Accident Analysis and Prevention 42 (2010) 12401249

Contents lists available at ScienceDirect

Accident Analysis and Prevention

journal homepage: www.elsevier.com/locate/aap

Age, skill, and hazard perception in driving

Avinoam Borowsky , David Shinar, Tal Oron-Gilad
Ben Gurion University of the Negev, Beer Sheva, Israel

a r t i c l e

i n f o

Article history:
Received 20 October 2009
Received in revised form 28 January 2010
Accepted 1 February 2010
Keywords:
Hazard perception
Potential hazards
Driving experience
Older drivers
Novice drivers
Eye movements

a b s t r a c t
This study examined the effects of age and driving experience on the ability to detect hazards while
driving; namely, hazard perception. Studies have shown that young-inexperienced drivers are more
likely than experienced drivers to suffer from hazard perception deciencies. However, it remains to be
determined if this skill deteriorates with advancing age. Twenty-one young-inexperienced, 19 experienced, and 16 elderly drivers viewed six hazard perception movies while connected to an eye tracking
system and were requested to identify hazardous situations. Four movies embedded planned, highly hazardous, situations and the rest were used as control. Generally, experienced and older-experienced drivers
were equally procient at hazard detection and detected potentially hazardous events (e.g., approaching an intersection, pedestrians on curb) continuously whereas young-inexperienced drivers stopped
reporting on hazards that followed planned, highly hazardous situations. Moreover, while approaching T intersections older and experienced drivers xated more towards the merging road on the right
while young-inexperienced drivers xated straight ahead, paying less attention to potential vehicles on
the merging road. The study suggests that driving experience improves drivers awareness of potential
hazards and guides drivers eye movements to locations that might embed potential risks. Furthermore,
advanced age hardly affects older drivers ability to perceive hazards, and older drivers are at least partially
aware of their age-related limitations.
2010 Elsevier Ltd. All rights reserved.

1. Introduction
The ability to identify hazardous situations while driving is an
important skill, which enables the driver to overcome complex cognitive demands that the trafc environment dictates. This ability
is known as hazard perception. Mills et al. (1998) dened hazard
perception as the ability to read the road. Horswill and McKenna
(2004) added that hazard perception may be regarded as situation
awareness for hazardous situations. Within this framework, novice
drivers have difculties in forming a holistic understanding of their
current driving situation (e.g., Benda and Hoyos, 1983; Brown and
Groeger, 1988; Armsby et al., 1989).
According to Horswill and McKenna (2004), of the different
components of driving skill, only hazard perception has been correlated with trafc-accident involvement across a number of studies
(e.g., Peltz and Krupat, 1974; McKenna and Crick, 1991). Like any
skill, it is reasonable to assume that drivers improve their ability
to detect hazards as their experience grows. Research has shown
that novices are slower in detecting hazards, and that they often
detect fewer hazards than experienced drivers (e.g., Underwood et

Corresponding author at: Department of Industrial Engineering and Management, Ben Gurion University of the Negev, Ben-Gurion Avenue 1, P.O. Box 653, Beer
Sheva 84105, Israel. Tel.: +972 8 647 2247; fax: +972 8 647 2958.
E-mail address: Borowsky@bgu.ac.il (A. Borowsky).
0001-4575/$ see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.aap.2010.02.001

al., 2005). Chapman and Underwood (1998) found that experienced

drivers adapt their visual scanning patterns to different road situations, while novice drivers tend to use the same scanning patterns
for all road types, and their xation durations are usually longer.
It appears that feedback from the environment is the best trafc guide, and it improves hazard perception (e.g., Matthews and
Moran, 1986; Naatanen and Summala, 1976). Indeed, some studies
have shown that experienced drivers relate to environmental cues
which may indicate an upcoming hazard more than to a specic
maneuver or an actual discerned hazard. For example, Armsby et
al. (1989) asked participants to classify pictures of different trafc situations, and found that fog for example, which is a situation
where only potential hazards are found, was considered more hazardous among experienced drivers than among novice drivers. Finn
and Bragg (1986), who used a similar method, found that novice
drivers rated a pedestrian walking along the road (an actual hazard)
as more hazardous than experienced drivers but rated tailgating (a
potential hazard) as less hazardous than experienced drivers. Moreover, Benda and Hoyos (1983) found that, in general, novice drivers
pay attention to unimportant details within the environment. They
argued that experienced drivers have a holistic perception of the
trafc environment whereas young-inexperienced drivers assess
the level of hazardousness in a scene base on a single dimension.
Like young drivers, older drivers tend to be over involved in
vehicle crashes (e.g., Maycock et al., 1991) although HakamiesBlomqvist et al. (2005) showed that the increase in older drivers

