Knowledge and Data
in Road Safety Management
- Research at the Center for Road Safety
International Seminar on Road Safety Research
La Universidad de los Andes
November 19, 2012, Bogota, Colombia
by
Andrew Tarko
Professor of Civil Engineering
Director of Center for Road Safety
School of Civil Engineering
Purdue University
West Lafayette, Indiana, USA
tarko@purdue.edu
Can Road Safety be Managed?
• It can if:
– The current safety is known
– The consequences of potential actions can be
predicted
– The relevant countermeasures exist and can be
identified
– The safety can be monitored to verify the
predictions and the correctness of actions
Measuring Safety
• Bad news – safety is difficult to measure
• The biggest challenge is to measure safety here
and now (probability of crash and expected
severity of its outcome)
• It is much easier to measure aggregate safety
afterwards:
– in long periods
– in large areas
Acquiring Knowledge
• Statistical analysis of observational data
(causal relationship may be questioned)
• Experiments with postulated relevance to
safety and without exposure context
• Good news – modern surveillance, naturalistic
driving (still observational)
Safety Factors Contribution
(Great Britain and USA)
6
Human factors (93%)
57%
27%
3%
Roadway
factors (34%)
6% 3%
1%
2%
Vehicle factors (12%)
Source: http://en.wikipedia.org/wiki/Traffic_collision
What Knowledge Is Transferrable?
• Qualitative knowledge explaining safety-related
behavior and performance
• General classification of which countermeasures
work based on principles of physics,
biomechanics, etc.
• Methods and processes for safety management
based on mathematics, statistics, and
Example - Speed Selection
Example - Speed Selection
Speed selected = 130 km/h if speed limit is ignored
Example Transferrable Knowledge
• Perceived risk grows slower than actual risk
• Enforcing speed limits is needed
• Explains aggressiveness of young drivers (low
risk perception)
• Explains ignoring unreasonably low speed
limits (too high cost of complying)
• Consistent with observed fast driving under
time pressure (high value of time)
Another Example - Power Model
Crash Modification Factor =
(Speed after/Speed before)
Exponent
Power model exponent was believed to be
universal – only dependant on the severity of
crash outcome
Another Example - Power Model
Exponents
Freeway Crash Severity
States comparison for freeway crashes – many conditions controlled
State
Injury Odds Ratio
Run-off-road
Crash
Multiple-vehicle
Same-dir Crash
IN
1
1
OH
0.79
1
IL
1.28
1.16
MO
1.97
1
WA
3.29
2.56
NY
2.76
4.47
OR
3.52
2.59
CO
2.15
2.27
What Knowledge May not be
Transferrable?
• Safety Performance Functions
• Crash Modification Factors
• Cost of crashes
• This knowledge is also subject to aging
Two Uses of Data
1. Research: To acquire knowledge and to
develop methods useful for safety
management, and
2. Management: To identify safety problems
and to propose most effective
Safety Data Needs
Minimum Data Requirement
• Crash reports
• Basic infrastructure inventory
• Traffic volumes
Crash Reports
• Primary source of safety information
• Determines the quality of safety management
• Vary between countries and jurisdictions
• Underreporting
• Most important data are most troublesome:
– Crash location
(Indiana: 1995 – linear reference, 45-50% known;
2012 – GIS coordinates, 80% known; future – “point and click”)
– Injury severity
(Indiana: fatalities – 30 day update, good quality; other
outcomes determined by investigating police officers)
Crash Reports
• Address
• Linear reference
• GPS receiver
• Point & Click
Crash location
(Where)
GIS
Geo-code
Crash Data Geo-coding
Geocode
Crash
database
Crash Data – Electronic Report and
Point and Click Technique
Crash
database
Crash Reports –
Injury Data
Crash
Injury Severity
K
Fatal
A
Incapacitating
injury
B
Non-incap.
