You are on page 1of 8

Safety Science 56 (2013) 36–43

Contents lists available at SciVerse ScienceDirect

Safety Science
journal homepage: www.elsevier.com/locate/ssci

Missing links? The effects of distraction on driver situation awareness


Kristie L. Young ⇑, Paul M. Salmon 1, Miranda Cornelissen 2
Monash University Accident Research Centre, Monash Injury Research Institute, Building 70, Monash University, Clayton, Victoria 3800, Australia

a r t i c l e i n f o a b s t r a c t

Article history: Achieving situation awareness is critical to safe and efficient driving. One factor that can limit drivers’
Available online 5 December 2012 situation awareness and thereby affect driving performance is distraction. This study aimed to investi-
gate, under real-world driving conditions, whether and how visual–manual distraction disrupts drivers’
Keywords: situation awareness. Twenty-three participants drove an instrumented vehicle around an urban test
Driver distraction route while distracted (i.e., performing a visual detection task) and while not distracted. A network anal-
Situation awareness ysis procedure, based on verbal protocols provided by drivers en-route, was used to describe and analyse
On-road study
participants’ situation awareness. The findings indicate that engaging in a visually distracting task did not
alter the structure of drivers’ SA, but did change the content of their awareness. When distracted, drivers
tended to place less emphasis on visual scanning element in favour of focussing on elements related to
vehicle control tasks. These results have implications for the design and regulation of technologies being
introduced into vehicles.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction understanding is required of how distraction affects aspects of hu-


man cognition and how this translates into impaired performance.
Over the past decade, an intense research effort has been de- One aspect of cognition that is likely to be affected by distraction is
voted to the study of driver distraction. This has largely been dri- drivers’ situation awareness (SA) (Salmon et al., 2011). The present
ven by the increasing popularity of in-vehicle devices and study aimed to investigate, under real-world driving conditions,
subsequent concerns that drivers’ use of these systems may com- whether and how visual-manual distraction disrupts driver SA.
pete for limited cognitive resources, resulting in impaired driving
performance and safety (Regan et al., 2009; Young and Lenné, 1.1. The role of situation awareness in driving
2010). A large body of research has documented the myriad ways
in which distraction by in-vehicle devices impact driving perfor- Not surprisingly given its origins, the study of SA has tradition-
mance, including reduced longitudinal (Rakauskas et al., 2004; ally been carried out within the aviation domain (Stanton et al.,
Strayer and Drews, 2004) and lateral control (Engstrom et al., 2001). Notwithstanding this, SA has been identified as a critical
2005; Reed and Green, 1999) and impaired response times to traf- component of human performance in any complex, dynamic envi-
fic signals and roadway hazards (Lee et al., 2001; Liang and Lee, ronment, including driving (Endsley, 1995b). Driver SA is defined
2010). Moreover, research indicates that secondary task engage- as activated knowledge, regarding road user tasks, at a specific
ment, including interacting with in-vehicle devices, is a contribut- point in time (Salmon et al., 2011). This knowledge is derived from
ing factor in up to 23% of crashes and near-crashes (Klauer et al., schema driven sampling and perception of the driving environ-
2006). ment, which in turn directs driver behaviour, the outputs of which
Much of the driver distraction research to date has focussed on confirm or modify drivers’ schema and so on. To safely navigate
the effects of distraction on driving performance measures. Few through a continuously and often rapidly changing roadway, driv-
studies have examined the cognitive and perceptual roots of dis- ers must perceive and attend to relevant information (e.g., vehicle
traction; that is, the mechanisms by which it leads to the perfor- locations and changing headway distances) and use this to antici-
mance impairments observed (Young and Salmon, 2012). In pate and react to changes and events in the environment to avoid
order to develop effective distraction countermeasures, a greater conflicts with objects and other road users (Gugerty, 2011).
Achieving SA involves a range of cognitive processes including
perception and pattern recognition (Kass et al., 1991), attention
⇑ Corresponding author. Tel.: +61 3 9905 1258; fax: +61 3 9905 4363.
and comprehension (Kass et al., 2007; Wickens and Hollands,
E-mail addresses: kristie.young@monash.edu (K.L. Young), paul.salmon@
monash.edu (P.M. Salmon), miranda.conrelissen@monash.edu (M. Cornelissen).
2000), and decision-making (Endsley, 1995b; Ma and Kaber,
1
Tel.: + 61 3 9905 1907. 2005). Evidence suggests that engagement in distracting activities
2
Tel.: + 61 3 9905 4669. that compete for these same cognitive resources can lead to a

0925-7535/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ssci.2012.11.004
K.L. Young et al. / Safety Science 56 (2013) 36–43 37

