TRANSPORTATION
IDENTIFICATION OF FACTORS IN ROAD ACCIDENTS
THROUGH IN-DEPTH ACCIDENT ANALYSIS
Mouyid BIN ISLAM
Kunnawee KANITPONG
Research Associate, Thailand Accident Research Center
Asian Institute of Technology
Pathumthani, Thailand
Assistant Professor, Transportation Engineering Program
Manager, Thailand Accident Research Center
Asian Institute of Technology
Pathumthani, Thailand
(Received February 4, 2008)
The rising trend of motorization and improving socio-economic status of Thai people directly influences the aggravating road
safety situation with fatalities and permanently disabled injuries of about 130,000 and 500,000 respectively over the past decades. An
estimated annual cost from road crashes amounts to about US$2,500 million, 3.4 percent of Gross National Product (GNP), undoubtedly inflicts Thailand with a burning public health concern in the South East Asian region. This paper addresses an in-depth study
through crash investigation and reconstruction which has not yet been practised in Thailand to identify the contributory factors in road
crashes by the concerned authorities. This research attempts to establish the linkage between the causes and consequences with
event classification of an investigated case by highlighting the dynamic driving situation with initial traveling speed, pre-impact and
post-impact speed of the involved vehicles to describe the crash scenario. Moreover, inaccurate risk assessment and late evasive
action, absence of street-light facilities, inadequate lane marking and visibility were also outlined as major risk factors increasing the
severity of crash and injury in this investigated case.
Key Words: Investigation, Reconstruction, In-depth analysis, Event tree, Factors
58
IATSS RESEARCH Vol.32 No.2, 2008
1.00
50%
0.90
45%
0.80
Injuries per accident
40%
0.70
35%
0.60
30%
0.50
25%
20%
0.40
0.30
Fatality Index
0.00
15%
10%
0.20
0.10
Fatality Index
Road safety becomes a major public health concern
when the statistics show that more than 3,000 people
around the world succumb to death daily due to road traffic injury1. In addition, road crashes lead to the global
economic losses as estimated in road traffic injury costs
of US$518 billion per year 2. The huge economic losses
are an economic burden for developing countries. It is reflected that the road crash costs are estimated to be US$100
billion in developing countries which is twice the annual
amount of development aid to such countries 2. Considering within South East Asian countries, the economic
growth rate of Thailand continues to move upward with
an aggravating road traffic situation due to the heavy negative impact of a higher level of motorization. Over
130,000 fatalities and nearly 500,000 people were permanently disabled due to road crashes over past decades 3.
The economic losses due to the road crashes are; therefore, considerably high, costing approximately US$2,500
million per year (about US$0.3 million per hour), or 3.4
percent of the Gross National Product (GNP).
An Asian Development Bank country report 3 focused on the seriousness of the road accident problem
which is shown in Figure 1 with an upward trend of inju-
ries per accident whereas fatalities per accident remained
constant with small fluctuations from 1993-2002. However, the fatality index declined to 16 percent in 2002
from 27 percent in 1993 over this period of time.
The collection and use of accurate and comprehensive data related to road accidents is very important to
road safety management 4. The road accident data are
necessary not only for statistical analysis in setting priority targets but also for in-depth study in identifying the
contributory factors to have a better understanding of the
chain-of-events. Having the inconsistencies in the aims
of the police and the road safety engineers, the data analFatalities or Injuries per Accident
1. INTRODUCTION
Fatality per accident
1993 1994 1995 1996 1997 1998 1999 2000 2001 2002
5%
0%
Fig. 1 Trends in casualties per accident and fatality
index (Tanaboriboon 2004)
IDENTIFICATION OF FACTORS IN ROAD ACCIDENTS THROUGH IN-DEPTH ACCIDENT ANALYSIS
ysis and its interpretation usually does not result in proper countermeasures. Sometimes a lack of proper knowledge
of crash and proper training of the police officers in
charge on systematic data collection procedures from a
crash scene adds to the diverging nature of the role of the
police and the road safety professionals. These problems
have become a burning issue for developing countries
addressing road safety without completed crash data due
to the negligence of the concerned authorities. A study 5
clearly indicates this limitation - “the reactions are mostly on major accidents, but the interests would fade away
rapidly and the problem still remains”.
2. BACKGROUND
The identification of factors affecting road crashes
obtained from the crash investigation and reconstruction
has not been conducted in practice in the Asian countries.
The goal of this study was to initiate this road safety
practice in Thailand by addressing the timely need for an
in-depth study for road accidents.
The accident investigation involves the inspection
of crash scenes and the documentation of all necessary
and available information of each component (i.e. human, vehicle, and road-environment). Accident reconstruction is defined by Baker and Fricke 6 as “…the efforts
to determine from whatever information is available, how
the accident occurred”. Accident reconstruction approach
works backward from the evidence of the crash investigation and the remains of the crash to look into the scenario
of before (pre crash), during (crash) and after the crash
(post crash). The sequential analysis of end results to the
initial condition of the events can establish “how” and
“why” a particular type of crash occurs. Mathematics and
Newtonian physics are applied in this analysis. It can be
stated that crash reconstruction goes back to investigate
the contributory factors and/or causes behind the crash
event based on major and minor physical clues left behind at the crash scene.
The techniques of crash reconstruction, trajectory
and damage based analysis by using physics simplifies
the determination of many important parameters of crash
events. Moreover, to obtain a reliable conclusion, detailed
information encompassing the system components needs
to be thoroughly investigated. The information necessary
for reconstruction starts with the crash scene 7. The answers to the questions of ‘why’, ‘what’, ‘when’ and ‘how’
should lead the reconstruction process to build up the real
scenario of the pre-crash, crash, and post crash 7. Photographing of important clues and videotaping of the crash
M. BIN ISLAM, K. KANITPONG
scene plays a vital role for the reconstruction. Injury information from occupant medical reports can be verified
with the trajectory of the occupants found inside the involved vehicles at the scene. Therefore, an “open mind”
investigative attitude is very crucial to search for all the
detailed information from the scene 7.
3. OBJECTIVES
The purpose of this study was to conduct an indepth study focusing on the application of event analysis
through crash investigation and reconstruction. The objectives of this study were the followings:
1. To identify the contributory factors based on the findings obtained from crash investigation and reconstruction by using a case study;
2. To apply an event analysis in establishing the links between the events to describe the crash scenario based
on the available information.
This case was selected to conduct an in-depth analysis of the crash investigation and reconstruction because
of the following reasons:
1. A fatal case between a bus and a pickup truck vehicle
to understand the crash mechanism of two structurally
different (incompatible) vehicles and crash and injury
severity of the involved vehicles.
2. Good example for the evasive actions taken by the
driver (i.e. bus driver) in the form of the skid marks on
the road surface to calculate back the traveling speed
of the bus.
3. The total number of buses and pickup truck vehicles
combined involved in all crashes increased from
34,650 in 2003 to 36,816 in 2005 8. But the severity of
injury from an angled head-on collision like this case
is very challenging to investigate the leading factors
to prevent such crashes.
4. METHODOLOGY
The in-depth analysis for the contributory factors
through event classification requires a scientific methodology to follow systematic research. The conceptual framework designed for this study is presented in Figure 2.
The investigation team was always on the alert for
crash news. Independent News Network (INN) was considered a primary source of road accidents where the
team was motivated to investigate the case. It should be
noted that the safety precautions of the investigation team
were maintained while conducting the investigation proIATSS RESEARCH Vol.32 No.2, 2008
59
Crash
Notification
Crash News Report
Crash Data
Collection
TRANSPORTATION
Crash Investigation
resent possible ways for a dysfunctional behavior to manifest itself in the dimensions of time, space and energy
whereas the causes are interpreted from the reasoning 9.
To apply DREAM analysis in this study, the possible
connections between factors behind the events were established which attempts to explain the observed consequences or the event phenotype9. Figure 2 shows the steps
followed in this study to figure out the complete analysis.
5. ACCIDENT IN-DEPTH ANALYSIS
- A CASE STUDY
Crash
Analysis
Crash Reconstruction
5.1
Results and
Findings
Development of Event Tree
Contributory Factors
Fig. 2 Conceptual framework for the study
cess. After the arrival at the crash location, the necessary
information was collected with field sketches, drawings,
and photographs of the crash scene and damaged vehicles
from different angles. The physical evidence (e.g. tire
masks, broken glass, oil and blood spatter, etc) at the crash
scene were carefully collected particularly for the skid
marks that were measured according to direction and coordinates from a reference point. The roadside infrastructure was referenced and taken into consideration. In
addition, the concerned police station was contacted and
their reports were gathered. Damaged vehicles taken to
the police station were also carefully investigated.
The evidence at the crash scene, interviews of eyewitnesses in the vicinity of the crash location, and the
police reports were gathered and interpreted to visualize
the events prior to the crash according to the available
information. The trajectory based reconstruction was carried out to determine the traveling, pre-impact and postimpact speed including direction of force acting on the
involved vehicles.
Based on the concept of Driving Reliability and
Error Analysis Method (DREAM), the genotypes (i.e.
cause) and phenotypes (i.e. consequences) of the case
study were segregated. The inseparable driver behavior
from the context was the main focus which encompassed
the effects and the causes of the effects. The effects rep-
60
IATSS RESEARCH Vol.32 No.2, 2008
General information of the case
The crash occurred between a bus and a pickup truck
on an undivided 2-lane 2-way straight section of Chiang
Rak-Bang sai Arts and Craft Center rural highway at about
9 p.m. A total of nine casualties, eight fatalities and one
serious injury, were reported in this angled head-on crash
between two vehicles. All of the causalities were the occupants of the pickup truck. The pickup truck truck was
severely damaged with little deformation in the bus.
5.2 Pre-crash information
Driver Information
Pickup truck: The pickup truck driver was 37 yearold male. Nine passengers were inside the pickup truck.
A seatbelt was used by the driver according to the evidence
from the investigation process. He was traveling at about
55 kph outbound to Bangsai.
Bus: The bus driver was male (age not reported).
He was trying to pass the motorcycle in front of him. He
was traveling at about 65 kph inbound to Bangsai.
Vehicle Information
Pickup truck: The 4-wheel pickup truck was locally
modified to provide a roof and seats for the passengers in
the back. The seating rows were arranged along both sides
of the vehicle for the convenience of passengers. The pickup truck was used for public transport. The body was silver color.
Bus: The body structure of the 6-wheel bus was locally modified. It was white and blue painted.
Road-Environment Information
Geometry: Chiang Rak-Bang sai Arts and Crafts
Center Road is a 2-lane 2-way road in a rural setting. The
lane width is 2.7 meters in each direction with 1.8 meter
wide shoulders in both directions. The road was level.
Surface: The pavement surface was asphalt, and the
surface condition was dry during the investigation.