injury
C
Possible
injury
O
Property
damage
23Injury Data
Police-based vs. Hospital-based
Linked Police-hospital Data
(Indiana CODES Project)
• Better measurement of injury level
• Hospital data available only for more severe cases
• Selection biased addressed via bivariate model of
outcome and selection
Road Infrastructure Data
• Road classification
– Administration
– Functional
• Cross-section information
• GIS representation
– Segments
– Intersections
– Bridges
– Ramps
– Interchanges
26Road Representation
Segments - splitting
27
Traffic Data
• Critical exposure information
• Annual Average Daily Traffic (AADT)
• Typically, available only for major roads
• Mitigation for local roads: use proxy
exposure such as land use, proximity of
arterial roads, etc.
CRS Road Safety Database
Data
Details
Source
Road network 440,000 state and local roads segments, 200,000 intersections Indiana Department
of Transportation
Road data
Cross-section data for state and major local roads
AADT
State road segments
Crash data
24 years, 8 years geo-coded, 300,000 records/year
Indiana State Police
Hospital data 2003-2010, 4.5 million hospital discharge records/year,
between 40,000 and 63,000 annual crash-hospital links
Indiana Hospital
Association
Death data
2003-2008, 55,000 deaths/year, death certificates data
Use of seat
belts
113 sites, 3 times/year, 2001-2012
New survey 190 sites twice/year 2013 onwards
Indiana Criminal
Justice Institute
Driver data
Registered motorcycle owners and all drivers born after 1978,
Citations, suspensions, registration information
Indiana Bureau of
Motor Vehicles
Census data
2000, 2010 years, demographic, socioeconomic, economic, and
other statistical data. Linear features such as roads, railroads,
rivers, and legal boundaries (TIGER).
US Census Bureau
Bridge data
System 1 Bridges – Bridges found on System 1 roads that
include interstate highways, U.S. highways, state routes, ramps.
INDOT, National
Bridge Inventory
Weather data US National Weather Service shape files. Weather stations in
Indiana + weather historical data
USNWS,
National Climatic
Data Center
Households
2 million households as of 2010, basic socio-demographics
CAS Inc. (commercial
source)
Safety Database Renewal
Source Data 1
Source Data 2
Source Data n
Acquiring
New Data
Reformatting
Source Data
Reformatted
Source Data 1
Reformatted
Source Data 2
Reformatted
Source Data n
Converting
Data
Processed
Data A
Processed
Data X
Linking Data
Linked
Database
Approaches to Safety Management
32Normative
Objective
Analyze
Safety Data
Comply
to Standards
Subjective
Follow
Expert Judgment
32Data-driven Management Cycle
33
Analysis
Action
Data
Indiana Safety Management
• Strategic Highway Safety Plan (INDOT, ICJI)
• Hazard Elimination Program
– Road screening for hazards
– Site investigation of high-crash roads
– Economic evaluation and selection of safety interventions
• Targeted Safety Programs
– Road screening for certain deficiencies
– Implementation of safety interventions
Safety Needs Identification Program -
SNIP
• A comprehensive safety evaluation tool to address
statewide safety investment needs
• Identify safety needs based on an excessive number
of crashes in user-defined categories
• Develop a method of screening the road network for
road deficiencies
Roads with Speeding Problem
SNIP2
(1) Improved
interface
(2) Improved
features
(3) Matching
identified
safety needs
with safety
programs
within the
budget
Road Hazard Analysis Tool
RoadHAT
The RoadHAT is a computer implementation of the Guidelines for
Highway Safety Improvements in Indiana (Tarko and Kanodia,
2004) for analyzing high-crash roads.
RoadHAT2 Forms
RoadHAT2
• The user can defined its own:
– Types of roads
– Safety Performance Functions
Transferrable
General Safety Measures
• Managing exposure to risk through transport and
land-use policies
• Designing roads for road injury prevention
• Providing visible, crash-protective, “smart”
vehicles
• Securing compliance with safety rules and
promoting safe behavior
• Delivering post-crash care
Do All Countermeasures Work
Everywhere?