breakdown in drivers’ SA and, ultimately, impaired performance the freeze probe technique is that it only captures data on drivers’
(Kass et al., 2007). A driver engaged in a visual-manual phone dial- knowledge of SA elements: it does not measure the integration of
ling task, for example, may sample the visual environment less of- these elements or relationships made between them. The implica-
ten, leading to information on a rapid change in headway not being tions of these limitations are, first, that further research on the
detected in time to avoid a rear-end collision. Understanding how relationship between distraction and SA is needed, and, second,
distraction can lead to failures in driver SA may therefore play a that new methodological approaches are required in order to fulfil
crucial role in explaining the link between distraction and the this research need.
decrements in driving performance observed in previous studies. A recently proposed account of SA within sociotechnical sys-
tems views SA as comprising information elements, or concepts,
1.2. Previous research on distraction and situation awareness and relationships between them (Salmon et al., 2009). From this
viewpoint, the relationship between concepts is arguably more
While a number of studies infer that the driving impairments important than the concepts themselves. For example, in the driv-
observed when distracted may be due to a loss of SA, only a hand- ing domain there is evidence that in experts, SA may be more than
ful of studies have directly examined the affect of distraction on merely the sum of its informational parts, and also that experts can
driver SA (Kass et al., 2007; Ma and Kaber, 2005; Rogers et al., achieve appropriate levels of SA without accessing all relevant
2011). Ma and Kaber (2005), for example, investigated the effects information in the environment (e.g., Salmon et al., in press; Walk-
of Adaptive Cruise Control (ACC) and mobile phone use on SA er et al., 2009). When using a SAGAT style approach, it could con-
whilst performing a vehicle following task in a driving simulator. ceivably be that drivers who perceive many elements from the
Drivers’ SA was assessed using the Situation Awareness Global driving environment are characterised as having high levels of SA
Assessment Technique (SAGAT; Endsley, 1995a), whereby drivers regardless of whether or not they integrate the elements and make
were asked a series of SA queries at various ‘frozen’ (simulation connections between them. On the other hand, drivers who per-
stopped and all screens and displays blanked) points in the drive ceive fewer elements but integrate them better could be character-
(e.g., recall car locations and colours). Ma and Kaber (2005) found ised as having poor levels of SA. In the context of distraction, a
that while driver SA improved in the ACC condition, SA was de- driver who compensates for being distracted by using a parsimoni-
graded when drivers conversed on the mobile phone in terms of re- ous model of the situation (i.e., fewer, but well-integrated informa-
duced comprehension of the driving environment and degraded tion elements) will be characterised as performing worse than
projection of future driving environment states. It was believed drivers who take on board many elements but do not integrate
that the mobile phone conversations competed for limited mental them well. Thus the method only provides a partial picture of dri-
resources, leading drivers to pay less attention to the roadway. ver SA and may not accurately reflect breakdowns in SA due to dis-
They concluded, on the other hand, that ACC may allow drivers traction, particularly if the failures relate to drivers integration of
to achieve higher levels of SA by relieving them of vehicle monitor- perceived information into a situational model. Network-based
ing and control tasks. Kass et al. (2007) also investigated the effect analysis of driver SA, drawn from verbal protocols, where SA is rep-
of mobile phone use on novice and experienced driver SA and per- resented as information ‘elements’ and the relationships between
formance. Using a SAGAT measure, they found that overall, novice them, can overcome these limitations by allowing both the infor-
drivers had poorer SA and made more driving infractions (e.g. col- mation elements and their relationships to be examined (Walker
lisions, speeding violations, missed stop signs, lane keeping errors) et al., 2011). In recent times using driver verbal protocols along
than expert drivers; however, they also found that all drivers were with network analysis has become popular as an approach to
not able to maintain the same level of SA as other drivers not en- investigate driver SA on road in the real world (Walker et al.,
gaged in a phone conversation. Other studies exploring the effects 2011, 2008).
of mobile phones and navigation systems on driver SA have also In summary, our understanding of driver SA and its links with
observed similar reductions in driver SA when distracted (e.g., distraction is incomplete. Further, there has been a lack of on-road
Gugerty et al., 2003; Rogers et al., 2011). studies investigating the relationship between SA and distraction.
Previous studies examining driver SA have a range of limita- The present study aimed to investigate, via on-road study, the ef-
tions that constrain our understanding of the concept and its links fects of distraction on driver SA. Driver SA was modelled through
with driver distraction. The first is that most of the previous stud- propositional networks, which were constructed based on a con-
ies conducted on SA and distraction are simulator-based and thus tent analysis of verbal protocols provided by participants while
suffer the limitation that drivers’ allocation of attention between negotiating a pre-defined urban test route. Participants drove the
the driving and distracter task may not be the same as it would test route twice, once while distracted by the Visual Detection Task
be on real roads where the consequences of poor driving perfor- (VDT; Engström and Mårdh, 2007) and once while not distracted.
mance are real. Differences in attention allocation strategies may The study was expected to provide a more comprehensive picture
have particularly large implications for the study of driver SA than currently exists of the impact of distraction on driver SA by
and distraction, because both concepts are so closely linked to examining not only the content (e.g., information elements) of
attention. The second limitation is that most studies have at- the SA networks, but also their structure (e.g., the relationships be-
tempted to assess SA against a normative ideal, allocating an SA tween elements). Based on existing literature from the distraction
score to participants based on their responses to queries regarding and SA domains (e.g., Ma and Kaber, 2005; Walker et al., 2011) and
their knowledge of the situation when distracted and not dis- the conceptualisation of SA as concepts and the relationships be-
tracted. Although useful in that a degradation in SA can be identi- tween them (Salmon et al., 2009), it was predicted that being en-
fied, changes in the actual content of SA in terms of knowledge and gaged in a distracting task would lead to degraded driver SA.
information from the driving environment cannot be examined. Based on the use of a network analysis measure to describe and as-
Moreover, in real world scenarios it is not possible to determine, sess SA (see Section 2.4), it was hypothesised that this degradation
beforehand, what SA should comprise. The third key limitation is in SA would manifest itself in the form of fewer links or relation-
that most studies have used a SAGAT freeze probe style method ships being made between the information elements being sam-
for measuring driver SA. This method has various limitations, pled from the environment. Previous network analysis-based SA
including that it only captures driver SA at specified points in time assessments, for example, suggest that networks with lower con-
and can only be applied in simulated environments since it re- nectivity between information elements are indicative of poorer
quires the driving scenario to be frozen. Another concern with SA (e.g., Salmon et al., 2009; Walker et al., 2011) It was also pre-
38 K.L. Young et al. / Safety Science 56 (2013) 36–43