A. Borowsky et al. / Accident Analysis and Prevention 42 (2010) 12401249

accident involvement is smaller than the increase in older drivers

presence in trafc. This nding may be partially because older
drivers who are aware of their age-related limitations (e.g., physical, visual, etc.) adopt a self-regulation policy or related strategies
such as driving slower, avoiding difcult conditions, reducing
night driving, and driving only in familiar areas (Oxley et al.,
2006).
In the context of hazard perception one should ask whether
older drivers, though they are very experienced drivers, suffer from
an age-related decline in their ability to detect hazards. It is not yet
clear whether age-related trafc risks include hazard perception
deciencies. According to Schacter (1996) age-related atrophy of
the frontal lobes affects recall but not recognition memory. Thus,
older people might experience difculties in recalling past actions
they should take but with the proper retrieval cues their performance may be as good as that of young people. Bolstad and Hess
(2000) suggested that most of the attentional problems attributed
to older adults typically occur in novel contexts. They added that
old people who are very experienced in driving depend on their
schema. It enables them to focus their attention on the appropriate
information and direct their attention according to a probabilistic
information attitude. The schemata they use are only minimally
affected by aging and in most cases remain intact. They added that
episodic memory functions may be moderated by environmental
support when external guidance such as trafc signs are present
allowing the driver to encode or retrieve information. Underwood
et al. (2005) showed that 6075 years old drivers who observed hazard perception movies scanned the road in a manner similar to that
of younger experienced drivers (3045), but detected more hazards
than younger experienced drivers. Thus, in the context of hazard
perception older drivers should demonstrate similar scanning patterns as experienced drivers and they should perceive hazards at
least as well as experienced drivers.
Finally in a recent study, we (Borowsky et al., 2009) had younginexperienced, experienced, and older-experienced drivers view
six movies of trafc scenarios and classify them into an arbitrary
number of groups according to the similarity in their hazardous
situations. We found that both older-experienced and experienced
drivers tended to classify the movies according to similarities in the
trafc environment (e.g., grouping all movies that included residential trafc environments) whereas young-inexperienced drivers
tended to classify the movies according to similarities in the hazard instigators (e.g., grouping all movies that included pedestrians
crossing the road). These ndings suggest that experienced adults
and old drivers perceive hazards in a similar holistic manner and
that possibly this skill remains intact among older drivers.
The present study was designed to examine the effects of
age and driving experience on the ability to detect hazards
while viewing videos of driving scenes, namely, hazard perception (HP) in driving. Many studies have used short trafc scene
movies, presented on a computer screen, to examine differences
in response times to hazardous events between novice and experienced drivers (e.g., Chapman and Underwood, 1998; Sagberg and
Bjrnskau, 2006; Crundall et al., 1999; Wallis and Horswill, 2007;
see also Horswill and McKenna, 2004 for a review). Evidence from
these studies has yielded conicting results. Some studies have
shown that experienced drivers detect hazards sooner than younginexperienced drivers (e.g., Horswill and McKenna, 2004) whereas
others did not nd any differences (e.g., Sagberg and Bjrnskau,
2006; Crundall et al., 1999; Chapman and Underwood, 1998). Thus,
further investigation is required to determine the source of discrepancy among studies.
Six hazard perception movies videotaped from the cab of a
moving car were presented to three groups of drivers: younginexperienced, experienced, and older drivers (over 65 years old).
Participants were instructed to observe the movies (while con-

1241

nected to an eye tracking system) and to press a button each time

they detected a hazardous situation.
The study had two main hypotheses. First, older-experienced
and experienced drivers will be more sensitive to potential hazards
and will press the hazard button more often than young drivers.
Second, older and experienced drivers will direct their xations
towards hazardous environmental cues (e.g., intersections) more
often than young-inexperienced drivers who will scan the environment in a more random fashion.
2. Method
2.1. Participants
Fifty-six participants, 21 young drivers (1718 years old, with
an average of 2.7 months of driving experience), 19 experienced
drivers (2230, with an average of 7.3 years of driving experience)
and 16 older drivers (6572, with an average of 37.5 years of driving experience) participated in this study. Participants received
monetary compensation for their participation. All participants had
uncorrected Snellen visual acuity of 6/9 (20/30) or better, and normal color vision. The older drivers were mostly retired people who
lived in proximity to the university. The experienced drivers were
students in Ben Gurion University and the young-inexperienced
drivers were recruited through driving schools in the city of Beer
Sheva.
2.2. Apparatus and stimuli
2.2.1. Screen and setup
A 19 in. LCD screen with 1024 768 pixels, connected to a Pentium 4 PC was used in order to display the movies. Participants sat
at a distance of 70 cm from the LCD, which provided them with a
visual eld of 22 vertically and 26 horizontally.
2.2.2. Eye tracking and xations calculation algorithm
The visual scan pattern was recorded with an eye tracking system (ETS) (Applied System Laboratories, Model 504), sampling
the visual gaze at 50 Hz, with a nominal accuracy of 1 . Fixations analysis was made using the dispersion methodology applied
by Gitelman (2002) via ILAB. The dispersion algorithm has three
parameters: minimum xation duration, minimum dispersion considered a xation ( ), and maximum consecutive sample loss.
These parameters were set to 100 ms, 1 visual degree, and innity
(default), respectively.
2.2.3. Hazard perception movies
Six driving scene movies were shown to the participants. Four
of the movies included planned-staged hazardous situations and
the other two were control movies with no planned hazardous situations. All movies were lmed in an urban area in the center of
Israel and from a drivers perspective. The duration of each movie
was on average 30 s. The six driving scenarios included:
Movie 1 (M1). This movie was taken on a high-speed urban road
with a relatively open eld of view. The camera was located in a
car that tailgated a red car. At a certain point (after 21 s) the red
car braked and swerved unexpectedly toward a parking space on
the right without signaling.
Movie 2 (M2). This control movie was similar to Movie 1 in terms
of roadway characteristics but had no lead car. At a certain point
(after 20 s) the participants car approached a signalized intersection and stops at the red light.
Movie 3 (M3). This movie was taken in a densely populated residential one-way street with cars parked on both sides. The scenario

1242

A. Borowsky et al. / Accident Analysis and Prevention 42 (2010) 12401249

Fig. 1. Planned hazardous situations. The white car symbolizes the participants car location prior to the hazardous situation. Movie 1 (M1) was lmed in an urban area and
M3, M4, and M5 were lmed in a residential area. The planned hazards included: a car-following episode in which the leading car suddenly brakes and turns right towards a
free parking space (M1), a roller-blader enters the road in front of the participants car (M3), a bicyclist entering the participants car path to avoid a parked car whose driver
opened its door (M4), and a car-following episode in which the leading car had to suddenly brake as a result of a third car who entered from the right side of the intersection
(M5).