Lane markings: The yellow marking (dashed) was
a lane separating the two lanes, and the white marking
IDENTIFICATION OF FACTORS IN ROAD ACCIDENTS THROUGH IN-DEPTH ACCIDENT ANALYSIS
(solid) was a lane-shoulder separation in both directions.
The lane markings were not clearly visible considering the
road surface as a background.
Roadside furniture and area: The small trees and
electricity transmission poles were found along the straight
section of the road. In addition, election campaign boards,
information boards, and some traffic signs were observed
along the road near the crash scene. There was a sheltered
‘Bus Stop’ on the opposite side of an election campaign
board. No street light was found on the road. Two minor
connecting roads (i.e. access road) were also observed on
the opposite side of the road section at the crash scene. A
cut-section was on both sides of the embankment of the
road section.
5.3 In-crash information
Driver Information
Pickup truck: As the pickup truck was traveling in
its own lane, there was no sudden expectancy to slow
down. Suddenly, the bus appeared into the right of way
(lane) of the pickup truck. Due to ‘sudden and unexpected’ situation, the pickup truck did not have time to brake
and avoid the collision (no skid marks by the pickup truck
were found on the driving lane of the road). The pre-impact speed of the pickup truck was about 55 kph. The
direction of force (PDOF) passing through the centriod
of the damaged portion was 11.5 degree (clockwise) with
respect to the longitudinal axis of the vehicle. Figure 3
shows the driving direction of the pickup truck.
Bus: Consequently, the bus was going to the right
lane from its driving lane (left lane). Suddenly, the bus
D
E
C
B
A
Note: A: Initial Direction of Travel (pickup truck); B: Rear Right and
Left Tire at Point of Rest (pickup truck); C: Blood; D: Debris and
Oil Spatter and E: Initial Direction of Travel (bus).
Fig. 3 Driving path of pickup truck (southeast bound)
M. BIN ISLAM, K. KANITPONG
J
I
H
F
G
Note: F: Initial Direction of Travel (bus); G: Right Front Tire Mark (bus); H:
Rear Right Tire Mark (bus); I: Rear Right and Left Tire Mark at
Point of Rest (bus) and J: Initial Direction of Travel (pick-up).
Fig. 4 Driving path of bus (northwest bound)
driver found that the pickup truck was approaching at a
very close distance. Therefore, the bus driver applied the
brakes in 0.63 seconds and made 9.8 meters of skid marks
before the crash. However, the bus could not avoid the
collision with the pickup truck. The bus was traveling at
about 65 kph and slowed down to 47 kph within a very
short time (i.e. 0.63 seconds). Figure 4 shows the driving
direction of the bus.
5.4 Post-crash information
Driver Information
Pickup truck: The post-impact speed of the pickup
truck was estimated to be about 37 kph. Due to severe intrusion and damage of the passenger compartment there
were eight fatalities and one serious injury of the occupants.
Bus: The post-impact speed of the bus was estimated to be about 24 kph. Due to the geometry and mass incompatibility between the vehicles, the bus sustained
minor damage compared to the pickup truck.
Vehicle Information
Pickup truck: Since pickup truck collided with the
bus, the damage was very severe. The impact force of the
bus was high enough to crush the major portion of crush
zone and passenger compartment of the pickup truck. The
measurements were not directly made due to the crash severity of the pickup truck. The external damage with internal intrusion of the damage resulted in eight fatalities
out of nine passengers inside the pickup truck. The calculated delta-V for the pickup truck was about 89 kph. Figure 5 (a) shows the extent of damage of pickup truck from
different angles.
IATSS RESEARCH Vol.32 No.2, 2008
61
TRANSPORTATION
Bus: The bus was higher than the pickup truck in
terms of vehicle geometry and only the right side of the
bus was damaged at the bumper and the right front fender.
The direction of force (PDOF) passing through the centriod of the damaged portion was 8.4 degree (clockwise) with
respect to the longitudinal axis of the vehicle. The extent
of damage of bus was comparatively less. The calculated
delta-V for the bus was about 24 kph. Figure 5 (b) shows
the extent of damage of the bus from different angles.
Road-Environment Information
Pickup truck: The pickup truck was stopped after
going backward on the left side for 6.8 meters from the
point of impact. The rest position of the pickup truck was
found at the crash scene on its driving lane but close to
the shoulder.
Bus: The bus stopped after crossing 2.8 meters
ahead from the point of impact on the opposing lane (right
lane which is right-of-way of pick-up).
Figure 6 shows a schematic drawing of the crash
scene when the vehicles were at their rest positions.
5.5
Reconstruction by simulation
The trajectory of the crash involved vehicles was
determined by using hand calculations. The results were
obtained and used as input in the reconstruction simulation software, PC-Crash, to simulate different events to
demonstrate the consequences leading the crash. In the
simulation package, 3D features were applied in the reconstruction process. The 100 meters-straight section of
road following the curve was set as an input for the crash
scene. Trees, electric poles, information sign boards, advertisement boards, two abutting roads, and the small
sheltered ‘Bus Stop’ for the passengers were included as
inputs for the crash scene data. Length of broken lines
and gap between them were set as 3 meters and 6 meters,
respectively. The plan view is illustrated in Figure 7.
a)
1.8m
2 Lans @2.7m
1.8m
20m
N
b)
10m
Level Grade
Cross Slope
5 percent
Reference Point
Electric Pole
B1A
1F-07
10K
11.4m
Fig. 5 a) Damaged pickup truck (left: front) and (right: right-corner), and
b) Damaged bus (left: front) and (right: right-corner)
Fig. 6 Schematic drawing of crash
scene
A
D
C
B
Note: A: Initial position of bus, B: Initial position of pickup truck, C: Point of rest, D: Initial position of motorcycle.
Fig. 7 Plan view of crash scene input
62
IATSS RESEARCH Vol.32 No.2, 2008
IDENTIFICATION OF FACTORS IN ROAD ACCIDENTS THROUGH IN-DEPTH ACCIDENT ANALYSIS
M. BIN ISLAM, K. KANITPONG
The traveling speed, path, direction, and relative positions of the vehicles (i.e. pickup truck, the bus and motorcycle) were set according to the investigation and
values obtained from reconstruction. Figure 8 shows the
snapshot of simulation when the bus was trying to start
the passing maneuver. Figure 9 shows the separation of
the bus and the pickup truck after full impact and both
vehicles were starting to slow down to their respective rest
positions. As shown in Figure 10, the speed vs. distance
relationship of the bus shows the events for bus before
collision (e.g. Perception-Identification-Emotion-Volition
(PIEV) distance and pre-crash braking), at collision (e.g.
point of impact) and after collision (e.g. point of rest).
Pickup truck: It was traveling at about 55 kph along
the straight section of the undivided 2-lane 2-way rural
road. The pickup truck driver was observing the oncoming vehicles in the opposite lane. There was a motorcycle
coming in the opposite lane. During night time driving
under no street light condition, the driver could have a
glare problem to some extent. Consequently, the pickup
truck driver paid less attention to the following vehicle
(i.e. bus). The driver did not expect any opposing vehicle
would enter the right-of-way of the pickup truck. However, the driver had little time (e.g. due to un-expectancy
and glare) to react and no time to apply the brakes. Since
no skid marks by the pickup truck were found during the
investigation, the speed vs. distance profile of the pickup
truck was not plotted.
Fig. 8 Snapshot of starting passing maneuver of bus
(Left: Plan view, Right: Bus driver’s view)
Fig. 9 Snapshot of separation of the bus and pickup
truck after full impact (Left: Plan view, Right:
Bus driver’s view)
70
Initial Travel Speed (V* = 65 kph)
60
Speed (kph)
50
40
30
20
Pt. of No Escape
Pt. of Actual Perception
10
Pt. of Operator
Action
POI
POR
0
0
5
10
15
20
25
30
35
40
45
Perception/Recation (PIEV = 1.5 sec)
11.08 m
50
55
60
65
Braking
9.8 m
27.12 m
11.08 m
SSD = 48 m
SSD = 48 m
Distnace (m)
Note: PIEV: Perception-Identification-Emotion-Volition, POI: Point of Impact, POR: Point of Rest, SSD: Safe Stopping Distance.
Fig. 10 Speed vs. distance profile for bus
IATSS RESEARCH Vol.32 No.2, 2008
63
TRANSPORTATION
Bus: It was traveling at about 65 kph after crossing
the curve upstream. The required safe Stopping Sight
Distance (SSD) was 48 m. Considering PIEV time of 1.5
second, PIEV distance was about 27 m and pre-impact
skidding took 9.8 m which made about 37 m in total with
initial traveling speed of 65 kph. According to the calculations, the bus driver could avoid the crash if he started
braking at about 11 m before the actual braking position
as shown in Figure 10. At x = 38 m, the bus driver applied
braking and collided with the pickup truck at x = 48 m
with about 47 kph.
Due to the misjudgment of the bus driver to the
pickup truck speed which was traveling (pickup truck: 55
kph) 10 kph lower than the bus (bus: 65 kph) and also
with no street lighting resulted in too late decision making by the bus driver to pass the leading vehicle (i.e. motorcycle). The misjudgment of a gap between the vehicles
(i.e. bus and motorcycle) and high traveling speed of the
bus resulted in passing and entering to the opposing lane.
The late decision of passing maneuver of the bus led to
the collision with the oncoming pickup truck even though
the bus was applying the brakes. All these factors influenced the bus driver to cross the point of no escape at
x > 0 as shown in Figure 10.
5.6
Application of ‘Event Tree’ concept
The sequences of the events presented in Figure 11
(for Pickup truck) and 12 (for Bus) are based on the phys-
Events Prior to Crash
Lack of street light facilities
Less conspicuity due to absence of
street lights, presence of shadow of
trees and advertisement boards
Observation of oncoming vehicles
Observing oncoming small
profile-motorcycle in the opposite lane
Misjudgment of speed of motorcycle
Fast closing distance of single headlight
of small profile-motorcycle and pickup
truck
Driving along straight section
Traveling in long straight section for
long time with speed (55 km/hr) in
undivided 2-lane, 2-way road
Point of No Escape
ical evidence found at the crash scene, damaged vehicles,
information of the crash investigation, results of the crash
reconstruction, interview of eye witnesses and local people, and police reports. Following the concept of DREAM
analysis, possible factors and observable consequences
were presented in Figures 11 and 12. The branching out of
these consequences from their causes led to the collision.
Pickup truck: Lack of street light facilities, misjudgment of speed of oncoming vehicles, and normal driving expectancy on the straight road affected unclear
vision of the pickup truck driver, less attention to following vehicles and unexpected state for the pickup truck
driver. The later caused less time to react and literally no
time to brake to avoid the oncoming bus from the opposite direction. All these factors and consequences are interlinked to one another. The event tree for the pickup
truck is presented in Figure 11 with events prior to crash
vs. time line.