dicted that the type of elements sampled from the environment The VDT was piloted prior to testing to ensure that it provided a
would differ when drivers were distracted compared to when sufficient level of cognitive demand, as determine through objec-
undistracted. tive measures of driving performance and subjective ratings. When
distracted, drivers’ mean (Walker et al., 2011) and standard devia-
tion of speed (distracted = 13.1 km/h, undistracted = 12.6 km/h)
2. Method
were significantly higher than when not distracted (t(19) = 2.39,
p = .028 and t(19) = 2.33, p = .031, respectively) and drivers made
2.1. Participants
a greater number of lane excursions (42 distracted and 34 undis-
tracted). Using the NASA-RTLX, drivers rated their overall workload
Twenty-three drivers (10 males, 13 females) aged 19–51 years
as being significantly higher during the distraction drive (M = 8.9,
(mean = 28.9, SD = 8.6) participated in the study. Seventeen partic-
SD = 2.3) than during the undistracted drive (M = 6.9, SD = 2.5),
ipants held a valid Full driver’s license while the remaining six held
t(21) = 3.25, p = .004.
a valid Probationary (P2) license. Participants had held their driv-
The set-up for the VDT followed the general specifications de-
ers licence for an average of 10.1 years (SD = 8.9) and drove an
scribed in Engström and Mårdh (2007). A single red LED light
average of 11.9 h (SD = 15.6) per week. Participants were recruited
was positioned so that the stimulus was presented in the driver’s
through the weekly on-line Monash University newsletter and
central field of view via reflection in the windshield (Fig. 1). The
were compensated for their time and travel expenses. The study
intensity of the LED was two candelas. Each visual stimulus was
was approved by the Monash University Human Research Ethics
presented for a maximum of two seconds and turned off as soon
Committee.
as the response button was pressed. Stimuli were presented with
a random temporal variation of 3–5 s. Participants responded to
2.2. Materials each stimulus by pressing a button attached to their left index fin-
ger. The button provided mechanical haptic feedback and an audi-
2.2.1. On-road test vehicle tory ‘click’ to indicate that the response had been registered. The
The On-Road Test Vehicle (ORTeV) is an instrumented vehicle presentation of the visual stimuli was controlled by an HP laptop
equipped to collect three main types of data: vehicle-related, eye which also recorded mean response time to the stimuli and hit
tracking data and roadway scene data. Vehicle data are acquired rate.
from the vehicle network and include: vehicle speed, GPS location,
accelerator and brake position, steering wheel angle, lane tracking
and headway logging, primary controls (windscreen wipers, indi- 2.2.3. Driver verbal protocols
cators), and secondary controls (GPS navigation, entertainment Verbal Protocol Analysis (VPA) was used to elicit data regarding
system). Driver eye movements can also be tracked and overlaid the cognitive processes undertaken by drivers while en route and
on a driver’s-eye camera view using the FaceLab and Scene Camera formed the primary measure of drivers’ SA. VPA is commonly used
eye tracking systems. Data on the roadway scene is derived from to investigate the cognitive processes associated with complex task
seven unobtrusive cameras which record forward and peripheral performance and has been used previously as the input to SA mod-
views spanning 90° each respectively as well as the driver, the elling procedures in various domains, including road transport
vehicle cockpit and the rear of the vehicle. In the current study, (Walker et al., 2007) and under high workload conditions including
all three forms of data were collected while drivers drove the OR- nuclear power plant malfunctions (e.g., Kirwan et al., 1995) and a
TeV twice around a pre-defined on-road route. For the purpose of complex process control task (Sanderson et al., 1989). In the pres-
this paper, data from the instrumented vehicle were used to ensure ent study, participants provided verbal protocols continuously as
that the distracter task had an observable effect on driving perfor- they drove the instrumented vehicle around the test route. Partic-
mance and, thus, was imposed a sufficient level of demand. Data ipants received extensive training on the VPA task, under both sta-
from the FaceLab system are not reported here. tic and driving conditions, and were directed to merely verbalise
they were doing and seeing, but not explain or rationalise their
behaviour, so as to minimise the effect of the verbal protocols on
2.2.2. Visual detection task driving performance. Given that drivers were not exposed to a ‘si-
The Visual Detection Task (VDT) was used as the distracter task. lent’ condition where they provided no verbal reports, it is not pos-
This task was selected as it is a low demand visual detection task sible to examine directly any effect that the VPA task may have had
that requires a manual rather than a verbal response from drivers,
which drivers were unable to provide given that they were provid-
ing concurrent verbal protocols (see Section 2.2.3). The VDT was
used as the distracter task in the current study. The VDT was devel-
oped as part of the SafeTE project conducted by Volvo Technology
and Swedish National Transport Research Institute (VTI) (Engström
and Mårdh, 2007). The VDT is an extension of the Peripheral Detec-
tion task (PDT; van Winsum et al., 1999) where the visual stimulus
is presented at increased intensity and in the driver’s central field
of view in order to provide a measure of pure cognitive demand
that is not affected by large visual eccentricity and, hence, periph-
eral interference. The VDT was originally developed as a non-intru-
sive measure of cognitive load rather than a distracter task,
however, its utility to serve as a low demand visual-manual dis-
tracter task for use in on-road studies was examined as part of this
proof-of-concept study. Research by Broström et al. (2009) has
suggested that the VDT may indeed affect driver performance,
finding that the VDT had a significant effect on secondary task
completion time and subjective measures of mental workload
and stress. Fig. 1. Approximate location of the VDT stimuli.
K.L. Young et al. / Safety Science 56 (2013) 36–43 39