included a person riding on roller blades on the right curb pavement partially hidden by parked cars, and trees. At a specic point
(after 20 s) the roller-blades rider moved into the road because
his way was blocked by a car that was partially parked on the
pavement. This scenario was planned to examine drivers ability
to integrate elements in the environment to achieve good situation
awareness and expect this movement.
Movie 4 (M4). This movie was similar to M3 in terms of roadway
characteristics. In this scenario a bicyclist is riding on the curved
road. The driver in the moving car can observe the bicyclist right
after the exiting spiral of the roads curvature. A car that straddled
the right lane and the curb was stopped with its brake lights on,
indicating that the driver was still in the car. At a certain point
(after 9 s) the driver of the parked car opened his door and the
bicyclist who was riding just ahead of the moving car swerved to
the center of the road to avoid striking the door. Later in the same
movie an unplanned event took place when the bicyclist had to
avoid another car parked half on the curb and half on the road.
Movie 5 (M5). This movie was similar to M3 and M4 in terms of
the roadway characteristics (i.e., driving in a narrow residential
street). However, the planned hazardous situation in this movie

involved following a white car while parked cars obstructed the

eld of view on both sides of the road. When the lead car reached
a crossing, a third car, approaching the intersection from the right,
made a right turn into the drivers road right in front of the lead
car. As a result the lead car had to brake suddenly. This movie
was designed to examine whether drivers anticipate the sudden
braking of the lead car as a result of the reckless driving of other
road users.
Movie 6 (M6). This control movie was similar to Movies 3, 4, and
5 in terms of the roadway characteristics except that it had no
planned hazardous situations. Thus, this movie was similar to M2
in the absence of planned hazardous events. The car drove straight
and then (after 14 s) made a right turn. Then it continued to drive
straight again and (after 23 s) made a second right turn.
Fig. 1 presents all hazardous planned events and Fig. 2 presents
the two control movies (M2 and M6).
2.2.4. Training movies
Two movies lmed in England that included hazard perception situations were used for training the participants.

Fig. 2. Control movies. The white car symbolizes the participants car initial location. M2 and M6 served as control movies for urban (M1) and residential (M3M5) movies
and did not include any planned hazards.

A. Borowsky et al. / Accident Analysis and Prevention 42 (2010) 12401249

2.2.5. Movie presentation program and controls

To synchronize the movies with eye xation patterns, all movies
were converted into a sequence of 760 576 pixels single frames. A
C++ program was used to send data to the ETS that indicated each
frame switch. Each button press (utilizing a special button) was
marked and linked the subjects identication of a hazard situation
with the relevant movie frame.
2.2.6. Hazard denition
Hazard was dened for all participants according to Haworth
et al. (2001, p. 3) denition: Any object, situation, occurrence or
combination of these that introduce the possibility of the individual road user experiencing harm. Hazards may be obstructions in
the roadway, a slippery road surface, merging trafc, weather conditions, distractions, a defective vehicle, or any number of other
circumstances. Harm may include damage to ones vehicle, injury to
oneself, damage to anothers property, or injury to another person.
2.3. Procedure
Participants were connected to the ETS after reading a
short description of the experiment. The experimental procedure
included a training session followed by the main task. Before the
training session participants gaze was calibrated and they read the
instructions. During the training session they were instructed to
observe the movie as if they were drivers driving down the road,
and to respond (by pressing the designated button) each time they
detected a hazardous situation. At the end of each movie the participants had to identify the hazard that triggered their response.
At the end of the training session, the experimenter veried that
the participant understood the experimental task. The participant
then read the hazard denition once more, and then viewed the
six hazard perception movies. The order of the movies was randomized for each participant. This procedure resembled Chapman
and Underwoods (1998) experimental procedure. During the video
presentations the participants eye movements were recorded and
synchronized with the button presses.
3. Results
3.1. General comparison among movies
The rst goal was to investigate whether movies M1, M3, M4,
and M5 were indeed perceived as more hazardous than the control
movies M2 and M6. In the framework of GLM (general linear models), a repeated measures two-way ANOVA on movie type (planned
hazard vs. control) and driver type with unbalanced observations
conrmed that the average number of button presses was signicantly higher for the four planned hazardous movies than for the
two control ones [1.65 (SD = 0.07) vs. 1.10 (SD = 0.09), respectively,
F1, 53 = 55, p < 0.001]. The interaction between driver groups and
movie type and the driver type main effect were not statistically
signicant.
The second goal was to examine whether xation durations in
the busier urban areas (M1 and M2) were longer than in the residential areas (M3M6). This hypothesis was based on Chapman
and Underwoods (1998) nding that residential areas, which are
denser in information than rural areas, produced larger amounts of
shorter xations compared to rural areas. This analysis was applied
to 15 young-inexperienced drivers, 15 experienced drivers and 9
old drivers. The other drivers had missing data in at least one of
the movies were removed from the analysis. Within the framework of GLM A similar repeated measures two-way ANOVA on
movie type and driver type showed only a signicant main effect for
road type, with average xation durations of 229 ms (SD = 31.5) for