Bus: Lack of street light facilities, together with
misjudgment of speed of oncoming vehicles, and less attention in driving caused a faulty line of vision of the bus
driver, misjudgment of gap between the leading and following vehicles, and temptation of higher speed. Consequently, the misjudgment of the gap caused improper plan
of driving and temptation of higher speed to pass the leading vehicles causing a late decision for passing. This late
decision of passing made the passing vehicle applied late
braking to avoid the collision with the oncoming vehicle
Unclear vision of pickup truck driver
Sudden oncoming headlights of bus
within short interval of motorcycle’s
single headlight
Less recovery time for pickup
truck driver
Sudden and unexpected maneuvering
of bus into the “wrong lane”
Less attention to following vehicle
Paying less attention to succeeding two
headlights of bus followed by single
headlight of motorcycle
Unexpected state
Keeping the proper lane for the vehicles
as designated for respective right-of-way
Point of Actual Perception
Time Line
Fig. 11 Event tree for pickup truck
64
IATSS RESEARCH Vol.32 No.2, 2008
No time to brake
Not having time to brake
to avoid collision with bus
Point of Operator Action
IDENTIFICATION OF FACTORS IN ROAD ACCIDENTS THROUGH IN-DEPTH ACCIDENT ANALYSIS
Events Prior to Crash
Speeding in straight section
Speeding up of bus in straight
long section after crossing the
curve upstream
Misjudgment of speed of pickup truck
Bus driver estimated speed of pickup
truck slower for safe passing distance
M. BIN ISLAM, K. KANITPONG
Passing got priority
Paying more attention to
pass the leading vehicle (i.e.
motorcycle) in undivided
2-lane 2-way road
Temptation of high speed
Bus driver was tempted to
drive fast to pass the slow
moving vehicles ahead
Misjudgment of gap
Bus driver estimated the gap
between bus and leading vehicle
(i.e. motorcycle) for safe passing
Late decision of passing
Late in starting to pass the
motorcycle in front of bus
Late braking in avoidance
“Flinching” due to sudden
appearance of hazard (i.e.
pickup truck) in the opposite
lane and applying brake late
Less attention in driving
Bus driver paid less attention in driving
during nighttime in undivided 2-lane
2-way road
Lack of street light facilities
Less conspicuity in the line of vision for
bus driver for safe passing sight distance
due to no street lights along the road
Faulty line of vision of bus driver
Going apart of single taillight of
motorcycle and oncoming two
headlights of pickup truck in the
dark section of 2-lane roadway
Point of No Escape
Improper plan in driving
Overtaking maneuver and
sudden appearance of
oncoming pickup truck
Point of Actual Perception
Point of Operator Action
Time Line
Fig. 12 Event tree for bus
from the opposite direction. The event tree for the bus is
presented in Figure 12 with events prior to crash vs. time
line.
6. SUMMARY OF FINDINGS
The summary of event analysis could lead to listing
the factors of the system components. The human factor
of the pickup truck, where the pickup truck driver could
have braked as an evasive action (but actually he did not
in the real situation) was influenced by the unexpectancy
of the oncoming vehicle (i.e. bus) into its (i.e. pickup truck)
own right of way. Less attentive driving on the undivided
2-lane 2-way highway at night in the absence of street
light facilities is also added. For the bus driver, the misjudgment of a gap between bus and leading vehicle (i.e.
motorcycle) together with late passing decision of the
passing vehicle led to late braking to avoid the collision.
For the vehicle aspect, the compatibility of the bus
and smaller vehicle such as a pickup truck in this case
could be taken seriously in terms of mass and geometry
of the vehicles. In addition, the locally modified body
structure of the bus could possibly have late braking response due to the old age of the vehicle.
For the road and environment aspect, no street light
facilities particularly during night time driving on the undivided 2-lane 2-way rural highway could lead to the dif-
ficulty in distinguishing the lane separation. Shadows of
static objects (e.g. campaign boards, information boards)
also contributed to the faulty decisions of the pickup
truck and bus drivers. In addition, no speed limit signs
along that road section were found during the investigation process.
Possible factors from the summary of event analysis
could be listed as follows under system components (i.e.
human, vehicle and road-environment):
Human
Pickup truck:
- Unexpected maneuvering of bus into “wrong lane” in
2-lane road
- Paying less attention to the oncoming vehicles in the
opposite lane in undivided road during nighttime
driving
Bus:
- Misjudgment of distance and speed of the leading vehicle (i.e. motorcycle)
- Late in overtaking for small profile vehicle (i.e. motorcycle)
- Inattentive driving for oncoming vehicles from the
opposite lane in 2-lane-2-way undivided road
Vehicle
Bus:
- Complete stop by applying brakes did not occur due
to having short braking time (0.63 sec)
- No “crash compatible design” between large vehicle
IATSS RESEARCH Vol.32 No.2, 2008
65
TRANSPORTATION
(i.e. bus) to smaller vehicle (i.e. pickup truck)
Road and Environment
Pickup truck:
- Lack of street lighting
- Lack of conspicuity of the static roadside objects
during nighttime
- Difficulty in distinguishing the lane separation clearly due to lack of reflective devices for 2-lane road
- No “Speed Limit” sign along the roadside
Bus:
- Lack of street lighting
- Unclear vision due to shadow of small trees, advertisement boards, information sign along road
- No “Speed Limit” sign or warning sign of curvature
ahead along the long straight section
7. CONCLUSIONS
The event analysis obtained from crash investigation and reconstruction can be applied to determine the
possible contributory factors in the fatal road crash. These
contributory factors are generalized according to the available information of the system components. The event
tree analysis mainly deals with what were the events and
factors that came to the drivers’ attention and decision
making to influence their behavior during driving in the
pre-crash stage. The factors found based on the event analysis can be concluded as follows:
Human Factor
- Cognitive behavior particularly judgment and decision-making based on analytical aspects of reaction
were found to be important particularly in this case.
This behavior was highly dependent on the inaccurate
risk assessment. Both drivers had other causes which
influenced them to take the risk.
- Sensorimotor behavior includes experiences related to
sensory and motor channels. This is another aspect of
human behavior, which highly contributes to crashes.
This effect was found to be important in this case where
evasive action (i.e. braking action) was perceived later
than required. In this case, the pick-up driver could not
react in an appropriate time to avoid collision when
there was a mistake in passing maneuver happened by
the bus.
Road and Environment Factor
- It was found to have a potential effect on road crashes.
The visibility, geometry, lane markings, surface condition, and street light facilities have a potential influence on the drivers to perceive and react in a dynamic
driving condition. The interaction of road and environ-
66
IATSS RESEARCH Vol.32 No.2, 2008
ment is quite complex with driving behavior and performance. In this case, absence of street lights, unclear
lane marking were found to increase the risk of crash
and its severity.
The vehicle factor particularly the vehicular defects
were not thoroughly investigated in this case. Nevertheless, the vehicle compatibility of bus and pickup truck
focusing on a safer vehicle design is highlighted in this
study. It is expected that more advanced vehicle investigation will be further analyzed in the future. Since this
study was undertaken as a pilot study for future in-depth
studies, the limitations and a more systematic way of
analysis will be adopted to understand better about the
pre-crash events and the contributory factors.
Human factors are supposed to be the leading contributory factor in any crash analysis including this case
study. Nevertheless, vehicle and road-environment are
also crucial in influencing driver behavior in pre-crash
and the crash phase. These two factors require more attention in investigation and in-depth analysis in particular
for developing countries where vehicle and road-environment are still not of the same standards as developed
countries10. This in-depth analysis through a demonstrated case study eventually indicates the improvements in
highway design and facilities need to be addressed to ensure safer roads in Thailand.
Since this study was initially conducted in Thailand, the findings and review of this study could play an
important role as a pilot study for further research studies. Considering this pilot study, the objectives and scope
of the research could be broadened to be implemented in
other projects to establish a firm research base. The benefits of such in-depth study envisioned with better understanding of the interrelationship of the system components
in road crashes eventually could be followed in other developing countries in Asia.
REFERENCES
1. World report on road traffic injury prevention: summary, http://
www.who.int/world-health day/2004/infomaterials/world_report/en/summary_en_rev.pdf, (July 24, 2004). (2004).
2. Pan American Health Organization (PAHO). http://www.paho.
org/English/DD/PIN/whd04_main.htm, (July 24,2004).
(2004).
3. Tanaboriboon, Y. The Status of Road Safety in Thailand. ADBASEAN Regional Safety Program, Country Report: CR09,
Final Report, Thailand. (2004).
4. Ogden, K.W. Safer Roads: A Guide to Road Safety Engineering, Ashgate Publishing Limited, England. pp.69-70. (2002).
5. Tanaboriboon, Y., Suriyawongpaisal, P., and Chadbunchachai,
IDENTIFICATION OF FACTORS IN ROAD ACCIDENTS THROUGH IN-DEPTH ACCIDENT ANALYSIS
M. BIN ISLAM, K. KANITPONG
W. Traffic Accident Study: Thailand and Japan Experiences.
Final Report Submitted to the Sumitomo Foundation, Japan.
(1997).
6. Baker, J., and Fricke, L., Process of Traffic Accident Reconstruction. Traffic Accident Reconstruction (L. Fricke, ed.),
Evanston, IL, Northwestern University Traffic Institute. (1990).
7. Van Kirk, D. J. Vehicular Accident Investigation and Reconstruction. CRC press. USA. pp.15-19. (2001).
8. Royal Thai Police crash statistics, www.royalthaipolice.go.th/
crime/traff_main.htm.
9. Huang, Y., H., and Ljung, M. Factors Influencing the Causation
of Accidents and Incidents. Human Factors in Design, edited
by D. de Waard, K.A. Brookhuis and C.M. Weikert, Maastricht,
Shaker. (2004).
10.
Islam, M.B., and Tanaboriboon, Y. “Crash Investigation and
Reconstruction…The New Experience in Developing Countries: Thailand Case Study”. Proceeding of the 13th International Conference on Road Safety on Four Continents
(CD-Rom), Warsaw, Poland, pp.874-884. (2005).
ACKNOWLEDGEMENTS
The authors sincerely acknowledge late Professor Yordphol
Tanaboriboon, who had the vision of such innovative thinking in
the context of Thai road safety research. This study was completely supervised by late Professor Yordphol Tanaboriboon. This
pilot study was made possible for the team members’, particularly
Mr. Sarttrawut Ponboon and Mr. Nuttapong Boontob, motivation
and hard work for such vigorous activity. The investigation team
described in this paper is presently working under Thailand Accident Research Center (TARC) in Thailand.