on driving performance without confounding results with the pres- is ‘red’. This process produced two networks for each participant,
ence or absence of the distracter task. ‘Think-aloud’ concurrent one representing their SA during the distracted drive and the other
verbalisation tasks such as VPA, where participants merely provide their SA during the undistracted drive. The Agna™ network analy-
verbal expression of their thoughts and actions without describing sis software was then used to interrogate their content and struc-
or explaining their behaviour, has been shown in a 94 study meta- ture using various network metrics.
analysis to have no significant effects on objective task perfor-
mance (Fox et al., 2011). Hughes and Cole (1986) also found that 3. Results
the process of providing concurrent verbal reports did not signifi-
cantly affect driver attentional demands under both real road 3.1. Verbal protocols: semantic extraction
and video-based experimental contexts. If participants fell silent
for more than 5 s, they were given a prompt to continue with their The word count of the participants’ verbal transcripts serves as
verbalisations. The verbal protocols were recorded using a digital an estimate of the amount of semantic content that drivers were
Dictaphone and transcribed verbatim post-trial. able to extract throughout the drive when distracted and not dis-
tracted. Word counts are currently one of the best estimates of
2.2.4. Driving route the amount of semantic content available and has been used pre-
The driving route comprised an 8.7 km urban route around the viously (Walker et al., 2011) as an estimate of the amount of
suburbs surrounding the Monash University Clayton Campus. The semantic content extracted by drivers. (Walker et al., 2011). The
test route comprised arterial (80, 70 and 60 km/h) and residential mean word counts across participants for the distracted and undis-
(50 km/h) roads and contained a total of 11 intersections with a tracted drives are presented in Table 1. The words counts were not
combination of fully, partially and un-signalised traffic controls. weighted for time taken to complete the test route given that the
The route took approximately 15–20 min to complete. To control time taken was almost identical across the drive conditions.
for traffic conditions, all drives were completed on weekdays be- The total word count was 1888 words or 8.3% lower when driv-
tween the hours of 9.30 am and 2.30 pm. These times had been as- ers were distracted compared to when not distracted. This suggests
sessed by the authors prior to the study to ensure that participants that drivers were able to extract less semantic content from the
would experience similar traffic conditions. Direction instructions same (or very similar) situations when they were distracted. The
were provided to participants by one of two in-vehicle observers. analysis now proceeds with an assessment of how this reduced
semantic content affected the content and structure of driver SA.
2.3. Procedure

3.2. Network analysis


A demographic (age, gender, license type, driving history) ques-
tionnaire was completed by participants prior to the study. After a
The two SA networks for each participant (distracted and undis-
VPA training session, participants were seated in the ORTeV and
tracted) were analysed quantitatively and qualitatively. The quan-
the data collection systems were initiated and FaceLab was cali-
titative analysis involved using the density, diameter and
brated. Participants then completed a static baseline run of the
sociometric status metrics to analyse the structure and content of
VDT while seated in the stationary vehicle, followed by a 20 min
each participants’ SA network under distracted and undistracted
practice drive around the Clayton campus to familiarise them-
conditions. T-tests were used to measure differences in these met-
selves with the ORTeV, the VPA method and the VDT. At the end
rics across the distraction conditions. The qualitative analysis in-
of the practice route, participants were informed that the test
volved identifying the concepts that were common across
had begun and that data collection had commenced. The partici-
networks and those that were unique to each distraction condition,
pants then completed the driving route twice, once while perform-
and also the key concepts underpinning SA in each condition.
ing the VDT (distracted) and once while not performing a
secondary task (undistracted). The order in which the distracted
and undistracted drives were completed was counterbalanced 3.2.1. Network structure: density and diameter
across participants using a standard Latin square design. Partici- Density is a measure of the overall interconnectivity of the net-
pants provided verbal protocols during each test drive. Two in- work in terms of the number of links between information ele-
vehicle observers were present in the vehicle to record the driving ments. The formula for calculating network density is:
errors made during each test drive. The findings from these error- 2L
based observations are reported separately in Young et al. (in Network density ¼
gðg  1Þ
press).
where L is the number of links in the network and g is the number of
2.4. Network creation and analysis concepts or information elements. Network density values range
from 0 (no concepts are connected) to 1 (every concept is connected
Network analysis was used to describe and analyse participants’ to every other concept). Thus, higher values indicate a richer set of
SA. The verbal protocols provided by participants were first tran- links between concepts and enhanced SA. The mean density values
scribed verbatim and then a content analysis of each transcript for participants’ networks across the distracted and undistracted
was performed using the text analysis software Leximancer™. Lex- drives are presented in the left column of Table 2. The mean density
imancer™ automates the creation of networks by extracting or interconnectedness of the networks was almost identical across
themes, concepts and their associated links from the verbal tran-
scripts in a five step process: conversion of raw text data, concept
Table 1
identification, thesaurus learning, concept location, and mapping
Mean and total word counts for drivers during the distracted and undistracted drives.
(i.e. visual representation of network). Leximancer has been used
successfully in previous studies to create networks representing Condition Mean (SD) Total Mean (SD) time to
count complete route
the SA held by different road users during on-road studies (Salmon
(mm:ss)
et al., in press; Walker et al., 2011). The networks produced com-
Distracted 946.96 (432.85) 21780 15:01 (1:53)
prise concepts, or ‘information elements’, and the links between
Not distracted 1029.04 (512.56) 23668 14:59 (1:22)
them. For example, ‘vehicle’ has ‘slowed’ its ‘speed’; ‘traffic light’
40 K.L. Young et al. / Safety Science 56 (2013) 36–43