According to various USA studies:
Widening paved shoulders from 3 ft to 8 ft reduces
crashes by 12 percent
Importance of Local Conditions
Transferrable Countermeasures?
CMF
30→10
=2.2/4.0 = 0.55
(Brown and Tarko, 1999)
(Li and Tarko, 2011)
CMF
Rear-End
= 0.06
Safety Measurement Revisited
Exceedance as a Crash Surrogate
Safety Measurement Revisited
CLOSURE
•
A healthy balance of transferrable knowledge and own
components should be applied to safety management
•
Good data support research (developing and updating
safety models) and effective management
•
Developing of a good database takes time and should
begin as soon as possible
•
Data collection and sharing was a starting point of
communication between Indiana organizations
•
There is still a long road to good safety management
even in countries who started earlier than others
•
Safety measurement methodology still requires
Selected Sources
Villwock, N. M., N. P. Blond, and A. P. Tarko. Risk Assessment of Various Median Treatments of Rural Interstates. Publication
FHWA/IN/JTRP-2006/29. Joint Transportation Research Program, Indiana Department of Transportation and Purdue University, West Lafayette, Indiana, 2008.
Elvik, R. The Power Model of the relationship between speed and road safety. Institute of Transport Economics, Norwegian Centre of Transport Research, TOI report 1034/2009, 2009.
Tarko, A.P. Modeling drivers’ speed selection as a trade-off behavior, Accident Analysis & Prevention, Volume 41, Issue 3, May 2009, Pages 608-616.
Persaud, B., D. Lord, and J. Palmisano. Calibration and Transferability of Accident Prediction Models for Urban Intersections,
Transportation Research Record 1784, 2003, pages 57-64.
Sacchi, E., M. Bassani, B. Persaud. Comparison of Safety Performance Models for Urban Roundabouts in Italy and Other Countries. Transportation Research Record: Journal of the Transportation Research Board, No. 2265, Transportation Research Board of the National Academies, Washington, D.C., 2011, pp. 253–259.
Sacchi , E. and M. Bassani. and B. Persaud. Assessing International Transferability of Highway Safety Manual Crash Prediction Algorithm and Its Components. Transportation Research Record: Journal of the Transportation Research Board, No. 2279, Transportation Research Board of the National Academies, Washington, D.C., 2012, pp. 90–98.
Tarko, A.P. and A. Issariyanukula, H. Bar-Gera, Model-Based Application of Abbreviated Injury Scale to Police-Reported Crash Injuries.
Transportation Research Record: Journal of the Transportation Research Board, No. 2148, Transportation Research Board of the
National Academies, Washington, D.C., 2010, pp. 59–68.
Tarko, A. and Md.S. Azam. Pedestrian injury analysis with consideration of the selectivity bias in linked police-hospital data. Accident
Analysis and Prevention 43 (2011), pages 1689–1695.
Tarko, A, and M. Kanodia. Effective and Fair Identification of Hazardous Locations. Transportation Research Record: Journal of the
Transportation Research Board, No. 1897, TRB, National Research Council, Washington, D.C., 2004, pp. 64–70.
Li, W. And A. Tarko. Effect of Arterial Signal Coordination on Safety. Transportation Research Record: Journal of the Transportation
Research Board, No. 2237, Transportation Research Board of the National Academies, Washington, D.C., 2011, pp. 51–59.
Tarko, A. Use of crash surrogates and exceedance statistics to estimate road safety. Accident Analysis and Prevention 45, 2012, pages 230– 240.
Tarko, A. and S. Azam. Safety Screening of Road Networks with Limited Exposure Data. Transportation Research Record: Journal of the
Transportation Research Board, No. 2102, Transportation Research Board of the National Academies, Washington, D.C., 2009, pp.
18–26.
Brown, H. and A. Tarko. Effects of Access Control on Safety on Urban Arterial Streets. Transportation Research Record: Journal of the
Transportation Research Board, No. 1665, Transportation Research Board of the National Academies, Washington, D.C., 1999, pp.