IATSS RESEARCH Vol.32 No.2, 2008
67
JOURNAL OF PREVENTIVE MEDICINE AND HYGIENE 2006; 47: 69-73
BRIEF
REVIEW
Biological and behavioral factors affecting
driving safety
R. VIVOLI, M. BERGOMI, S. ROVESTI, P. BUSSETTI, G.M. GUAITOLI
University of Modena and Reggio Emilia, Department of Public Health Sciences, Modena, Italy
Key words
Biological factors • Behavioral factors • Driving safety
The European Commission estimates that car crash-related costs in Europe are around 160 billion euros, approximately 2% of the Gross Domestic Product [1]. In
several countries, car crashes are the first cause of
death among subjects aged 15-30, with a direct heavy
impact on the years of life lost; in young subjects car
crashes also represent one of the major causes of disability [2].
Several driver characteristics and driving behaviors
due to age, diet, alcohol consumption, circadian
rhythms, drug intake and diseases may contribute to a
reduced alertness and induce drowsiness with dangerous consequences on driving ability thus increasing the
risk of car crashes. It can be estimated that human factors concerning the psychophysical condition of the
driver are involved in 60-80% of road accidents [2].
Crash involvement rates on a population basis are higher among males than females in all age groups [3]. This
observation may be related to the fact that females drive fewer kilometres/year, drive mainly in town and for
short journeys, rarely in bad weather and usually drive
small engine cars. Males drive for a higher number of
kilometres/year, on motorways for long distance driving and drive trucks or large engine cars.
Most accidents involve subjects under 25 years (35%),
whereas subjects aged over 70 years are involved in ap-
proximately 3% of car crashes, as expected considering
that the percentage of drivers over 70 years of age is
small compared to other age groups [3].
Taking into account the distance travelled (Fig. 1),
crash rates in older subjects are higher than in the middle-aged and comparable to those of young subjects;
crash rates in females are slightly higher than in males
in all age groups [3].
Our study on truck drivers disclosed that the reaction
times worsen (rS = 0.337; p = 0.034) and the number of
correct answers decreases as age increases (rS = -0.354;
p = 0.025) even in a limited age-range (Fig. 2) (unpublished data).
Socioeconomic factors such as low social class and low
educational level, family conditions (divorced or with
divorced parents), job loss and social isolation and several behavioral and psychophysical factors are considered predisposing factors to traffic accidents [4].
Several psychotropic substances taken for recreational
(alcohol and illicit drugs) or medical purposes can impair driving performance either by disturbing the information processing mental function, promoting risk taking behaviour, or by increasing response time [4].
Commonly administered therapeutic drugs, such as antihistaminics, antihypertensives, cardiac glycosides, diuretics, antidiabetic agents and antibiotics may cause
Fig. 1. Driver crash involvement by age and sex, Western
Australia 1989-1992 (crashes
per 100 million km driven) [3].
Reprinted from Accid Anal Prev
1998;30:379-87, with permission from Elsevier.
69
R. VIVOLI ET AL.
Fig. 2. Relationship between age and driver performance (reaction time and number of correct answers) by Vienna Determination Test.
weakness or other side effects, thereby increasing the
risk of a road accident [4]. Although relatively few reports have addressed the role of medical conditions in
road accident involvement, it has been suggested that
several diseases and disabilities may impair driving
performance [4].
Among behavioral factors, alcohol plays an important role in car crashes, and accidents involving alcohol are more likely to result in injuries and deaths
than crashes where alcohol is not a factor [4-6]. A
large proportion of accidents are attributable to alcohol (in Europe about 20%) mainly in young people:
the intake of alcoholic beverages when associated
with narcotics use may represent the most dangerous
combination that when is increases the risk of serious
crashes [4].
Sleepiness while driving and/or falling asleep at the
wheel are other important risk factors for injuries from
car crashes, though the exact role of these factors has
yet to be fully elucidated [7]. Estimates of the proportion of road traffic accidents due to sleepiness while
driving vary widely between nations, ranging from 1%
to 33% [7]. This contribution is most likely underesti-
Fig. 3. Reasons for sleep loss [11]. Reprinted from Accid Anal
Prev 1997;29:463-9, with permission from Elsevier.
mated, due to the lack of a standardised definition of
“sleep-related vehicle accidents” and/or due to insurance-related problems. The percentage of road traffic
accidents ascribed to sleepiness in Italy has been estimated at around 21% [8].
Individuals at highest risk of sleep-related accidents are
generally young subjects, particularly males, individuals with undiagnosed or untreated sleep disorders, subjects who use sleep-inducing drugs, shift workers or
truck drivers with long periods of driving, a high frequency of night driving and lack of adequate rest (< 6
hours of sleep) [9, 10].
Sleep loss is also an important risk factor in city accidents
occurring on short trips and the main reasons for sleep
loss (Fig. 3) are work and partying/social events [11].
As regards the distribution of sleep-related vehicle accidents by time of day, Figure 4 shows three peaks, two
during the early morning (at around 02-03, and 06-07)
when traffic flow rates are low, and another one in the
mid-afternoon (16-17) at a time of high traffic density
[12].
Sleepiness decreases performance, reducing the vigilance level and impairing attention and reaction times
[13]. Even a low and generally ‘safe’ level of alcohol
consumption may exacerbate driving impairment due
to sleepiness [14].
A poorly investigated aspect of road accidents is the
role of the driver’s chronotype in the induction of
sleepiness. Several studies have shown interindividual
differences in the spontaneous sleep-wake cycle and it
has been suggested that this variability may influence
individual levels of performance. In particular it has
been observed that performance differences between
the two chronotypes (morning-type or evening-type)
are related to the time of day [15].
Recent studies have shown that peaks of melatonin
synthesis occur on average three hours earlier among
individuals with morning chronotype compared with
evening chronotypes [16-18]. The phase and magnitude
of the melatonin secretion cycle, at the beginning or
70
FACTORS AFFECTING DRIVING SAFETY
Fig. 4. Incidence of sleep related vehicle accidents and traffic
flow rates by hour of day [12]. Reprinted from BMJ
1995;310:565-7, with permission from the BMJ Publishing
Group.
end of the work shift, may represent a marker of maladjustment and stress condition among vehicle drivers
[19, 20].
Among the human risk factors of road accidents, several psychological and behavioral factors seem to be associated with an increased risk of motor-vehicle accidents [4, 21]. Several attempts have been made to
analyse the role of these factors in triggering accidents
and it has been suggested that certain personality traits
can affect “driving style”. To shed more light on the
relative contribution of personality factors and driving
behaviors in accident involvement, a model distinguishing distal (i.e. personality factors) and proximal
(i.e. aberrant driving behaviors) predictors of traffic accidents has been proposed (Fig. 5) [21].
Some personality traits, such as impulsiveness and sensation seeking have been implicated as major factors in
the risk-taking disposition. In particular, the “Sensation
Seeking” trait related to pursuit of “strong” sensations
and of an adventurous and hazardous life, seems to increase the risk of road traffic accidents. “Sensation
Seeking”, a personality trait that seems to be genetically based, is more pronounced among young males, and
is related to behaviors such as reckless driving, sometimes associated with excess alcohol consumption [22,
23]. “Sensation Seeking” is directly correlated to hazardous driving behaviours such as high speed, infringement of safety distance and other rules of the highway
code, driving after excess alcohol intake, etc. [24, 25].
It has been hypothesized that the interaction between
stable personality factors and transient elements such
as stressful events, fatigue or drinking might play a major role in crash causation [26].
Normal car driving on the road, especially under difficult conditions, is considered one of the most significant stressors of everyday life and is influenced by several individual and environmental factors. Consequently, the driver’s performance, as well as road safety, may
be affected by the stress induced by driving. Significant
changes in stress hormones, such as catecholamines (in
particular adrenaline) and cortisol or both, have been
detected in studies carried out on bus, truck and racing
car drivers [27-33]. The finding that adrenaline excretion rates were significantly correlated to anxiety
scores in both truck and racing car drivers suggests that
the degree of adrenergic response is influenced by the
psychological profile [31, 33].
Driving motor-vehicles under stressful environmental
conditions (long-distance driving, traffic or weather
conditions) may trigger a major activation of the car-
Fig. 5. Proposed contextual mediated model [21]. Reprinted from Accid Anal Prev 2003;35:949-64, with permission from Elsevier.
71
R. VIVOLI ET AL.
diovascular system. Our study on truck drivers
showed an increase in heart rate and onset of
supraventricular extrasystoles during conditions of
traffic jams and fog [31]. In addition, the marked increase in urinary levels of thromboxane B2 found in
truck drivers at the end of the working-shift suggests
that the stressful conditions of long distance driving
might interact with the release of this modulator of
platelet function [32].
In conclusion, among the human factors related to driving safety, some individual characteristics such as age
and gender and lifestyle features such as alcohol and
drug intake increase the risk of being involved in motor-vehicle crashes.
Young male subjects with particular personality traits
(aggressiveness, sensation seeking) are likely to have
aberrant driving behaviors (driving speed, violations,
alcohol abuse) that increase the probability of road accidents.
A large proportion of traffic accidents can be ascribed
to drowsiness or falling asleep that usually hit the driver in the early morning hours.
Among the factors affecting sleepiness, driver chronotype may influence driving safety, mainly at certain
times of day, since individual variability related to the
sleep-wake cycle has been associated with changes in
performance rhythms. Although the large literature data on traffic accidents, the potential causative role of
several factors overviewed in this paper (personality
traits, chronotype and others) needs to be clarified in
future researches.
A better understanding of the human factors affecting
motor-vehicle accidents is required to adopt appropriate measures to increase driving safety.
References
[1] Commissione delle Comunità Europee. Comunicazione della
Commissione - Programma di azione europeo per la sicurezza
stradale - Dimezzare il numero di vittime della strada nell’Unione europea entro il 2010: una responsabilità condivisa.
http://europa.eu.int/scadplus/leg/it/lvb/l24257.htm
[2] Taggi F, Giustini M, Fondi G, Macchia T, Chiarotti M. L’epidemiologia degli incidenti stradali (I): i dati di base e i fattori
di rischio. In: Atti della 53a Conferenza del Traffico e della Circolazione, Stresa, 1-4 ottobre 1997, pp. 67-79.
[3] Ryan GA, Legge M, Rosman D. Age related changes in drivers’ crash risk and crash type. Accid Anal Prev 1998;30:37987.
[4] Petridou E, Moustaki M. Human factors in the causation of
road traffic crashes. Eur J Epidemiol 2000;16:819-26.
[5] Fabbri A, Marchesini G, Morselli-Labate AM, Rossi F, Cicognani A, Dente M, et al. Positive blood alcohol concentration
and road accidents. A prospective study in an Italian emergency
department. Emerg Med J 2002;19:210-4.
[6] Krüger H-P, Vollrath M. The alcohol-related accident risk in
Germany: procedure, methods and results. Accid Anal Prev
2004;36:125-33.
[7] Connor J, Norton R, Ameratunga S, Robinson E, Civil I, Dunn
R, et al. Driver sleepiness and risk of serious injury to car occupants: population based case control study. BMJ
2002;324:1125-8.