Table 2 however, between two thirds and three quarters of the concepts
Mean (SD) network density and diameter across the distracted and undistracted contained in the master networks were shared across the dis-
drives.
tracted and undistracted drives. On the other hand, around one
Condition Density Diameter third of the concepts were unique to each particular distracted
Distracted .603 (.135) 2.30 (0.47) condition, suggesting that the content of driver SA did differ when
Not distracted .601 (.145) 2.39 (0.49) drivers were distracted.
p .915 .426 To further explore the content of drivers’ SA the shared and un-
ique concepts were compared and contrasted in greater detail
across the distracted and undistracted conditions (see Table 4).
the distracted and undistracted drives, which is reflected in the The concepts that were shared across the distracted and undis-
highly non-significant p value from the t-test. This suggests that tracted conditions included those related to checking or scanning
the level of connectedness within drivers’ SA networks did not dif- for other road users (e.g., checking, looking, watching, cars, traffic,
fer across the distracted and non-distracted drives. truck, pedestrian) both behind (e.g., behind and mirrors) and in
Diameter is another network-wide metric that is used to ana- front of the vehicle (e.g., ahead, coming and front). The shared con-
lyse the connections and path lengths between concepts within a cepts also included those relating to vehicle control tasks (e.g.,
network. It is given by the formula: speed, lane, indicating, changing (lanes), slowing) and anticipating
upcoming events or the behaviour of other road users (e.g., think-
Diameter ¼ max dðg i ; g j Þ
uy ing, wondering, waiting).
As seen in Table 4, the concepts unique to each condition varied
where d(gi,gj) is the largest number of concepts that must be across drivers, with many being unique to individual drivers. How-
crossed to travel from one concept to another. With regard to SA, ever, many of the individual concepts can be grouped into common
lower diameter scores are indicative of more enhanced SA because themes, such as the ‘van’, ‘ute’, ‘truck’ and ‘vehicle’ concepts in the
the routes through the network are shorter and more direct. Greater distracted condition all relating to surrounding vehicles, and
diameter values are indicative of lower SA because there are more ‘wide’, ‘narrow’, ‘corner’ and ‘merge’ concepts in the undistracted
concepts per pathway through the network (Walker et al., 2011). condition relating to roadway conditions. This variation in con-
The mean diameter values for the participants’ networks overall cepts therefore likely reflects differences in individual’s speech
and at the different roadway sections are presented in the middle and in the road and traffic conditions experienced. The concepts
column of Table 2. The diameter scores are quite low, suggesting that are unique to each distraction condition indicate some differ-
that the concepts contained in the SA networks were well inte- ences in the information elements focussed on during the drives.
grated, and did not differ significantly across the distracted and When not distracted, drivers’ unique concepts included those re-
undistracted drives. lated to the road environment and how these elements guided or
In summary, when drivers were distracted, the structure of affected their behaviour (e.g., shops, busier, wide, strip, narrow,
their SA networks was similar to that when they were not dis- cameras, and roadside). These types of roadway elements were
tracted. Both the density and diameter metrics suggest that the not evident in the concepts unique to the distracted drive. Rather,
participants’ networks contain a rich set of concepts with well inte- when distracted, the unique concepts appeared focussed on the ac-
grated links between them. However, it is possible that being dis- tions the driver was undertaking (e.g., making, cause, passing, putt-
tracted may change the content rather than the structure of ing, and driving). These findings suggest that when distracted,
drivers’ SA networks. The analysis of network content is the focus drivers’ focus shifted from external environmental elements to
of the next section. their own behaviour and the actions they needed to perform to
control the vehicle and negotiate the route. It is also interesting
3.2.2. Network content: shared, unique and key concepts to note that the concepts unique to distracted drivers included
The analysis of network content involved both qualitative and those related to checking or the areas to the side of the vehicle
quantitative interrogation of the networks. For the qualitative (e.g., besides, sides), which may be indicative of distracted drivers
assessment, two ‘master’ networks were created, one for distracted narrowing their scan area to those areas immediately surrounding
and one for the undistracted drives, using all of the elements con- the vehicle. Alternatively, the presence of other unique concepts
tained in every participants’ networks. Shared concepts (i.e. that such as ‘space’ and ‘distance’ may also indicate that drivers become
were present in both the distracted and undistracted master net- more attentive of the safety margins surrounding the vehicle when
works) and unique concepts (i.e. present only in one of the master distracted.
networks) were identified. The sociometric status metric for shared A further method for examining how the content of driver SA
concepts was then used to identify the relative importance of the may differ when distracted is to compare the sociometric status
shared concepts to drivers’ SA when distracted and not distracted of each shared concept across the distracted and undistracted con-
as well as the key information elements underpinning SA. ditions. Sociometric status provides a measure of how busy or
Table 3 shows the total number of concepts within each master important a concept is relative to the total number of concepts
network along with the number and percentage of unique and within the network and, there, indicates which concepts have more
shared concepts within each distraction condition network. The prominence in driver SA under each condition. It is calculated
undistracted condition network had a greater number of concepts using the formula:
than the distracted network and the undistracted condition also
1 X
g
contained a slightly higher proportion of unique concepts. Overall,
Sociometric status ¼ ðX ji þ X ij Þ
g  1 j1
Table 3
Number of shared and unique concepts across the distracted and undistracted drives. where g is the total number of nodes in the network, i and j are indi-
vidual nodes and are the edge values from node i to node j. Concepts
Condition Total no. of Mean (range) no. of Shared Unique
with high sociometric values are highly connected to other concepts
concepts concepts per driver
in the network, while concepts with low values often reside in the
Distracted 110 22.7 (13–27) 78 (70.9%) 32 (29.1%)
periphery of the network and are not well connected with other
Not distracted 118 21.7 (16–27) 78 (66.1%) 40 (33.9%)
concepts. The mean sociometric status of the shared concepts is
K.L. Young et al. / Safety Science 56 (2013) 36–43 41

Table 4
Shared and unique concepts across the distracted and undistracted drives.

Shared across distracted Unique to distracted Unique to


and undistracted undistracted
Checking Time Indicator (6) Aware (3)
Car Straight Driving (3) Nice (3)
Cars Looking Passing (2) Busy (2)
Driving Watching White (2) Car’s (2)
Behind Guy Beside (1) Scan (1)
Intersection Moment Access (1) Limits (1)
Speed Crossing Service (1) Doing (1)
Slow Hand Making (1) Guys (1)
Lane Past Cause (1) Shops (1)
Indicating Arrow Sun’s (1) Busier (1)
Light Parked Putting (1) Girl (1)
Road Change Van (1) Middle (1)
Thinking Pedestrian Stage (1) Hour (1)
Mirror Slowing Ute (1) Lots (1)
Traffic Man Stay (1) Wide (1)
Green Stopping Oncoming (1) Strip (1)
Changing Stopped Sides (1) Wrong (1)
km Wondering Train (1) Narrow (1)
Sure Bus Round (1) Forward (1)
Drive Worries Truck’s (1) Halt (1)
Front Lanes Somebody’s (1) Possibly (1)
Safe Cyclist Somebody (1) Hoon (1)
Lights Limit Directly (1) Cautious (1)
Coming Street Shelter (1) Relaxed (1)
Ahead Awareness Top (1) Stick (1)
Eye Look Observe (1) Head (1)
Turning Quick Vehicle (1) Braking (1)
Clear Approaching Anybody (1) Cameras (1)
Waiting Person Fast (1) Clayton (1)
Truck Observing Space (1) Eyes (1)
Red Ready Annoying (1) Slowly (1)
Direction Signal Distance (1) Corner (1)
Moving Nobody Blind (1)
Pulling Quiet Electronic (1)
Indicate Walking Confusing (1)
Someone Pulled Wipers (1)
People Pull Yellow (1)
Turned Wait Roadside (1)
Sign Watch Area (1)
Merge (1)
Non-relevant concepts: keeping, trying, Non-relevant Non-relevant
generally, hopefully, having, seems, concepts: one’s, concepts: old,
slightly, probably, giving used, feel, anyway, happening, case
ease