1243

the urban scenes and 209 ms (SD = 25.8) for the residential scenes
(F1, 36 = 32.5, p < 0.001).
3.2. Within-movies events analysis
An examination of how age and skill affect drivers ability
to detect hazardous events was conducted. Participants detected
and responded to many more hazardous events than the initially
planned events. All reported events are presented in Table 1.
As can be seen from Table 1, in addition to the four planned
hazardous events there were fteen additional unplanned events.
Notably, the average number of responses to each planned event
(see highlighted cells in Table 1) in some movies exceeded 1.0. This
is because some drivers responded to a planned event more than
once, partitioning a single planned event into two or more distinct
events (e.g., roller-blader on curb followed by roller-blader on
road). In general, the levels of agreement among drivers on the
planned events, in which the hazards materialized, were high (indicated by the relatively high response percentages to these events
by all three groups of drivers). The large number of unplanned
events presented in Table 1 suggested that it would be inappropriate to analyze only the planned events as such an analysis
might present an incomplete picture regarding experienced-based
hazard perception differences. However, because most unplanned
events were noted by a relatively small number of participants
it was inappropriate to evaluate response time (RT) differences.
Notably, some of the inconsistencies reported in earlier studies with respect to reaction time differences/similarities between
experienced and young-inexperienced drivers might be attributed
to the type of events added to the RT analysis. It was therefore decided to analyze all events (planned and unplanned)
broadly according to a procedure presented in the following two
paragraphs.
3.2.1. Event characteristics and their analysis procedure
Four dependent measures were available for each event:
response time, response sensitivity (driver-group response distribution, see rows M1 Res through M6 Res in Table 1), the verbal
description of the hazard instigator, and the eye movements pattern. In order to systematically decide which of the rst three
dependent measures was most appropriate to use in the analysis of each event (eye movements patterns were examined for all
analyzed events), two denitions were made (event support and
response sensitivity) and the following rule of thumb was established accordingly.
Event support (%). An event was considered as having high support if 30% or more of all respondents identied it as hazardous.
This cut-off was chosen arbitrarily but suggests that a meaningful number of participants considered the event as hazardous. It
might be argued that other cut-offs may have been chosen. The
cut-off chosen here, however, merely points out that it is important to distinguish between high and low support events and that
they all should be included in the analysis. Similarly, an event was
considered as having low support if less than 30% of the respondents
identied it as hazardous [see Table 1 rows M1M6 (right-hand side
data in parenthesis)]. For example, M5 E1 (Event 1 in Movie 5) had
low support because only 11% (6/56) of the participants considered
this event as hazardous.
Response sensitivity (%) within driver-groups support. For each
event, a group-specic support score was calculated using the percent of drivers in each group who responded to that event (see
Table 1 rows M1 ResM6 Res). For example, event M5 E1 was
noted by 1/21, 3/19, 2/16 of the young, experienced, and olderexperienced drivers, respectively, corresponding to 4.8%, 15.8%, and
12.5% of support.

1244

A. Borowsky et al. / Accident Analysis and Prevention 42 (2010) 12401249

Table 1
Description of all events noted by 21 young, 19 experienced and 16 old drivers in each movie, with the number of observed responses by each age group below in the response
row.
Movie

Sequence of events detected in each movie and the proportion responses to it in each group
E1

E2

E3

E4

M1

First parked truck A

truck occluded the
participants car FOV
on the right curb (6,
11%)

Second parked truck

Another truck
occluded the
participants car FOV
(5, 9%)

Intersection The
participants car
approached a signaled
intersection (11, 20%)

M1 Res
M2

Y 2; E 2; O 2
Passing lane The
participants car passed
from the right to the
left lane (8, 14%)

M2 Res
M3

Y 1; E 4; O 3/16
Old lady An old
lady was walking on
the road on the right
(12, 21%)

Y 2; E 1; O 2
Intersection The
participants car
approached a
signalized intersection
(33, 59%)
Y 10; E 14; O 9
Roller on curb A
Roller-blader was
skating on the right
curb (7, 13%)

*Red car braked

The lead car braked
suddenly and diverted
to the right towards a
free parking space
without signaling (50,
82%)
Y 17; E 17; O 12

Intersection The
participants car
approached a T
intersection (10, 18%)

M3 Res
M4

Y 4; E 3; O 5
Bicyclist detection
The participants car
entered a curve when
a bicyclist was seen
riding on the road in
front of the car (27,
48%)

Y 0; E 4;O 3
*Cars Door opens
The bicyclist had to
avoid a parked car (by
diverting to the center
of the lane whose
driver opened the
door without looking
(43, 77%)

M4 Res
M5

Y 11; E 11; O; 5
First intersection
The participants car
crossed the rst
intersection (6, 11%)

M5 Res
M6

Y 1;E 3; O 2
First intersection
The participants car
approached a T
intersection (44, 79%)

M6 Res

Y 17; E 14; O 13

Y 18; E 11; O 14
*White car brakes
The participants car
followed a car that
had braked before a T
intersection due to a
third car that made a
careless right turn
(54, 96%)
Y 21; E 19; O 14
Car in front When
turning right in the
previous intersection a
car that was driving on
the opposite way
appeared in front of the
participants car (10,
18%)
Y 3/21; E 4; O 3

*Roller on road
The Roller-blader
entered the road from
between two parked
vehicles in front of
the participants
vehicle (49, 88%)
Y 20; E 14; O 15
Bicyclist avoids a
parked car The
bicyclist had to avoid
a parked car (by
diverting to the center
of the lane) on the
right side of the road
with two wheels on
the curb and two on
the road (35, 63%)
Y 14; E 13; O 8
White car turns
right The white
lead car signaled and
made a right turn into
the intersection (15,
27%)

E5

Y 1; E 6; O 4

Y 0; E 5; O 5
Children on
Pavement Two
children were walking
on the edge of the right
side of the curb due to
a parked car blocking
their path. They didnt
enter the road (5, 9%)

Y 0; E 2; O 3
Second intersection
The participants car
made a right turn in a T
intersection (27, 48%)

Y 10;E 2; O 3
Second intersection
The participants car
approached a second
intersection (24, 43%)

Y 8;E 12; O 7
After second
intersection The
participants car made
a wide right turn at the
intersection into a
narrow carriageway
street (4, 7%)

Y 10; E 9; O 5

Y 1; E 1; O 2

Parked cars
Parked cars on
both sides of the
curb occluded the
participants line of
sight (5, 9%)

Y 1; E 2; O 2

Note: The data in the response rows (e.g., M1 Res) presents the distribution of responses across 21 young, 19 experienced and 16 older drivers, respectively. For example,
M4 Res (Event 1), indicates that 11out of 21 young participants responded to this event (i.e., 52.4% of the young drivers group), similarly, 11 out of 19 experienced and 5
out of 16 older drivers responded. Planned events are in those highlighted in the gray, bolded cells. For each sequence of planned events one cell is marked with an asterisk
that indicates the time where the majority of participants responded in that sequence. The numbers in parenthesis indicate the total number of participants who responded
when the hazard appeared and the overall percent of participants who responded to it, respectively.