[8] Garbarino S, Nobili L, Beelke M, De Carli F, Balestra V, Ferrillo F. Sleep related vehicle accidents on Italian highways. G
Ital Med Lav Erg 2001;23:430-4.
[9] Stutts JC, Wilkins JW, Osberg JS, Vaughn BV. Driver risk factors for sleep-related crashes. Accid Anal Prev 2003;35:32131.
[10] Horne J, Reyner L. Vehicle accidents related to sleep: a review.
Occup Environ Med 1999;56:289-94.
[11] Fell DL, Black B. Driver fatigue in the city. Accid Anal Prev
1997;29:463-9.
[12] Horne JA, Reyner LA. Sleep related vehicle accidents. BMJ
1995;310:565-7.
[13] Dinges DF. Probing the limits of functional capabilities. The effects of sleep loss in short duration tasks. In: Broughton RJ,
Ogilvie RD, eds. Sleep, Arousal and Performance. Boston
(MA): Birkhauser 1992, p. 176-88.
[14] Horne JA, Reyner LA, Barrett PR. Driving impairment due to
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
72
sleepiness is exacerbated by low alcohol intake. Occup Environ
Med 2003;60:689-92.
Mecacci L, Righi S, Rocchetti G. Cognitive failures and circadian typology. Pers Individ Dif 2004;37:107-13.
Carrier J, Monk TH. Circadian rhythms of performance: new
trends. Chronobiol Int 2000;17:719-32.
Rosenthal L, Day R, Gerhardstein R, Meixner R, Roth T, Guido P, et al. Sleepiness/alertness among healthy evening and
morning type individuals. Sleep Med 2001;2:243-8.
Griefahn B. The validity of the temporal parameters of the daily
rhythm of melatonin levels as an indicator of morningness. Chronobiol Int 2002;19:561-77.
Coblentz A, Cabon P, Bougrine S. Sleep deprivations and irregular work schedules in transport operations. In: Proceedings
of the 2nd Pan-Pacific Conference on Ergonomics in Occupational Safety Health, Wuhan (China), 1-5 November 1992, p.
135-40.
Bougrine S, Mollard R, Ignazi G, Coblentz A. Appropriate use
of bright light promotes a durable adaptation to night-shifts
and accelerates readjustment during recovery after a period of
night-shifts. Work Stress 1995;9:314-26.
Sümer N. Personality and behavioral predictors of traffic accidents: testing a contextual mediated model. Accid Anal Prev
2003;35:949-64.
Zuckerman M, Kuhlman DM. Personality and risk-taking:
common biosocial factors. J Pers 2000;68:999-1029.
Iversen H, Rundmo T. Personality, risky driving and accident
involvement among Norwegian drivers. Pers Individ Dif
2002;33:1251-63.
Jonah BA. Sensation seeking and risky driving: a review and
synthesis of the literature. Accid Anal Prev 1997;29:651-65.
Jonah BA, Thiessen R, Au-Yeung E. Sensation seeking, risky
driving and behavioral adaptation. Accid Anal Prev
2001;33:679-84.
Elander J, West R, French D. Behavioral correlates of individual differences in road-traffic crash risk: an examination of
methods and findings. Psychol Bull 1993;113:279-94.
van der Beek AJ, Meijman TF, Frings-Dresen MH, Kuiper JI,
Kuiper S. Lorry drivers’ work stress evaluated by catecholamines excreted in urine. Occup Environ Med 1995;52:464-9.
Sluiter JK, van der Beek AJ, Frings-Dresen MHW. Work stress
and recovery measured by urinary catecholamines and cortisol
excretion in long distance coach drivers. Occup Environ Med
1998;55:407-13.
FACTORS AFFECTING DRIVING SAFETY
[29] Matthews G, Dorn L, Hoyes TW, Davies DR, Glendon AI, Taylor RG. Driver stress and performance on a driving simulator.
Hum Factors 1998;40:136-49.
[30] Vivoli G, Bergomi M, Caselgrandi E. Biochemical and psychological study of stress in bus drivers. In: Proceedings of the
Workshop on Effects of Automation on Operator Performance
(A. Coblentz Ed.), Paris, 27-28 October 1986, Commission of
the European Communities, Medical and Public Health Research Programme, p. 80-96.
■ Received on June 17, 2005. Accepted on January 5, 2006.
■ This paper was presented at the Fourth European Conference on
Travel Medicine, Rome, 29-31 March 2004.
■ Correspondence: Dr Roberto Vivoli, Department of Public
Health Sciences, University of Modena and Reggio Emilia, via
Campi 287, 41100 Modena, Italy. Tel. +39 059 2055460 – Fax
+39 059 2055483 – E-mail: vivoli@unimo.it
73
[31] Vivoli G, Bergomi M, Rovesti S, Carrozzi G, Vezzosi A. Biochemical and haemodynamic indicators of stress in truck drivers. Ergonomics 1993;36:1089-97.
[32] Bergomi M, Vivoli R, Rovesti S, Malagoli C, Pecone L, Vivoli
G. Biological indicators of stress in lorry drivers. Epidemiology 2002;13:S170-1.
[33] Rovesti S, Vivoli R, Bergomi M, Vivoli G. Biological indicators of stress in racing car drivers. Heavy Vehicle Systems,
Special Series, Int J Vehicle Design 1997;4:340-52.
Accident Analysis and Prevention 34 (2002) 357– 365
www.elsevier.com/locate/aap
Effects of road geometry and traffic volumes
on rural roadway accident rates
Matthew G. Karlaftis *, Ioannis Golias
Department of Transportation Planning and Engineering, Faculty of Ci6il Engineering, National Technical Uni6ersity of Athens,
5 Iroon Polytechnion Street, 157 93 Zografon, Athens, Greece
Received 16 May 2000; received in revised form 1 February 2001; accepted 28 February 2001
Abstract
This paper revisits the question of the relationship between rural road geometric characteristics, accident rates and their
prediction, using a rigorous non-parametric statistical methodology known as hierarchical tree-based regression. The goal of this
paper is twofold; first, it develops a methodology that quantitatively assesses the effects of various highway geometric
characteristics on accident rates and, second, it provides a straightforward, yet fundamentally and mathematically sound way of
predicting accident rates on rural roads. The results show that although the importance of isolated variables differs between
two-lane and multilane roads, ‘geometric design’ variables and ‘pavement condition’ variables are the two most important factors
affecting accident rates. Further, the methodology used in this paper allows for the explicit prediction of accident rates for given
highway sections, as soon as the profile of a road section is given. © 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Accident rates; Rural roads; Hierarchical tree based regression
1. Introduction
Road safety modelling has attracted considerable
research interest in the past four decades because of its
wide variety of applications and important practical
implications. Public agencies, such as State Departments of Transportation, may be interested in identifying accident-prone areas to promote safety treatments.
Transportation engineers may be interested in identifying those factors (traffic, geometric, etc.) that influence
accident frequency and severity to improve roadway
design and provide a safer driving environment.
The very high cost of highway accidents paid by
societies around the world makes highway safety improvement an important objective of transportation
engineering. Highway safety specialists can influence
traffic safety either through means such as road rules,
law enforcement, and education, or by applying local
traffic control and geometry improvements. An over* Corresponding author. Tel.: + 30-1-6711203; fax:
7721327.
E-mail address: mgk@central.ntua.gr (M.G. Karlaftis).
whelming majority of previous studies have indicated
that improvements to highway design could produce
significant reductions in the number of crashes. Recognizing this, the Federal Highway Administration
(FHWA) promotes safety and accident investigation by
encouraging States to pursue the development of Safety
Management Systems (SMS). And, although SMSs are
not Federally required as of 1996, most States continue
to work on their development, suggesting the need for
improving on existing empirical models for accident
measurement.
Following a long line of studies concerned with identifying major factors contributing to highway accidents,
this paper revisits the problem of the relationship between rural road geometric characteristics, accident
rates and their prediction, using a rigorous non-parametric statistical methodology known as hierarchical
tree-based regression (HTBR).1 The goal of this paper
is not only to develop a methodology that quantitatively assesses the effects of various rural road geomet-
+ 30-11
For information on the process of functional road classification
the reader should refer to US DOT (1969).
0001-4575/02/$ - see front matter © 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 0 0 1 - 4 5 7 5 ( 0 1 ) 0 0 0 3 3 - 1
358
M.G. Karlaftis, I. Golias / Accident Analysis and Pre6ention 34 (2002) 357–365
ric characteristics on accident rates, but also to provide
a straightforward, yet fundamentally and mathematically sound way of predicting accident rates. The ability
to predict accident rates is very important to transportation planners and engineers, because it can help in
identifying hazardous locations, sites which require
treatment, as well as spots where deviations (either
higher or lower rates) from expected (predicted) warrants further examination. The remainder of the paper
is organized as follows. The next section provides some
background necessary for the development of the
methodology used in this paper. Following this, the
data and methodology that were used, the estimation
results, and examine the effects of various geometric
characteristics on accident rates are presented and discussed. The final section of the paper summarizes the
findings and offers some concluding remarks.
2. Background
Much literature exists that addresses the problem of
accident rate estimation, and the identification of the
various factors affecting this rate. Joshua and Garber
(1990) used multiple linear and Poisson regression to
estimate truck accident rates using traffic and geometric
independent variables. Jones and Whitfield (1991) used
Poisson regression with data from Seattle to identify
the daily characteristics (traffic, weather, etc.) that may
influence the number of traffic accidents. Miaou et al.
(1992) used Poisson regression on traffic data from
8779 miles of roadway from the Highway Safety Information System (HSIS) to establish quantitative relationships between truck accident rates and highway
geometric characteristics. Their results indicate that surrogate measures for mean absolute curvature (for horizontal alignment) and mean absolute grade (for vertical
alignment) are the most important variables for accident rate estimation.
In a study of approximately seven thousand miles of
roadway logs in Utah, Mohamedshah et al. (1993) used
linear regression to predict truck accident involvement
rate per mile per year, based on average Average
Annual Daily Traffic (AADT) and truck AADT per
lane, shoulder width, horizontal curvature, and vertical
gradient. The results suggest that truck involvement
rate increases with AADT and truck AADT, degree of
curvature and gradient. Hadi et al. (1993), using data
from the Florida Department of Transportation’s
Roadway Characteristics Inventory (RCI) system, estimated negative binomial (NB) regression for accident
rates on various types of rural and urban highways
with different traffic levels. Their results suggest that
higher AADT levels and the presence of intersections
are associated with higher crash frequency, while wider
lanes and shoulders are effective in reducing crash rates.
In that paper, the authors also provide an extensive
review of earlier findings relating accident rates and
geometric characteristics.