Numbers in parentheses indicate number of participants who had concept in SA network.

presented in Table 5 for each distraction condition. Only those when drivers were distracted. Given that the network sizes are
shared concepts that featured in five or more participants’ net- quite small (maximum of 27 concepts) the differences observed
works, and thus were those most important for driver SA, were in- in the sociometric status values show a meaningful difference in
cluded in the sociometric analysis to ensure their prominence and the importance of the described concepts to driver SA across the
to ensure that the results extracted were meaningful, understand- distracted and undistracted conditions. These findings suggest that
able and were most relevant to the drivers’ SA. This process yielded when distracted, drivers ‘shed’ visual scanning tasks and instead
24 shared concepts that were examined. The concepts excluded focus their awareness more on elements or tasks related to main-
from the analysis were mentioned by fewer participants and thus taining control of the vehicle.
it was deemed that they were less important for SA. The impact
of excluding them from this analysis is therefore not expected to
influence the conclusions made. 4. Discussion
A comparison of the mean sociometric status across the dis-
tracted and undistracted conditions revealed that, when not dis- Achieving SA is crucial for safe driving and this concept is likely
tracted, information elements associated with scanning the to play an important role in explaining how engagement in dis-
surrounding environment and road users were more prominent. tracting tasks leads to driving performance decrements observed
This is evidenced by undistracted drivers having higher sociomet- in the literature (Young et al., 2008). This exploratory study aimed
ric values for concepts such as ‘checking’ ‘looking’, ‘cars’, ‘behind’, to investigate the effects of visual-manual distraction on driver SA
‘front’, ‘ahead’, and ‘coming’, all of which relate to scanning the sur- to improve our understanding of how distraction can lead to fail-
rounding roadway and traffic. In contrast, when drivers were dis- ures in SA and, consequently, impaired driving performance.
tracted, concepts related to vehicle control tasks featured more Importantly, both the structure and content of driver SA was exam-
prominently in their SA networks. For example, concepts such as ined in the present study, providing a more comprehensive picture
‘speed’, ‘lane’, ‘light’, and ‘slowing’ had higher sociometric values of the links between SA and distraction.
42 K.L. Young et al. / Safety Science 56 (2013) 36–43

Table 5 does not fully explain the shift towards more vehicle control type
Mean sociometric status of common shared concepts across the distracted and elements when distracted. This shift in SA content from scanning
undistracted drives.
elements to control based elements may indicate that drivers
Concept Distracted Undistracted ‘shed’ what they believe to be non-critical scanning tasks due to
Counta Sociometric status Count Sociometric status the increased load of the distracter task. However, drivers did not
Concepts more prominent for undistracted drivers
shed all scanning tasks, rather their scanning behaviour appeared
Checking 14 2.80 10 3.20 to narrow or become more tightly focussed on those areas imme-
Car 20 3.05 21 3.59 diately surrounding the vehicle; as evidence by concepts such as
Cars 16 3.02 17 3.82 ‘sides’ and ‘beside’, rather than ‘roadside’ and ‘shops’, featuring in
Behind 15 3.00 12 3.67
their unique concepts.
Slow 6 1.79 8 2.07
Road 17 2.11 14 3.33 These findings are congruent with the results of previous re-
Green 17 2.51 18 3.03 search that has found an effect of distraction on drivers’ visual
Sure 8 2.00 7 2.66 scanning behaviour (Greenberg et al., 2003; Harbluk et al., 2007;
Front 11 1.82 12 2.66
Victor et al., 2005). These studies have found that distraction has
Lights 19 2.62 20 3.63
Coming 19 2.35 18 2.69
a strong influence on drivers’ visual scanning patterns, leading to
Ahead 11 3.14 13 4.01 the concentration of gaze to the road centre, reduced scanning of
Turning 18 2.37 19 2.73 peripheral regions and impaired event detection. The suggestion
Moving 5 2.40 5 2.93 from the present study that this gaze concentration effect may
Looking 8 4.51 6 6.21
be due to a reduced focus or emphasis on visual scanning of envi-
Concepts more prominent for distracted drivers ronmental elements, especially those in the periphery, accords
Intersection 7 2.38 7 2.06
well with recent evidence that this phenomenon is caused by gen-
Speed 11 2.80 13 2.22
Lane 17 3.36 20 3.21 eral interference in top-down selective attention rather than a per-
Light 13 2.18 13 1.77 ceptual narrowing (Victor et al., 2009).
Traffic 17 3.16 14 2.97 Taken together, the findings on SA structure and content sug-
Clear 7 1.92 8 1.61 gest that distracted drivers do not take less information from the
Red 11 2.47 10 2.34
Slowing 9 2.33 8 1.71
driving environment or make fewer connections between con-
Look 6 2.02 6 1.44 cepts, but rather that they focus on different aspects of the envi-
a
ronment or driving task. Evidence that distraction changes the
Number of participants who had concept in their SA network.
content of driver SA of course begs the question as to how these
changes manifest in terms of degraded driving performance and
poor safety outcomes and how environmental factors such as road-
Contrary to expectation, the findings indicate that engaging in a way design might moderate this. An examination of the correlation
visual-manual distracter task did not alter the structure of drivers’ between driver SA and driving performance was beyond the scope
SA; that is, drivers were still able to integrate concepts in order to of the current study but this should be the focus of further research
generate awareness whilst distracted. However, as predicted, to extend our knowledge of the nexus between driver distraction
engaging in a distracting task did change the content of their and driver SA and what factors may moderate this relationship.
awareness; that is, the concepts underpinning SA were different Given its exploratory nature, this study contained some limita-
when distracted versus when not. Specifically, with regard to net- tions that are worthy of note. The study utilised a small sample
work structure, the findings suggest that the level of interconnec- size, only explored one type and level of distraction and, impor-
tivity between the information elements extracted from the tantly, the drivers were exposed to only a limited range of driving
driving environment did not differ when drivers were distracted situations and traffic scenarios under light traffic conditions. When
versus not distracted. Given that the same drivers were used across exposed to a greater range of traffic conditions and scenarios, fur-
the distracted and undistracted conditions, it is not completely ther affects of distraction on the structure and content of driver SA
unexpected that SA network structure did not differ appreciably may become apparent. Likewise, the use of different types and lev-
across conditions, because the basic building blocks of SA (i.e., els of distracter tasks may also reveal further differences in the nat-
how information is integrated) are heavily influenced by drivers’ ure of driver SA when distracted. Another issue lies with the
individual cognitive capacity, strategies and schema. Hence, it is measurement of driver SA itself. While many information elements
possible that being distracted may not affect so much how drivers’ that underpin driver SA can be reported by drivers, there may still
achieve SA, but rather what environmental elements they sample be elements that are not amenable to verbal reporting, but never-
and focus onto achieve SA (i.e., SA content). This finding could also theless form an important part of driver SA (Walker et al., 2008). It
be an artefact of the VPA method used to assess SA, which relies is possible that distraction might affect these more implicit aspects
totally on drivers’ verbalisations. Given that people essentially talk of SA and such effects would not have been detected through the
the same way under varying conditions, this may have led to the verbal protocol method used here. Future research on distraction
high similarities in the structure of the SA networks across the dis- and SA should seek to utilise a range of SA measures that can assess
traction conditions. the implicit as well as explicit aspects of driver SA.
Analysis of drivers’ network content revealed that drivers’ SA
content did indeed differ when they were distracted, with around
one third of the information elements extracted being unique to 5. Conclusion
each drive condition. When not distracted, drivers’ networks con-
tained more information elements associated with scanning the Drivers will continue to be exposed to an array of complex, mul-
surrounding environment and road users. In contrast, when they ti-functional devices in the vehicle that will compete for their lim-
were distracted, drivers tended to reduce or place less emphasis ited cognitive resources. Continued research efforts are therefore
on visual scanning in favour of focussing on elements related to required to establish how drivers’ ability to achieve and maintain
specific vehicle control tasks. This might be a direct effect of the SA is affected by the influx of driver support and information tech-
VDT, which may have led drivers to maintain their gaze toward nologies into the vehicle. Such knowledge is essential for informing
the road centre in order to detect the visual targets; however, this the safe design and regulation of current and future technologies,
K.L. Young et al. / Safety Science 56 (2013) 36–43 43