All high support events (support 30%) both planned and

unplanned were analyzed by all four dependent measures. low
support events were treated somewhat differently. First, these
events were not analyzed in terms of their elicited response times
because of the small number of respondents who responded to
these events. Second, as previously mentioned, low support
events may be of signicance for specic age groups. For example,
Event 4 in M3 shows a total support of 18% which means it is a low
support event. However, the driver-group response sensitivity
shows 26% and 31% for experienced and older drivers, respectively,
and 0% for young drivers. The rule of thumb was therefore to ana-

lyze low support events by means of response sensitivity if one

or two of the drivers groups (but not all three) had response sensitivity of 20% or more. Again, the 20% cut-off was chosen arbitrarily
to convey (1) that a substantial number of participants in a specic
group noted this event and (2) that the event cannot be considered
a false alarm.
3.2.2. Statistical analysis of dependent measurements
The events (planned and unplanned) were divided into three
categories: pedestrians car-following, and intersections in either
urban or residential area. Car-following category included only high

A. Borowsky et al. / Accident Analysis and Prevention 42 (2010) 12401249

1245

Table 2
Type of analysis for low and high support events.
Measure

Used for

Description

Example

Analysis used

Average response
latency (ARL)

High support events

Suppose the experienced

drivers, responded on average
in frame 500 and the young
drivers responded on average
in frame 600. Then the ARL was
100 frames or 4 seconds (100
frames at the rate of 25 fps).

A one-way ANOVA
with ARL as the
dependent variable
and driver group as the
independent one.

Response sensitivity
(%) driver-group
support.

All events however

statistical differences
were found only in
low support events
All events, however,
only those with
signicant differences
are presented.

The baseline response time is set

according to the group who on average
responded rst to a specic event.
Average response latency (ARL) is
calculated by subtracting the average
response time of the two other groups
from the baseline.
Examines whether one or two groups
of drivers considered a specic event
as less or more hazardous than each
one of the other groups of drivers
Each participant was asked to write the
hazard instigator (cause) of each
hazardous event. These hazard
instigators were coded and classied
into the three categories of hazard
instigators. Then, each participants
explanation was classied into one of
these hazard instigator categories.
Examines whether one or two groups
of drivers described the hazard
instigator differently. Obviously, this
comparison was made only among
drivers who responded to the relevant
events.
A xation was dened as a period of
time where the eye rested for more
than 100ms in a specic location
within a square of 1 visual arc degree
square (using ILAB software; Gitelman,
2002). The dependent variable was the
average horizontal location of each
participants xations relative to the
center point

Hazard instigator
verbal description

Eye movements
(xations) analysis

Low support events.

support events whereas pedestrians and intersections categories

included both high and low support events. The type of analysis for
all events is described in Table 2.
3.2.2.1. Pedestrians. The pedestrians category included three high
support planned events (M3 E3, M4 E2, and M4 E3), and one low
support event (M4 E4). Roadway characteristics were similar for
all events and included driving in residential areas. The three high
support events showed no statistical difference among the groups
in any of the dependent measures. The low support event included
two children walking on the right curb. Though all drivers xated on these children only 3 out of the 16 older drivers, and 2
out of the 19 experienced drivers responded to this event. Furthermore, no young driver thought this event was hazardous. 2
analysis revealed that the older drivers were marginally signicant
different from the young drivers while experienced drivers were
not signicantly different from any other group of drivers (old (O)
vs. experienced (E), 21 = 0.48, p = NS; young (Y) vs. O, 21 = 4.28,
p = 0.03 (p-adjusted = 0.115); E vs. Y, 21 = 2.33, p = NS).
3.2.2.2. Car-following. The car-following category included two
high support planned events M1 E3 and M5 E2. These events were
different in their roadway characteristics (urban vs. residential,
respectively).
For M5 E2 there were no statistical differences among the
groups in ARL, response sensitivity, or the verbal description of
the hazard instigator. A high number of participants from all three
groups responded to this event (21/21, 19/19, and 14/16 for young,
experienced, and older drivers, respectively). In contrast there were
signicant differences among the groups in the visual scan patterns.
Because this hazardous event also occurred in an intersection, the
results of the eye movements analyses will be presented in the
context of the intersections events.

2 analysis

For example, the sudden

breaking and swerving of the
lead car toward the right in M1
produced different hazard
instigator denitions. I should
have kept distance from the
lead car or The lead car did
not signal when it swerved
right. These descriptions were
coded as keep distance and
No signaling, respectively.

2 analysis

One-way ANOVA with

driver group as the
independent variable.

For M1 E3 there was a signicant difference in both ARL and

in the verbal description but not in response sensitivity. Older
drivers (12/16) were on average slower than young (17/21) and
experienced (17/19) drivers to respond (F2, 40 = 14.147, p < 0.001).
No difference was found between the experienced and younginexperienced drivers. The older drivers responded on average
2.15 s (SD = 1) later than the experienced drivers and 1.91(1) s later
than the young-inexperienced drivers. Fig. 3 illustrates the average
point in time when the older-experienced drivers responded to the
hazard (right picture) and the averaged point in time when younginexperienced and experienced drivers responded to the hazard
(left picture). Note that older drivers responded later, when the
vehicle had already diverted toward the parking space on the right
side of the curb, whereas the other two driver groups responded
when the car started braking.
Older drivers also differed from the young and experienced
drivers in their verbal description of the hazard instigator. The
majority of young and experienced drivers described the hazard
instigator as a sudden brake of the lead car (16/17 and 14/17,
respectively) while the majority of older drivers (10/12) claimed
that the lead car did not signal when it turned right towards the
free parking space. 2 analysis (1 if a participant mentioned the
lack of signaling and 0 otherwise) showed that older drivers were
more likely than young and experienced drivers to mention signaling as the problem (O vs. E, 21 = 12.27, p < 0.01; Y vs. O, 21 = 17.92,
p < 0.01; E vs. Y, 21 = 1.13, p > 0.2). Adjusted p values using Holms
(1979) method remained signicant.
3.2.2.3. Intersections. The intersections category included ve
unplanned events: M1 E4, M2 E2, M3 E4, M5 E4, M6 E1, M6 E3. All
intersection events in the control movies (M2 and M6) were High
support events and did not show any response sensitivity or verbal
description differences among the groups. In all of these events

