• Developing safety projects
• Implementing selected safety projects
• Evaluating safety improvements

3
Identification of Hazardous Road Segments
AADT in thousand vehicles/24 hours
Number of crashes during 5 years
Task Select the best candidate segment for
further in-depth safety analysis and potential
road improvement
4
Identification of Hazardous Locations

• Possible approaches
• Select the road segment with the highest number
  of crashes
• Select the road segment with the highest crash
  rate
• Select the road segment where some safety norm
  (typical number of crashes) is exceeded at most

5
Identification of Hazardous LocationsNumber of
Crashes
6
Identification of Hazardous LocationsCrash Rate
33,0001.13655/100,000,000 0.662 (100 mln
veh-mi)
7
Identification of Hazardous LocationsCrash Rate
19/0.381 49.8 acc/100 mln veh-mi
8
Identification of Hazardous LocationsCrash Rate
9
Identification of Hazardous Locations Excess
over Typical Number of Crashes
44.80.381 17.1 crash/5 years
Typical Crash Rate 126/2.814 44.8 acc/100 mln
veh-mi
10
Identification of Hazardous LocationsExcess over
Typical Number of Crashes
4-1.5 2.5 crash/ 5 years
11
Identification of Hazardous Locations Excess
over Typical Number of Crashes
12
Site AnalysisCollisions Diagram
1. Prevailing types of crashes 2. Drivers
actions leading to these crashes
13
Site AnalysisConditions Diagram
1. Prevailing types of crashes
2. Drivers actions leading to these crashes
3. Conditions leading to such actions
4. Changes reducing the likelihood of such
actions
14
Safety Management at Locations for States,
Counties, and Cities

• Identifying hazardous locations
• Evaluating safety at selected locations
• Identifying effective countermeasures
• Developing safety projects
• Implementing selected safety projects
• Evaluating safety improvements

15
Before-and-After AnalysisRandomness of Crash
Occurrence
No. of Crashes
96 97 98 99 00 01 02 03
04 05 06 07
Year of Improvement
16
Before-and-After AnalysisRandomness of Crash
Occurrence
No. of Crashes
96 97 98 99 00 01 02 03
04 05 06 07
17
Before-and-After AnalysisTest Conditions

• The before and after periods must be
  multiples of one year  
• For individual road facilities, the recommended
  length is 3 years before and 3 years after the
  safety improvement
•  
• The investigated location should not be affected
  by factors other than associated with the safety
  improvement

18
Before-and-After AnalysisTest Based on Normal
Distribution (approximation)

• Ab no. of accidents during n years before the
  treatment
• Aa no. of accidents n years after the treatment
• Crash counts are Poisson-distributed

1. Test for two normally-distributed variables
   (large counts)

• For the one-tailed test and significance level ,
  the critical value Z 1.65 for ? 0.05 (1.28
  for ? 0.10)
• The change is not random if Z gt Z

19
Before-and-After AnalysisNormal Distribution
Test Example
County X had 690 crashes involving trucks during
1988-90. The new policy of routing trucks in
this county became effective in January 1991.
The next three years (1991-93) brought 591
crashes with truck involvement. Is this
reduction significant?   Ab 640, Aa 591   Z
(640-591)/(640591)1/2 49/35.1 1.40   Z
1.65 (?0.05), Z 1.28 (?0.10)   The crash
reduction is not significant at the 0.05 level of
significance, but is significant at the 0.10
level of significance.
20
Before-and-After AnalysisPoisson and Chi-Square
Tests
21
Before-and-After AnalysisPoisson Test
During three years prior to the installation of
traffic signals, a given intersection was a scene
of 15 crashes. After the signals are installed,
the number of crashes dropped to 6 crashes during
the next three years. Is the safety improvement
significant?
43
22
Before-and-After AnalysisPoisson Test Example
  To be significant, the drop must be at least
43 according to the Poisson test (Figure 8-9 for
15 crashes).   The actual reduction is
Thus, the reduction is significant at the 0.08
level of significance.