as well as designing training programs aimed at enhancing driver Ma, R., Kaber, D.B., 2005. Situation awareness and workload in driving while using
adaptive cruise control and a cell phone. International Journal of Industrial
SA generally and under conditions of distraction or increased load.
Ergonomics 35, 939–953.
Rakauskas, M.E., Gugerty, L.J., Ward, N.J., 2004. Effects of naturalistic cell phone
Acknowledgements conversations on driving performance. Journal of Safety Research 35 (4), 453–
464.
Reed, M.P., Green, P.A., 1999. Comparison of driving performance on-road and in a
This study was funded by the Monash University Researcher low-cost simulator using a concurrent telephone dialing task. Ergonomics 42
Accelerator Program. Dr. Salmon’s contribution to this article was (8), 1015–1037.
part funded by his Australian National Health and Medical Re- Regan, M.A., Lee, J.D., Young, K.L., 2009. Driver Distraction: Theory, Effects and
Mitigation. CRC Press, Boca Raton, Florida.
search Council Post Doctoral Fellowship. We thank Johan Engström Rogers, M., Zhang, Y., Kaber, D., Liang, Y., Gangakhedkar, S., 2011. The effects of
for his advice on the Visual Detection Task and Nebojsa Tomasevic visual and cognitive distraction on driver situation awareness. Paper presented
and Ashley Verdoorn for their assistance with the ORTeV. at the HCI International 2011, Orlando, FL.
Salmon, P.M., Stanton, N.A., Walker, G.H., Jenkins, D.P., 2009. Distributed Situation
Awareness: Advances in Theory, Measurement and Application to Teamwork.
References Ashgate, Aldershot, UK.
Salmon, P.M., Stanton, N.A., Young, K.L., 2011. Situation awareness on the road:
Broström, R., Chilakapati, R.K., Rydström, A., 2009. Intrusiveness of a visual review, theoretical and methodological issues, and future directions.
detection task on secondary and driving task performance. Paper presented at Theoretical Issues in Ergonomics Science. http://dx.doi.org/10.1080/
the First International Conference on Driver Distraction and Inattention, 1463922X.2010.539289.
Gothenburg, Sweden. Salmon, P.M., Lenné, M.G., Young, K.L., Walker, G.H., in press. An on-road network
Endsley, M.R., 1995a. Measurement of situation awareness in dynamic systems. analysis-based approach to studying driver situation awareness at rail level
Human Factors 37, 65–84. crossings. Accident Analysis and Prevention.
Endsley, M.R., 1995b. Towards a theory of situation awareness in dynamic systems. Sanderson, P.M., Verhage, A.G., Fuld, R.B., 1989. State-space and verbal protocol
Human Factors 37, 32–64. methods for studying the human operator in process control. Ergonomics 32
Engström, J., Mårdh, S., 2007. Safe TE Final Report, Publication No. 2007:36. (11), 1343–1372.
Vägverket, Sweden. Stanton, N.A., Chambers, P.R.G., Piggott, J., 2001. Situational awareness and safety.
Engstrom, J., Johansson, E., Ostlund, J., 2005. Effects of visual and cognitive load in Safety Science 39 (3), 189–204. http://dx.doi.org/10.1016/s0925-
real and simulated motorway driving. Transportation Research Part F: Traffic 7535(01)00010-8.
Psychology and Behaviour 8 (2), 97–120. Strayer, D.L., Drews, F.A., 2004. Profiles in driver distraction: effects of cell phone
Fox, M.C., Ericsson, K., Best, R., 2011. Do procedures for verbal reporting of thinking conversations on younger and older drivers. Human Factors 46 (4), 640.
have to be reactive? A meta-analysis and recommendations for best reporting van Winsum, W., Martens, M., Herland, L., 1999. The Effect of Speech Versus Tactile
methods. Psychological Bulletin 137 (2), 316–344. Driver Support Messages on Workload, Driver Behaviour and User Acceptance
Greenberg, J., Tijerina, L., Curry, R., Artz, B., Cathey, L., Grant, P., et al., 2003. (TNO-Report TM-99-C043). Soesterberg, Netherlands.
Evaluation of driver distraction using an event detection paradigm. Journal of Victor, T.W., Harbluk, J.L., Engstrom, J.A., 2005. Sensitivity of eye-movement
the Transportation Research Board, No. 1843. measures to in-vehicle task difficulty. Transportation Research Part F: Traffic