1246

A. Borowsky et al. / Accident Analysis and Prevention 42 (2010) 12401249

Fig. 3. Drivers response to the braking car episode. Young and experienced drivers responded approximately 2 s earlier (left picture) than older drivers (right picture).

except M5 E4 older drivers responded signicantly later (closer

to the intersection) than experienced drivers. Young drivers were
not signicantly different from either one of the other two groups.
This pattern may indicate that experienced drivers responded rst,
then young drivers, and lastly the older drivers. All low support
intersection events showed that older-experienced and experienced drivers tended to respond more than young drivers. No
verbal description differences were found among drivers. M5 E4
differed from the other intersection events as response sensitivity
for this event was exceptionally high for all drivers and no signicant differences were found amongst groups. Table 3 summarizes
the statistical analysis of response time (average response latency
ARL) and response sensitivity for intersection events.
As mentioned in the car-following events, M5 E2 was categorized by the participants as a car-following event even though
it included a T intersection, most likely because the prominent
hazard instigator in this planned event was the brake of the lead
vehicle right in front of the intersection. The scanning patterns in
M5 E2 and in a similar event (M3 E4) showed that both experienced and older drivers xated towards the right in the direction of
the merging road whereas young drivers gazed directly forward. As
previously mentioned, the number of participants included in the
analysis varied such that only participants with less than 20% of eyedata loss in the examined movie were included. For both events,
a one-way ANOVA revealed that the mean horizontal xations
location for both experienced and older drivers was signicantly
more toward the right side of the road than that of the young
drivers. No signicant difference was found between experienced
and older drivers (M5 E2: F2, 41 = 8.07, p < 0.001; M3 E4: F2, 40 = 6.48,
p < 0.004). Fig. 4 contains the xation patterns of the three driver
groups for the two events and the mean horizontal xation loca-

tion (black vertical line). The mean horizontal xation location on

M5 E2 was 462.6 pixels (visual degrees from the center of screen
(VDCS) = 1.24 (a negative value indicates left from the center of the
screen), 512.8 pixels (VDCS = 0.03), and 523.5 pixels (VDCS = 0.3) for
young, experienced, and older drivers, respectively. The mean horizontal xation location on M3 E4 was 526 pixels (VDCS = 0.35),
561 pixels (VDCR = 1.24), and 589.1 pixels (VDCS = 1.95) for young,
experienced, and older drivers, respectively.
3.2.3. Temporal analysis of response sensitivity to unplanned
events
This analysis examined differences in attitude towards
unplanned hazardous situations among drivers. Specically, we
examined whether there were differences among the groups
in relation to perceiving potential hazards before and after a
planned event materialized. Unplanned events which preceded
the planned events were called before planned events. Similarly, unplanned events which followed the planned events were
called after planned events. In total there were 4 low support
unplanned events (M1 E1, M1 E2, M3 E1, and M5 E1) preceding
the planned events in Movies M1, M3, M4 and M5 (Movies M2
and M6 were not included in this analysis because they did not
include planned events) and 5 low support unplanned events
(M1 E4, M3 E4, M4 E4, M4 E5, M5 E4) following the planned
events in Movies M1, M3, M4 and M5. A two-way ANOVA with
repeated measures with driver group (3) timing (2; before and
after planned events) as the explanatory variables was conducted on the average response ratio. Two signicant main
effects emerged, driver group (F2, 53 = 4.078, p < 0.023) and Timing
(before planned events vs. after planned events) (F1, 53 = 12.405,
p < 0.001) as well as a second-order interaction between driver

Table 3
Statistical analysis for average response latency (ARL) and response sensitivity in intersection events.
Event/support

Average response latency (ARL) [s]

Response sensitivity

Respondents [%]

M1 E4/low

Not examined

O vs. E, 21 = 0.18, p = NS

Y vs. O, 21 = 3.18, p = NS
E vs. Y, 21 = 4.97, p = 0.026 (p* < 0.077)

O 4/16, 25%
E 6/19, 31.5%
Y 1/21, 5%

M2 E2/high

F2,

M3 E4/low

Not examined

O vs. E, 21 = 0.1, p = NS

Y vs. O, 21 = 7.6, p < 0.01 (p* = 0.017)
E vs. Y, 21 = 6.3, p < 0.025 (p* = 0.024)

M5 E4/high

No signicant differences

No signicant difference

M6 E1/high

F2, 37 = 4.753, p < 0.015

O (1.6) Y (1)

No signicant difference

M6 E3/high

F2, 20 = 3.808, p < 0.04

O (1.5)Y (0.6)

No signicant difference

23

= 3.522, p < 0.046

O (2.5)Y (1.5)

No signicant difference
O 5/16, 31%
E 5/19, 26%
Y 0/21, 0%

Note: Average response latency (ARL) represents the response time interval of each group with respect to the experienced drivers (that were always the earliest to respond).
Not examined means that low support events were not analyzed by means of response time. All adjusted p values (p*) for multiple comparisons were using Holms (1979)
method.

A. Borowsky et al. / Accident Analysis and Prevention 42 (2010) 12401249

1247

Fig. 4. Visual scanning patterns at T intersections. M5 E2 (upper panel) includes xations from 16 young drivers (left), 17 experienced drivers (middle) and 11 older drivers
(right) from frames 160178, superimposed on frame 170. M3 E4 (Lower panel) includes xations from 16 young drivers (left), 16 experienced drivers (middle) and 11 older
drivers (right) from frames 605630, superimposed on frame 605.