More recently, Ivan and O’Mara (1997), using NB
regression on 1991– 1993 data from the Traffic Accident Surveillance Report of Connecticut found that
annual average daily traffic was a critical accident
prediction variable, while geometric design variables
and speed differential measures were not found to be
effective predictors of accident rates. Karlaftis and
Tarko (1998), based on a county accident data set from
Indiana, estimated macroscopic accident models that
attempt to explicitly control for cross-section heterogeneity in NB regression that may otherwise seriously
bias the resulting estimates and invalidate statistical
tests. Data collected from the States of Minnesota and
Washington on rural two-lane highways, estimated accident models for segments and three-legged and fourlegged intersections stop- controlled on the minor legs.
Independent variables for their models included traffic,
horizontal and vertical alignments, lane and shoulder
widths, roadside hazard rating, channelization, and
number of driveways. Results imply that segment accidents depend significantly on most of the roadway
variables collected, while intersection accidents depend
primarily on traffic.
This brief review of some of the existing literature
suggests that a variety of traffic and design elements
such as AADT, cross-section design, horizontal alignment, roadside features, access control, pavement conditions, speed limit, lane width (LW), and median
width, affect accident rates. And, most of these results
have been based on multiple linear or Poisson and NB
regression models.
Much of the early work in the empirical analysis of
accident data was done with the use of multiple linear
regression models. As the literature has repeatedly
pointed out, these models suffer from several methodological limitations and practical inconsistencies in the
case of accident modelling (Lerman and Gonzales,
1980). To overcome these limitations, several authors
used Poisson regression models that are a reasonable
alternative for events that occur randomly and independently over time. Despite its advantages, Poisson regression assumes equality of the variance and mean of
the dependent variable. This restriction (which, when
violated, leads to invalid t-tests of the parameter estimates), can be overcome with the use of NB regression
which allows the variance of the dependent variable to
be larger than the mean. As a result, most of the recent
literature has used NB regression models to evaluate
accident data.
But, while NB regression has been instrumental in
overcoming most of the problems associated with models involving count data, it still remains a parametric
procedure requiring the functional form of the model to
M.G. Karlaftis, I. Golias / Accident Analysis and Pre6ention 34 (2002) 357–365
be specified in advance, it is not invariant with respect
to monotone transformation of the variables, it is easily
and significantly influenced by outliers, it does not
handle well discrete independent variables with more
than two levels, and it is adversely affected by multcollinearity among independent variables (Hadi et al.,
1993; Mohamedshah et al., 1993; Tarko et al., 1996;
Karlaftis and Tarko, 1998). It is likely, for example,
that while the accident models have been correctly
specified, multicollinearity has inflated the variance of
some of the independent variables coefficient estimates,
leading to lower t-statistic values and to coefficients
that are not significant and/or are counter-intuitive.
In this paper a methodology which attempts to recognize the existence of the above mentioned problems
and develop a framework to account for them is introduced. This methodology, known as HTBR or as Binary Recursive Partitioning (BRT) (Breiman et al.,
1984), can be of assistance in overcoming some of the
problems associated with multiple linear and NB regression. It should be noted that besides overcoming
the above, rather theoretical problems, the proposed
methodology has three additional strengths. First, it
allows for straightforward and quantitative assessment
of the effect of various rural road geometric characteristics on accident rates; second, it allows for the quick
estimation of predicted accident rates for a given rural
road section; and, third, it is easily amenable to ‘if-then’
statements for incorporation in expert systems which
have become increasingly popular and useful in safety
management. The strengths and weaknesses of the proposed methodology are demonstrated using Indiana
State Police Accident Information records and Indiana
Department of Transportation’s Road Inventory database. The combined database includes five years (1991–
1995) of crashes on Indiana rural roads, along with the
geometric and traffic characteristics for these roads.
3. Data and methodology
3.1. The data
The data used in this paper concern rural roads and
come from two sources: the Road Inventory database,
from the Indiana Department of Transportation (INDOT), and the Accident Information Record form the
Indiana State Police. The first database contains a list
of road sections and various traffic and geometric
characteristics for those sections. The second database
contains a description of the location and type of
accidents that occurred on Indiana’s roads. Combining
these two yields a database that contains five years
(1991– 1995) of accident data for Indiana along with
the traffic and geometric characteristics for the location
of each accident.
359
The availability of such data allows for inferences to
be drawn on the effects of traffic and geometric characteristics on highway accidents. Further, to avoid the
possibility of heterogeneity among roads with different
number of lanes and based on previous findings in the
literature (Hadi et al., 1993; Mohamedshah et al., 1993;
Karlaftis and Tarko, 1998), road sections were grouped
into two main categories: rural two-lane and rural
multilane. The variables available for model estimation
appear in Table 1.
3.2. The methodology
As previously mentioned, NB regression has accounted for most of the theoretical issues in count data
research. Nevertheless, there still remain a number of
issues that have not been addressed (Hadi et al., 1993;
Mohamedshah et al., 1993; Tarko et al., 1996; Karlaftis
and Tarko, 1998). First, NB regression, much like
multiple linear and Poisson regression, is a parametric
procedure requiring for the functional form of the
model to be known in advance. Second, it is easily and
significantly affected by outliers. Third, it cannot handle missing data well. Fourth, it does not treat satisfactorily discrete variables with more than two levels.
Fifth, it does not deal well with multicollinear independent variables.
HTBR is a tree-structured non-parametric data analysis methodology that was first used in the 1960s in the
medical and the social sciences (Morgan and Sonquist,
1963). An extensive review of the methods used to
estimate the regression trees and their applications can
be found in Breiman et al. (1984). HTBR is technically
binary, because parent nodes are always split into exactly two child nodes, and is recursive because the
process can be repeated by treating each child node as
a parent. In essence, the HTBR algorithm proceeds by
iteratively asking the following two questions: (i) which
of the independent variables available should be selected for the model to obtain the maximum reduction
in the variability of the response (dependent variable);
and (ii) which value of the selected independent variable (discrete or continuous) results in the maximum
reduction in the variability of the response. These two
steps are repeated using a numerical search procedure
until a desirable end-condition is met. In mathematical
terms, deviance D is initially defined as2
2
In this section only the essential parts of the HTBR methodology
formulation that may be of interest to the reader are presented.
Readers interested in the details of the formulation are encouraged to
refer to Breiman et al. (1984) for an in-depth treatment, or Washington and Wolf (1996) and Washington et al. (1996) for a presentation
of the methodology in the context of engineering applications. The
discussion of HTBR presented in this paper is based on Washington
and Wolf (1996).
M.G. Karlaftis, I. Golias / Accident Analysis and Pre6ention 34 (2002) 357–365
360
L
Da = % (yia −x̄a )2,
(1)
l=1
where Da is the total deviance of a variable y at node a,
or the sum of squared error (SSE) at the node, yia is the
observation on dependent variable y in node a and is
the mean of L observations in node a.
A split of the observations can be found at node a on
a value of an independent variable x1 that results in two
branches and corresponding nodes b and c, each containing M and N of the original L observation (M+
N = L). The goal of HTBR is to find the variable x1 at
its optimum split (i ) so that the reduction in deviance
is maximized, or more formally when
Z(Öx) = maximum.
(2)
The maximum reduction occurs at some x1(i ) (independent variable x1 at value i ). When the data are split
at this value of x, the remaining two samples have
much smaller variance of y than the original data set.
Numerical search procedures are employed to maximize
Eq. (2).
The HTBR methodology has several attractive technical properties: it is non-parametric and does not
require specification of a functional form; it does not
require variables to be selected in advance since it uses
a stepwise method to determine optimal splitting rules;
its results are invariant with respect to monotone transformations of the independent variables; it can handle
data sets with complex (non-homegeneous) structure; it
is extremely robust to the effects of outliers; it can use
any combination of categorical and qualitative (discrete) variables; and, it is not affected by multicollinearity between the independent variables. Further, and as
it pertains to this research, HTBR can straightforwardly yield predictions for the ‘dependent’ variable
(y), incorporating the optimal splitting rules in an
‘if-then’ series of statements, making the incorporation
of the results in an expert system rather simple.
4. HTBR model estimation and interpretation
As previously mentioned, HTBR partitions the data
into relatively homogeneous (low standard deviation)
terminal nodes, and it takes the mean value observed in
each node as its predicted value. In general, HTBR
models can be fairly complex and detailed, and therefore difficult to illustrate mathematically. Nevertheless,
the methodology lends itself to graphical ‘tree’ like
representations well.
The model shown in Fig. 1 is the result of the HTBR
methodology applied to crashes on rural two-lane
roads. Interpreting the tree, both for explanatory and
predictive purposes, is rather straightforward. The top
of the tree, or root node, shows that the first optimal
split for crashes on rural two-lane roads occurs on
Table 1
Independent variables available for model estimation
Variable
Symbol
Type
Description
Section length
Number of
lanes
Lane widths
Shoulder widths
Median width
Shoulder type
Pavement type
L
NoL
Continuous
Count
Length of the road section were an accident occurred
Number of moving traffic lanes in the section
LW
SW
MW
ST
PT
Continuous
Continuous
Continuous
Qualitative
Binary
Concrete
pavement
Median type
CP
Binary
MT
Qualitative
TL
NoC
Binary
Count
Widths of the northbound, southbound, and average lane widths
The widths of the left, right, inside, and outside shoulders
Width of the median (or 0 if median not available)
Dummy variables for type of shoulder (paved, earth, stabilized)
The variables takes the value of 1 if the road surface is bituminous concrete, sheet or rock
asphalt, and 0 otherwise
The variable takes the value of 1 if the road surface is Portland concrete cement and 0
otherwise
The variable takes the value of 0 if there is no median, 1 for grass or sod, 2 for bituminous
concrete, and 3 for non-mountable barrier median
These variables indicate the presence of left, right, left and right, and continuous turn lanes
The number of curbs on the road section (0, 1, 2)
NoPL
Count
The number of parking lanes on the section (0, 1, 2)
FR
Continuous
Coefficient of wet sliding (skidding) FR at 40 mph between a wet pavement surface and a
standard tire
Takes the value of 0 for dirt and gravel roads, 1 for very poor, 2 for deteriorated, 3 for fair, 4
for good, and 5 for very good pavements
Turn lanes
Number of
curbs
Number of
park lanes
Friction
Pavement
Serviceability
Index
Access control
SI
Qualitative
A
Qualitative
AADT
AADT
Continuous
Takes the value of 1 for no access control, 2 for partial access control, and 3 for full access
control
Annual Average Daily Traffic
M.G. Karlaftis, I. Golias / Accident Analysis and Pre6ention 34 (2002) 357–365
Fig. 1. Regression tree for accidents and geometric characteristics on rural two-lane roads.