Gugerty, L., 2011. Situation awareness in driving. In: Lee, J., Rizzo, M., Fisher, D., Psychology and Behaviour 8 (2), 167–190.
Caird, J. (Eds.), Handbook for Driving Simulation in Engineering, Medicine and Victor, T.W., Engström, J., Harbluk, J.L., 2009. Distraction Assessment methods based
Psychology. CRC Press, Boca Raton, FL. on visual behaviour and event detection. In: Regan, M.A., Lee, J.D., Young, K.L.
Gugerty, L., Rando, C., Rakauskas, M.E., Brooks, J., Olson, H., 2003. Differences in (Eds.), Driver Distraction: Theory, Effects and Mitigation. CRC Press, Boca Raton,
remote versus in-person communications while performing a driving task. FL.
Paper presented at the 47th Annual Meeting of the Human Factors and Walker, G.H., Stanton, N.A., Young, M.A., 2007. Easy rider meets knight rider: an on-
Ergonomics Society, Santa Monica, CA. road exploratory study of situation awareness in car drivers and motorcyclists.
Harbluk, J.L., Noy, Y.I., Trbovich, P.L., Eizenman, M., 2007. An on-road assessment of International Journal of Vehicle Design 45 (3), 307–322.
cognitive distraction: impacts on drivers’ visual behavior and braking Walker, G.H., Stanton, N.A., Young, M.S., 2008. Feedback and driver situation
performance. Accident Analysis and Prevention 39 (2), 372–379. awareness (SA): a comparison of SA measures and contexts. Transportation
Hughes, P.K., Cole, B.L., 1986. What attracts attention when driving? Ergonomics 29 Research Part F: Traffic Psychology and Behaviour 11 (4), 282–299. http://
(3), 377–391. dx.doi.org/10.1016/j.trf.2008.01.003.
Kass, S.J., Herschler, D.A., Companion, M.A., 1991. Training situational awareness Walker, G.H., Stanton, N.A., Kazi, T.A., Salmon, P.M., Jenkins, D.P., 2009. Does
through pattern recognition in a battlefield environment. Military Psychology 3, advanced driver training improve situation awareness? Applied Ergonomics 40
105–112. (4), 678–687.
Kass, S.J., Cole, K.S., Stanny, C.J., 2007. Effects of distraction and experience on Walker, G.H., Stanton, N.A., Salmon, P.M., 2011. Cognitive compatibility of
situation awareness and simulated driving. Transportation Research Part F: motorcyclists and car drivers. Accident Analysis and Prevention 43, 878–888.
Traffic Psychology and Behaviour 10, 321–329. Wickens, C.D., Hollands, J.G., 2000. Engineering Psychology and Human
Kirwan, B., Kaarstad, M., Hauland, G., Follesoe, K., 1995. See no evil, hear no evil, Performance, third ed. Prentice Hall, Upper Saddle River, NJ.
speak no evil: verbal protool analysis, eye movement analysis, and nuclear Young, K.L., Lenné, M.G., 2010. Driver engagement in distracting activities and the
power plant diagnosis. In: Roberston, S.A. (Ed.), Contemporary Ergonomics. strategies used to minimise risk. Safety Science 48, 326–332.
Taylor & Francis, London, UK. Young, K.L., Salmon, P.M., 2012. Examining the relationship between driver
Klauer, S.G., Dingus, T.A., Neale, V.L., Sudweeks, J.D., Ramsey, D.J., 2006. The Impact distraction and driving errors: a discussion of theory, studies and methods.
of Driver Inattention on Near-Crash/Crash Risk: An Analysis Using the 100-Car Safety Science 50, 165–174.
Naturalistic Driving Study data. Virginia Tech Transportation Institute, Young, K.L., Regan, M.A., Lee, J.D., 2008. Measuring effects of driver distraction:
Blacksburg, Virginia. direct driving performance methods and measures. In: Regan, M.A., Lee, J.D.,
Lee, J.D., Caven, B., Haake, S., Brown, T.L., 2001. Speech-based interaction with in- Young, K.L. (Eds.), Driver Distraction: Theory, Effects and Mitigation. CRC Press,
vehicle computers: the effect of speech-based e-mail on drivers’ attention to the Boca Raton, FL.
roadway. Human Factors 43, 631–640. Young, K.L., Salmon, P.M., Cornelissen, M., in press. Distraction-induced driving
Liang, Y., Lee, J.D., 2010. Combining cognitive and visual distraction: less than the error: an on-road examination of the errors made by distracted and
sum of its parts.. Accident Analysis and Prevention 42 (3), 881–890, doi: http:// undistracted drivers. Accident Analysis and Prevention.
dx.doi.org/10.1016/j.aap.2009.05.001.

You might also like