Table 4
Percent of drivers responding to unplanned events in each group as a function of its
timing (before vs. after planned event).
Driver group

Before (4 events) (%)

After (5 events) (%)

Experienced
Young
Older

11.8
10.7
17.2

27.4
9.5
26.3

group and timing (F2, 53 = 3.085, p < 0.054). Table 4 presents the percentage of drivers from each group who responded to the before
and after the planned events.
From Table 4 it can be seen that the response rate was relatively
low and similar for all three driver groups with regard to events that
appeared before planned events, however, older and experienced
drivers responded nearly three times more than young drivers to
events that appeared after planned events.
4. Discussion
4.1. Driving experience and hazard perception
The present study showed that regardless of the measure used
experienced and older drivers are more sensitive to potential
hazards than young-inexperienced drivers. Potentially hazardous
situations, were those unplanned events in which a hazard did
not materialize (e.g., intersections, and pedestrians walking on
the curb) and had low support. The results showed no differences in the driver groups attitude toward the planned events
(actual hazards). The level of agreement within and between the
drivers groups (response sensitivity) concerning the hazardousness of planned events was high (as shown in Table 1). The most
salient characteristic of the planned events was that the hazard
instigator (e.g., a roller-blader entering the road) posed an imminent threat because it interfered with the drivers maneuverability
and therefore forced him to act in order to avoid a crash.

The ability to read the environment and to anticipate a

possible hazard which may (or may not) materialize is experiencerelated. Young drivers who lack the appropriate experience-based
knowledge (e.g., Endsley, 1995; Logan, 1985) have difculties in
anticipating potentially hazardous situations because they have
not yet accumulated enough feedback (experience) from similar situations in which the potential hazard materialized. Hazards
must be salient and pose an imminent and obvious threat before
young-inexperienced drivers decide to take action. Older and experienced drivers, who have much more driving experience as well
as more established experience-based knowledge, were able to
indicate (anticipate) potentially hazardous situations even when
the probability that they will materialize was low. In support of
this notion we found that both experienced and older-experienced
drivers xated on potentially hazardous situations such as a merging road at a T intersection even when no salient hazard was
visible. These results are consistent with those of Pollatsek et
al. (2006) who found that novice drivers are not searching for
hazards in areas along the road where potential hazards are
obscured.
Moreover, young drivers were less likely to detect potentially hazardous situations when they occurred after a planned
(actual) event while, both older-experienced and experienced
drivers continued to search for hazards. This analysis might reinforce the lack of sensitivity among young-inexperienced toward
potential hazards. That is, at rst, while observing the videos younginexperienced drivers pressed the hazard button more frequently.
Then, once the actual hazard occurred they may have used its occurrence as a baseline for decide which situations are hazardous and
which are not. The lack of a continuous search for hazards among
young-inexperienced indicates that although potential hazards
are always present in the driving environment (and experienced
drivers are constantly searching for them) novice drivers lack this
situation awareness and rely primarily on prominent events in the
trafc environment.

1248

A. Borowsky et al. / Accident Analysis and Prevention 42 (2010) 12401249

As noted above, in a previous study (Borowsky et al., 2009) experienced drivers classied movies according to the similarity in their
trafc environment and showed a much more coherent and holistic attitude toward hazards than the inexperienced-young drivers
who classied the movies according to the similarity in their hazard instigator. The classication patterns may be partially explained
by the present results. In this study experienced drivers, regardless
of age, detected more potential hazards than young-inexperienced
drivers. It can be argued that paying attention to potential hazards
means that one is sensitive to the trafc environment and to the
hazards it may create.
4.2. Age and hazard perception
Older drivers in the present research had more than 37 years of
driving experience on average. Consistent with Bolstad and Hess
(2000), and Underwood et al. (2005) our older and experienced
drivers detected more potential hazards than young drivers. Older
drivers ability to perceive hazardous situations was not signicantly affected by age-related problems. This nding is consistent
with the results of some other studies (e.g., Bolstad and Hess, 2000).
Event M1 E3 demonstrated that older drivers responded on average 2.15 s later than experienced and young drivers to events, and
they also described the hazard instigator as being caused by another
driver (e.g., lack of signaling of the lead vehicle) and not by assuming own responsibility (as I had to brake) like the majority in
the other two groups. However, this nding needs to be further
examined on a wider array of events.
In approaching intersections older drivers were signicantly
slower than experienced drivers to respond (except in M5 E4).
Young drivers usually responded, on average, later than experienced drivers and sooner than older drivers but were not
signicantly different from either one of these groups. In fact, for
three intersection episodes older drivers responded on the average 1.5 s later or 30 m closer to the intersection than experienced
drivers. This highlights the difference between perceptual skills
that seem to remain intact and motor skills that seem to deteriorate with age. This nding needs to be further examined on the
road or in a driving simulator. Consistent with the car-following
episode M1 E3, it also seemed that older drivers relied more on
signage provided by other road users (e.g., signaling) in order to
decrease elements of surprise.
4.3. Response time and response sensitivity measurements
The present study suggested a broader approach toward analyzing drivers responses to hazardous events. In order to better understand hazard perception skill differences between experienced and
young-inexperienced drivers all hazardous events should be analyzed according to a certain cut-off criterion. That is, actual hazards
or prominent events that produce high rate of responses can and
should be analyzed in terms of response times. However, potentially hazardous situations, that generally produce fewer responses
should not be excluded from the analysis but rather be analyzed in
terms of response sensitivity, i.e., whether or not a specic group
of drivers decided to respond to that event. Because our approach
is new it is not argued that the cut-off criteria chosen are the
most appropriate to distinguish between high and low support
hazardous events but rather that such a distinction should be conducted if one wishes to get a more holistic picture of HP differences
between young-inexperienced and experienced drivers.
5. Conclusions and recommendations
Detecting environmental cues, which may indicate upon an
upcoming danger, is key in hazard perception. Experienced and