361
362
M.G. Karlaftis, I. Golias / Accident Analysis and Pre6ention 34 (2002) 357–365
AADT, sending cases (road sections) with less than or
equal to 8020 to the left and all others to the right. In
other words, the single best variable to explain the
variability in total crashes on rural two-lane roads is
AADT. Assume for the moment the interest is in rural
roads with AADT larger than 8020. Conditional on
this, the next best explanatory variable is LW. For LW
less than or equal to 12.5 ft the road sections go to the
left, where for LW larger than 12.5 ft the road sections
go to the right forming what is called a terminal node,
or leaf of the tree. For these road sections the tree
predicts an average of 32 accidents (normalized on
section length). The remaining splits, for the road sections with LW less than or equal to 12.0 ft, are made on
Friction (FR) and Serviceability Index (SI). In general,
an estimate on the number of accidents is obtained by
continuing down the branches of the tree in similar
fashion until a terminal node is reached. Recall that the
estimate provided at terminal nodes is the mean of the
sample at the node. This means that, since there is a
number of observations that fall within the characteristics of a terminal node, the expected number of accidents is the mean of those observations. For example,
there are 37 observations with AADT\ 8020 and
LW \12.5, and their mean is 32 accidents.
More importantly, since transportation planners are
very frequently interested in predicting accident rates
for given highway sections, the profile of a road section
can be examined, and the tree can be used to determine
the prediction. For example, assume a planner wants to
predict the expected number of accidents for a rural
two-lane road section with AADT of 4000, FR of 0.5,
and LW of 12.0 ft. Beginning at the root node (top of
the tree), we branch left (AADT0 8020), right
(AADT \2751), left (FR0 0.5), right (LW\ 11.5), to
get the estimate of 17.2 crashes for that highway
section.
It should be noted that, for the tree-structure, a
X 2-test was used to evaluate the ‘accuracy’ of the
predictions. Using a ‘hold-out’ sample of 120 randomly
selected observations, the tree structure was estimated
on the remaining 334 observations (for the rural twolane road case). Then, using the ‘if-then’ rules yielded
by the estimated tree, the accident rates for the 120
hold-out observations were estimated. At the 90% significance level for the X 2-test, the null hypothesis that
the difference between the actual and predicted rates
was zero could not be rejected.
Nevertheless, it should be noted that while the ‘holdout sample’ method is a rather popular approach to
validating the estimates yielded by the tree approach, it
does have a shortcoming. Because both the estimation
and validation samples are from the same general area
(the State of Indiana), it is not surprising that their
patterns are similar and hence the results of the model
validation process are good. As such, it would be
interesting to cross-validate the estimation results with
data from a different area (but from rural roads
nonetheless). In general of course, the process of randomly selecting a subsample for validation is the most
frequently used technique.
Looking closer at Fig. 1, it is clear that for lower
flows the parameter that seems to be more important is
the FR coefficient, while for higher flows LW seems to
have the greatest importance. This seems to be justified
by the fact that lower flows are related to higher speeds,
which render the slippery of the roads a critical
parameter. However, when flows are high the risk for
an accident seems to stem mainly from the interaction
of vehicles travelling at the same or opposite direction,
rendering LW the more important factor.
Fig. 2 shows the results of the HTBR methodology
applied to crashes on multilane rural roads. Interestingly, using again the X 2-test, the methodology yielded
a ‘simpler’ tree. Its first optimal split occurs on AADT.
Thus, it seems that AADT is the best variable to
explain crash variability in multilane roads as well.
What may also be noted is that for lower flows the
existence of a median is an important factor while when
it comes to higher flows the existence of access control
seems to be the more important factor safety wise.
Thus, vehicle interaction and vehicle maneuvering arrangements prove again to be important factors when
traffic demand increases. For predicti6e purposes, the
profile of a multilane road section can be examined
similar to that of a two-lane road section and the tree
can be used to determine the prediction. For example,
assume a prediction is needed for the number of accidents on a rural multilane road section with AADT of
8300, and No Access Control. Beginning at the root
node (top of the tree), we branch right (AADT\6851),
right (AADT\ 8075), left (A= 1), to get the estimate
of 27.1 crashes for that highway section.
It is interesting to note that some variables are
selected more than once in the estimation process. For
instance, taking all the left branches to the terminal
node (leaf), AADT appears three times. Since one of
the goals of HTBR is to develop a simple tree structure
for data, relatively few variables will appear explicitly
in the splitting criteria, and some highly important
variables will appear more than once (such as AADT in
this tree structure). While this could be taken to mean
that the other variables are not important in understanding or predicting the dependent variable, an independent variable could be considered highly important
even if it never appears as a primary node splitter. The
software used in this paper (CART 1995) keeps track of
surrogate splits in the tree-growing process, evaluating
the contribution a variable makes in prediction by both
primary and surrogate splits. That is, while the treestructure can be used, as previously shown, for predictive purposes, a different measure called variable
M.G. Karlaftis, I. Golias / Accident Analysis and Pre6ention 34 (2002) 357–365
Fig. 2. Regression tree for accidents and geometric characteristics on rural multilane roads.
363
364
M.G. Karlaftis, I. Golias / Accident Analysis and Pre6ention 34 (2002) 357–365
Table 2
Independent variable importance for crash rates (crashes normalized on highway section length)
Rural two-lane
Rural multilane
Variable
Relative importance (%)
Variable
Relative importance (%)
AADT
Lane width
Serviceability index
Friction
Pavement type
Access control
100
72
59
32
30
14
AADT
Median width
Access control
Friction
Lane width
Serviceability index
Pavement type
100
63
59
25
24
21
11
importance score should be used to estimate the importance of the effect of various geometric characteristics
on accident rates.
To calculate a variable importance score, the software looks at the improvement measure attributable to
each variable in its role as a surrogate to the primary
split. The values of these improvements are summed
over each node and totalled, and are then scaled relative to the ‘best’ performing variable. As a result, the
variable with the highest sum of improvements is scored
100, and all other variables will have lower scores
ranging downwards towards zero. The relative importance of the independent variables in explaining crash
rates on various types of roadways appear in Table 2
(for crashes normalized on highway section length), and
Table 3 (for crashes normalized on highway section
length and AADT).
It is interesting to note the differences in the variables
that ‘explain’ crashes on the two types of roadways.
While AADT is overall the most important variable
when crashes are normalized on section length (Table
2), the characteristics of subsequent importance vary
for the two types of roadway. For the rural two-lane
case, LW is the variable with the higher importance
after AADT. It is obvious that the proximity of the
opposing traffic streams renders the width of the lane
an important factor for safety. The next more important variables – SI, FR and pavement type – are
related to the road pavement conditions.
However, when it comes to multilane rural roads the
variables with the higher importance after AADT are
the existence of median width and of access control.
The importance of these two factors seems to be
justified mainly by the increased speeds on multilane
rural roads. This fact renders the above two factors
more important than LW and pavement condition variables, FR, SI and pavement type, which follow in
importance (Table 2). Furthermore, it should be noted
that when crashes are normalized on section length and
AADT (Table 3), the variables of importance are similar to those of Table 2 (normalization on section
length), the only new variable being the existence of a
left turn lane, for both two-lane and multilane rural
roads.
5. Discussion and conclusions
Much interest exists in the area of accident rate
estimation, and the identification of the various factors
affecting this rate. Much of the literature in this area
has concentrated in identifying the factors affecting
accident occurrence (accident rates), and secondarily in
predicting them. The ability to predict accident rates is
very important to transportation planners and engineers, because it can help in identifying hazardous
locations, sites which require treatment, as well as spots
where deviations (either higher or lower rates) from
expected (predicted) levels warrants further examination. The aim of this paper was twofold. First, it
developed a methodology that quantitatively assesses
the effects of various highway characteristics on accident rates. Second, it provided a straightforward, yet
fundamentally and mathematically sound way of predicting accident rates.
The methodology used in this paper, known as
HTBR, has a number of both theoretical and applied
advantages over multiple linear and NB regression that
have been commonly used in accident rate research. It
allows for the quantitative assessment of the effect of
various geometric characteristics on accident rates. It
allows for the quick estimation of predicted accident
rates for a given highway section. Finally, it is easily
amenable to ‘if-then’ statements for incorporation in
expert systems, which have become increasingly popular and useful in safety management. The methodology
was demonstrated using data from the Indiana State
Police Accident Information records and the INDOT’s
Road Inventory database.
The results of the investigation of the roadway characteristics that affect accident rates are of interest. It is
clear that for both rural two-lane and multilane roadways AADT is the most important variable. However,
the factors of subsequent importance vary for each
M.G. Karlaftis, I. Golias / Accident Analysis and Pre6ention 34 (2002) 357–365
365
Table 3
Independent variable importance for crash rates (crashes normalized on highway section length and AADT)
Rural two-lane
Rural multilane
Variable
Relative importance (%)
Variable
Relative importance (%)
Lane width
Serviceability index
Pavement type
Friction
Left turn
100
89
62
22
16
Median width
Access control
Friction
Lane width
Serviceability index
Left turn
100
73
55
25
19
16
case. Looking closely at the results of accident rates
normalized on AADT (which cancels out the effect of
AADT), it can be generally inferred that LW and
pavement condition factors – SI, pavement type and
FR – are the most important variables affecting
crash rates for the two-lane case. The importance of
LW seems to increase with higher flows. On the
contrary, the importance of pavement condition
factors seems to increase with lower flows due to higher
speeds.
For rural multilane roads, with the effect of AADT
cancelled out, median width and access control are the
most important factors followed by pavement condition
factors. It is worth mentioning that the importance of
access control seems to increase with heavier traffic
that renders vehicle maneuvering arrangements critical,
while the existence of a median becomes more
important in low flow conditions. Although the importance of isolated variables differs for the two roadway
types it is obvious that ‘geometric design’ captured
through LW and access control and ‘pavement condition’ captured through FR, SI and pavement type are,
as expected, the two most important factors affecting
accident rates.
The methodology used in this paper also allows the
explicit prediction of accident rates for given highway
sections. As soon as the profile of a road section is
given, predictions regarding the expected accident rates
can be obtained. In essence, when the AADT, LW, SI
and FR of a road section are known, predictions can be
obtained. Further, the ‘if-then’ rules for obtaining these
predictions can be easily incorporated in an expert
system that can automate the accident rate prediction
effort. The work presented in this paper is part of the
larger effort to tackle the problem of accident occurrence on the world’s roadways. The extremely high cost
of highway accidents paid by societies makes highway
safety improvement maybe the most important objective of transportation engineering. This effort’s eventual
goal is to reduce injuries and fatalities due to highway
design and maintenance deficiencies.
References
Breiman, L., Friedman, J., Olshen, R., Stone, C., 1984. Classification
and Regression Trees, Wadsworth International Group, Belmont,
CA.
Hadi, M.A., Aruldhas, J., Chow, L.F., Wattleworth, J.A., 1993.
Estimating Safety Effects of Cross-Section Design for Various
Highway Types Using Negative Binomial Regression. Transportation Research Record, 1500, TRB, National Research Council,
169 – 177.
Jones, I.S., Whitfield, R.A., 1991. Predicting injury risk with new car
assessment program crashworthiness ratings. Accident Analysis
and Prevention 6 (20), 411 – 419.