older-experienced drivers were able to detect more cues and thus

to perceive more potentially hazardous situations than novices.
Continuous feedback from the trafc environment guides the experienced drivers search for potential hazards which may or may not
materialize. This sensitivity to potential hazards may be viewed
as the holistic perception of the trafc environment which younginexperienced drivers lack. With an impoverished knowledge base
young drivers tend to respond primarily to actual hazards, a
response which may be too late or too infrequent to generate the
necessary holistic perception of the trafc environment. As Fisher
and his associates (e.g., Pollatsek et al., 2006; Fisher et al., 2006) suggest, practice with an array of hazard perception slides or movies
for the purpose of training young drivers may improve their hazard
perception abilities. Furthermore, assessment of hazard perception
should address the number of cues or potential hazards that young
drivers detect in potentially hazardous situations and not merely
their ability to respond to actual hazards.
With respect to the older drivers population, we found that in
general their hazard perception abilities remained intact. However,
they relied heavily on signs and signals (i.e., signals from other
vehicles) in order to cope with the trafc environment. Though
not statistically tested, from their verbal comments it seems that
older drivers tended to claim that other road users were responsible for putting them at risk and rarely considered themselves as
those responsible for the hazardous events. This point should be
further addressed in a more detailed study. Furthermore, drivers in
general, should be reminded of the importance of communicating
their intentions (e.g., by signaling) in order to improve older drivers
sense of safety on the road. Road designers should also develop
means to facilitate the conspicuity of environmental elements by,
for example, enlarging font size in important trafc signs or using
special markings where applicable.
Acknowledgments
This research was supported in part by the Ran Naor Foundation and by the Paul Ivanier Center for Robotics and Production
Management at Ben-Gurion University. We would like to thank the
anonymous reviewers for their supportive comments.
References
Armsby, P., Boyle, A.J., Wright, C.C., 1989. Methods for assessing drivers perception
of specic hazards on the road. Accident Analysis and Prevention 21, 4560.
Benda, H.V., Hoyos, C.G., 1983. Estimating hazards in trafc situations. Accident
Analysis and Prevention 15, 19.
Bolstad, A.C., Hess, M.T., 2000. Situation awareness and aging. Situation Awareness
Analysis and Measurement, 277302.
Borowsky, A., Oron-Gilad, T., Parmet, Y., 2009. Age and skill differences in classifying
hazardous trafc scenes. Transportation Research Part F 12, 277287.
Brown, I.D., Groeger, J.A., 1988. Risk perception and decision taking during the transition between novice and experienced driver status. Ergonomics 31, 587597.
Chapman, P.R., Underwood, G., 1998. Visual search of driving situations: danger and
experience. Perception 27, 951964.
Crundall, D., Underwood, G., Chapman, P., 1999. Driving experience and the functional eld of view. Perception 28, 10751087.
Endsley, M.R., 1995. Toward a theory of situation awareness in dynamic systems.
Human Factors 37, 3264.
Finn, P., Bragg, B.W.E., 1986. Perception of the risk of an accident by young and older
drivers. Accident Analysis and Prevention 18, 289298.
Fisher, D.L., Pollatsek, A.P., Pradhan, A., 2006. Can novice drivers be trained to scan
for information that will reduce their likelihood of a crash? Injury Prevention
12, i25i29.
Gitelman, D.R., 2002. ILAB: a program for postexperimental eye movement analysis.
Behavior Research Methods, Instruments and Computers 34 (4), 605612.
Hakamies-Blomqvist, L., Wiklund, M., Henriksson, P., 2005. Predicting older drivers
accident involvement Smeeds law revisited. Accident Analysis and Prevention
37, 675680.
Haworth, N., Symmons, M., Kowadlo, N., 2001. Hazard Perception by Inexperienced
Motorcyclists, Report No. 179. Monash University Accident Research Centre.
Holm, S., 1979. A simple sequentially rejective multiple test procedure. Scandinavian
Journal of Statistics 6, 6570.

A. Borowsky et al. / Accident Analysis and Prevention 42 (2010) 12401249

Horswill, M.S., McKenna, F.P., 2004. A cognitive approach to situation awareness:
theory and application. In: Banbury, S., Tremblay, S. (Eds.), Aldershot. Ashgate
Publishing, pp. 155175.
Logan, G.D., 1985. Skill and Automaticity: Relations, Implications, and Future Directions. Canadian Journal of Psychology 39 (2), 367386.
Matthews, M.L., Moran, A.R., 1986. Age differences in male drivers perception of
accident risk: the role of perceived driving ability. Accident Analysis and Prevention 18, 299313.
Maycock, J., Lockwood, C.R., Lester, J.F., 1991. The Accident Liability of Car Drivers
(No. 315). Transport and Road Research Laboratory, Crowthorne.
McKenna, F.P., Crick, J.L., 1991. Hazard Perception in Drivers: A Methodology for
Testing and Training. Final Report. Behavioural Studies Unit, Transport and Road
Research Laboratory, Crowthorne, UK.
Mills, K.L., Hall, R.D., McDonald, M., and Rolls, G.W.P., 1998. The effects of hazard perception training on the development of novice drivers skills. Report
to Department Environment, Transport and Regions, http://www.roads.detr.
gov.uk/roadsafety/hazard.
Naatanen, R., Summala, H., 1976. Road-user Behavior and Trafc Accidents. NorthHolland, Amsterdam.

1249

Oxley, J., Fildes, B., Corben, B., Langford, J., 2006. Intersection design for older drivers.
Transportation Research, Part F 9, 335346.
Peltz, D.C., Krupat, E., 1974. Caution prole and driving record of undergraduate
males. Accident Analysis and Prevention 6, 4558.
Pollatsek, A., Narayanaan, V., Pradhan, A., Fisher, D.L., 2006. Using eye movements
to evaluate a PC-based risk awareness and perception training program on a
driving simulator. Human Factors 48, 447464.
Sagberg, F., Bjrnskau, T., 2006. Hazard perception and driving experience among
novice drivers. Accident Analysis and Prevention 3, 407414.
Schacter, D.L., 1996. Searching for Memory: The Brain, the Mind, and the Past. Basicbooks, New York.
Underwood, G., Phelps, N., Wright, C., Van Loon, E., Galpin, A., 2005. Eye xations
scanpaths of younger and older drivers in a hazard perception task. Ophthalmic
Physiological Optics 25, 346356.
Wallis, T.S.A., Horswill, M.S., 2007. Using fuzzy signal detection theory to
determine why experienced and trained drivers respond faster than
novices in a hazard perception test. Accident Analysis and Prevention 39,
11771185.