Joshua, S.C., Garber, N.J., 1990. Estimating truck accident rate and
involvement using linear and poisson regression models. Transportation Planning and Technology 15, 41 – 58.
Karlaftis, M.G., Tarko, A., 1998. Heterogeneity considerations in
accident modeling. Accident Analysis and Prevention 30 (4),
425 – 433.
Lerman, S.R., Gonzales, S.L., 1980. Poisson regression analysis under
alternate sampling strategies. Transportation Science 14 (4), 346 –
364.
Miaou, S.P., Hu, P.S., Wright, T., Rathi, A.K., Davis, S.C., 1992.
Relationship between truck accidents and highway geometric
design: a poisson regression approach. Transportation Research
Record 1376, 10 – 18.
Mohamedshah, Y.M., Paniati, J.F., Hobeika, A.G., 1993. Truck
accident models for interstate and two-lane rural roads. Transportation Research Record 1407, 35 – 41.
Morgan, J.N., Sonquist, J.A., 1963. Problems in the analysis of
survey data, and a proposal. Journal of the American Statistical
Association 58, 415 – 434.
Ivan, J.N., O’Mara, P.J., 1997. Prediction of Traffic Accident Rates
Using Poisson Regression, Presented in the 1997. Transportation
Research Board Meeting, Washington, DC.
Tarko, A.P., Sinha, K.C., Farooq, O., 1996. A Methodology for
Identifying Highway Safety Problem Areas, Presented in the 1997.
Transportation Research Board Meeting, Washington, DC.
US DOT, 1969. National Highway Functional Classification Study
Manual. Federal Highway Administration, Washington, DC.
Washington, S., Wolf, J., 1996. Hierarchical Tree-Based versus Ordinary Least Squares Linear Regression Models: Theory and Example Applied to Trip Generation. Presented in the 1996.
Transportation Research Board Annual Meeting, Washington,
D.C.
Washington, S., Wolf, J., Guensler, R., 1996. A Binary Recursive
Partitioning Method for Modeling Hot-Stabilized Emissions from
Motor Vehicles. Presented in the 1996. Transportation Research
Board Annual Meeting, Washington, DC.
International
Journal of Civil Engineering
andOF
Technology
ISSN 0976 – 6308
(Print),
INTERNATIONAL
JOURNAL
CIVIL(IJCIET),
ENGINEERING
AND
ISSN 0976 – 6316(Online) Volume 4, Issue 4, July-August (2013), © IAEME
TECHNOLOGY (IJCIET)
ISSN 0976 – 6308 (Print)
ISSN 0976 – 6316(Online)
Volume 4, Issue 4, July-August (2013), pp. 146-162
© IAEME: www.iaeme.com/ijciet.asp
Journal Impact Factor (2013): 5.3277 (Calculated by GISI)
www.jifactor.com
IJCIET
© IAEME
THE INFLUENCE OF ROAD GEOMETRIC DESIGN ELEMENTS ON
HIGHWAY SAFETY
HameedAswad Mohammed
Civil Engineering Department – Anbar University- Iraq
ABSTRACT
Road safety is an issue of prime importance in all motorized countries. The road accident
results a serious social and economic problems. Studies focused on geometric design and safety aim
to improve highway design and to eliminate hazardous locations. The effects of design elements such
as horizontal and vertical curves, lane width, shoulder width, superelevation, median width, curve
radius, sight distance, etc. on safety have been studied. The relationship between geometric design
elements and accident rates is complex and not fully understood. Relatively little information is
available on relationships between geometric design elements and accident rates. Although it has
been clearly shown that very restrictive geometric elements such as very short sight distances or
sharp horizontal curve result a considerably higher accident rates and that certain combinations of
elements cause an unusually severe accident problem. In this paper, road geometric design elements
and characteristics are taken into consideration, and explanations are given on how to which extent
they affect highway safety. The relationship between safety and road geometric design are examined
through results of studies mad in different countries and it compares the results of studies in different
countries and summarizes current international knowledge of relationship between safety and the
principal non-intersection geometric design parameters. In general, there is broad international
agreement on these relationships.
Key words: Highway Safety, Geometric Design, Traffic Accident
1. INTRODUCTION
Geometric design elements play an important role in defining the traffic operational
efficiency of any roadway. Key geometric design elements that influence traffic operations include
number and width of lanes, the presence and widths of shoulders and highway medians, and the
horizontal and vertical alignment of the highway [1]. Generally speaking, any evaluation of road
safety, such as in the driving dynamic field, has been conducted more or less qualitatively. It is safe
to say, from a traffic safety point view, that no one is able to say with great certainty, or prove by
measure or number, where traffic accidents could occur or where accident black spots could develop
146
International Journal of Civil Engineering and Technology (IJCIET), ISSN 0976 – 6308 (Print),
ISSN 0976 – 6316(Online) Volume 4, Issue 4, July-August (2013), © IAEME
[2].However, everyone agrees that there exists a relationship between traffic safety and geometric
design consistency. By all means, alignment consistency represents a key issue in modern highway
geometric design. A consistency alignment would allow most drivers to operate safety at their
desired speed along the entire alignment. However, existing design speed-based alignment policies
permit the selection of a design speed that is less than the desired speeds of majority of drivers [2].
Much of the research in highway safety has focused on different factors which affect roadway safety.
The factors are categorized as traffic characteristics, road geometrics, road surface condition,
weather and human factors. Previous research has shown that geometric design inconsistencies,
operations (traffic mix, volume, and speed), environment, and driver behavior are the common
causes of accidents. Most of the studies have shown the influence of various geometric design
variables on the occurrence of accidents and have concluded that not all variables have the same
level of influence in all places [3]. From the relation of factor mentioned above, different researchers
have developed the relationship of roadway safety in terms of crash frequency and crash rates,
fatality and injury rates and the road elements, traffic characteristics, and pavement conditions. Many
of these previous studies investigated the relationship of crash rates or frequency in terms number of
lanes, lane width, presence of median, median width, type of median, shoulder width, access density,
speed limit, vertical grade, horizontal curvature, weather condition. The relationship between safety
on the highway and factors mentioned above is the primary focus in crash reduction and
predictions[3].
2. GEOMETRIC PARAMETERS AFFECTING ROAD SAFETY
An accident is always characterized by multiple causes. The alignment of road is an
important influence factor: dimension of radii, ratio of consecutive curves, dimension of vertical
curves and sight distance conditions. In many evaluation studies of safety effects of road design
elements it turns out the present poor capacity to explain accidentality phenomenon; in fact the main
causes of accident is behavior of driver, which is mainly influenced by his personality, skills, and
experience. Furthermore external impacts like weather conditions, road conditions, time of day, or
light conditions influence the driver behavior as well. It is out of question that analyzing accidents
and their dependence on technical values or human factors has always to consider these interactions.
The relation between accidents (all, property damage only, slight injuries, severe injuries, fatalies)
and road geometry is proved but it is also a question of the driving behavior, especially of the
velocity. Again and again investigations show that comparable curves (similar geometry) are
characterized by different accident occurrence. One reason could be a different driving behavior:
lower speeds are less critical than higher speeds in curves. Several studies, oriented to create
relationships between accidentality and independent variables, were obtained in a particular context;
so, in every other different conditions, weather conditions, user behavior, etc.) the influence of these
factors should be considered, e.g. calibration procedure. Summarized, accidents do not depend on
only one factor; accidents are caused rather by a combination of several factors [4].
3. THE RELATIONSHIP BETWEEN SPEED AND SAFETY
Speed is one of the major parameters in geometric design and safety is synonymous with
accident studies [5]. For example, Finch et al. [6] recently concluded that a reduction of 1.6 km/hr (1
mph) in the average speed reduces the incidence of injuries by about 5%. Also it is generally
accepted that there are substantial safety benefits from lower speed limits. For example, reducing
rural speed limits from 100 km/hr to 90 km/hr has been predicted to reduce casualties by about 11%
[7]. It is interesting to note that the relationship between the design speed and the speed limit is not
referred to in the geometric design standards of many countries [8].
147
International Journal of Civil Engineering and Technology (IJCIET), ISSN 0976 – 6308 (Print),
ISSN 0976 – 6316(Online) Volume 4, Issue 4, July-August (2013), © IAEME
4. EFFECTS OF CHARACTERISTICS OF ROAD GEOMETRICAL DESIGN ELEMENTS
ON TRAFFIC ACCIDENTS
Some of the primary geometric design elements that can affect on highway safety are
carriageway, grade, horizontal curvature, shoulder, median, vertical curve [9]. The relationship
between some characteristics of these elements and traffic accidents, including studies made in
different countries are classified into groups: Cross-section effects and Alignment effects.
4.1 Cross-Section Effects
The widths of the various cross section elements affect the capability of driver to perform
evasive maneuvers and determine the lateral clearances both between vehicles and between vehicles
and other road users [5]. In the existing literature are mentioned especially the following parameters:
4.1.1 Lane Width
Wider lanes are traditionally associated with higher operating speeds and increased safety.
The Highway Capacity Manual (HCM) documents that wider lanes for multilane highways result in
higher free-flow speeds [10]. On the other hand, very little has been found on the safety implications
of wider lanes. It is reasonable to assume that wider lanes may provide additional space to the driver
to correct potential mistakes and thus avoid crashes. However, a driver could be expected to adapt to
the available space, and the positive safety effects from the wider lanes may be offset by the higher
speeds [10].Generally, most studies agree that lower accident rates are attributed to wider lanes. But
it seems that there is an optimal lane width around 3.5m. Studies have also noted that approaches
should base on more parameters of the cross section, at least also on traffic volume [4]. However,
Hearne's results [11] suggested that there was a marginal increase in accident occurrence with an
increase in carriageway width. Hedman [12] noted that some results indicated a rather steep decrease
in accidents with increased width of 4m to 7m, but that little additional benefit is gained by widening
the carriageway beyond 7m. Zeger [13], Zeger/Council [14], and Mclean [15] have shown that width
of 3.4 3.7m show the lowest accident rates. This is supported by the NCHRP Repot 197 [16]
conclusion that there is little difference between the accident rate for 3.35m and a 3.65m lane width.
However, studies on low volume rural roads indicate that accidents continue to reduce for widths
greater than 3.65m, although at a lower rate [17]. TRB [18] pointed out lanes wider than 3.70m do
not contribute to a higher safety because they may result in unsafe maneuvers such as over taking
despite of oncoming traffic. Another reason is the higher speed on wider lanes which leads to more
accidents. Yagar and VanAerdo [19] found that the passage of a vehicle requires a minimum lane
width and that any additional width beyond this minimum allows one to drive faster and /or with a
greater measure and preception of safety. For lane widths from 3.3m to 3.8m, they reported that the
operating speed is decreased by approximately 5.7 km/hr for each 1m reduction in width of the road
[5]. Lamm et al.[20] found a significant d...
Purchase answer to see full
attachment