Content uploaded by Sogand Hasanzadeh
Author content
All content in this area was uploaded by Sogand Hasanzadeh on May 01, 2018
Content may be subject to copyright.
Measuring the Impacts of Safety Knowledge on Construction
Workers’Attentional Allocation and Hazard Detection Using
Remote Eye-Tracking Technology
Sogand Hasanzadeh, A.M.ASCE
1
; Behzad Esmaeili, A.M.ASCE
2
; and Michael D. Dodd
3
Abstract: Although several studies have highlighted the importance of attention in reducing the number of injuries in the construction indus-
try, few have attempted to empirically measure the attention of construction workers. One technique that can be used to measure worker atten-
tion is eye tracking, which is widely accepted as the most direct and continuous measure of attention because where one looks is highly
correlated with where one is focusing his or her attention. Thus, with the fundamental objective of measuring the impacts of safety knowledge
(specifically, training, work experience, and injury exposure) on construction workers’attentional allocation, this study demonstrates the
application of eye tracking to the realm of construction safety practices. To achieve this objective, a laboratory experiment was designed in
which participants identified safety hazards presented in 35 construction site images ordered randomly, each of which showed multiple hazards
varying in safety risk. During the experiment, the eye movements of 27 construction workers were recorded using a head-mounted EyeLink II
system. The impact of worker safety knowledge in terms of training, work experience, and injury exposure (independent variables) on eye-
tracking metrics (dependent variables) was then assessed by implementing numerous permutation simulations. The results show that tacit
safety knowledge acquired from work experience and injury exposure can significantly improve construction workers’hazard detection and
visual search strategies. The results also demonstrate that (1) there is minimal difference, with or without the Occupational Safety and Health
Administration 10-h certificate, in workers’search strategies and attentional patterns while exposed to or seeing hazardous situations; (2) rela-
tive to less experienced workers (<5 years), more experienced workers (>10 years) need less processing time and deploy more frequent short
fixations on hazardous areas to maintain situational awareness of the environment; and (3) injury exposure significantly impacts a worker’s
visual search strategy and attentional allocation. In sum, practical safety knowledge and judgment on a jobsite requires the interaction of both
tacit and explicit knowledge gained through work experience, injury exposure, and interactive safety training. This study significantly contrib-
utes to the literature by demonstrating the potential application of eye-tracking technology in studying the attentional allocation of construction
workers. Regarding practice, the results of the study show that eye tracking can be used to improve worker training and preparedness, which
will yield safer working conditions, detect at-risk workers, and improve the effectiveness of safety-training programs. DOI: 10.1061/(ASCE)
ME.1943-5479.0000526.©2017 American Society of Civil Engineers.
Introduction
Human error is a main contributing factor to accidents in the work-
place, especially among construction workers (Sanders and
McCormick 1998;Garrett and Teizer 2009). The main human error
that leads to accidents and injuries on a construction site is a work-
er’s lack of attention when detecting potential or active hazards,
which subsequently results in the worker’s failure to react properly
(Garrett and Teizer 2009;Rozenfeld et al. 2010). Accordingly, the
identification of variables that impact a worker’s attention and vis-
ual search strategy can have a transformational impact on construc-
tion safety performance.
To study attention, one needs a reliable means of measuring it.
Because visual cues have a direct impact on attentional allocation,
especially in the very first phase of exploration (Hallett 1986), one
scientific way of studying attention is to detect eye-movement pat-
terns. Eye tracking is widely accepted as the most direct and contin-
uous measure of attention, because where one looks is highly corre-
lated with where one is focusing his or her attention (Shepherd et al.
1986;Hoffman and Subramaniam 1995). This strong relationship
between eye movements and cognitive performance manifests in a
variety of domains and explains why eye tracking has been widely
implemented in the fields of neuroscience, psychology, and behav-
ioral research (Richardson and Spivey 2004).
Several studies have highlighted the importance of attention in
reducing the number of injuries in the construction industry,
although few have attempted to empirically measure the attention
of construction workers (Garrett and Teizer 2009;Lopez et al.
2010). Because eye tracking exhibits the immense potential of provid-
ing deeper insights into construction workers’hazard-identification
patterns, researchers have also recently begun to examine potential
applications of this novel technology in studying the attention and
hazard-identification skills of construction workers (Bhoir et al.
2015;Hasanzadeh et al. 2016;Dzeng et al. 2016). Although these
past studies have demonstrated the potential application of eye-
tracking technology in construction safety, they face some limita-
tions. For example, although Bhoir et al. (2015) and Hasanzadeh et
1
Ph.D. Student, Durham School of Architectural Engineering and
Construction, Univ. of Nebraska–Lincoln, 116 Nebraska Hall, Lincoln,
NE 68588. E-mail: smohammadhasanzadeh@huskers.unl.edu
2
Assistant Professor, Durham School of Architectural Engineering and
Construction, Univ. of Nebraska–Lincoln, 113 Nebraska Hall, Lincoln,
NE 68588 (corresponding author). E-mail: besmaeili2@unl.edu
3
Associate Professor, Dept. of Psychology, Univ. of Nebraska–Lincoln,
B82 East Stadium, Lincoln, NE 68588. E-mail: mdodd2@unl.edu
Note. This manuscript was submitted on September 6, 2016; approved
on December 28, 2016; published online on April 12, 2017. Discussion pe-
riod open until September 12, 2017; separate discussions must be submit-
ted for individual papers. This paper is part of the Journal of
Management in Engineering, © ASCE, ISSN 0742-597X.
© ASCE 04017024-1 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
al. (2016) did investigate the feasibility of using eye-tracking tech-
nology in studying hazard-detection skills of construction workers,
they did not investigate variables that might impact the attentional
allocation of construction workers at construction sites. Dzeng et
al. (2016) studied the impact of work experience on worker atten-
tion; however, their study was limited because stimuli consisted of
virtual images of a limited number of hypothetical construction
scenarios and not real images taken from constructions sites.
Furthermore, the images in Dzeng et al. (2016) did not control for
other variables at play in eye tracking (e.g., color, contrast, viewing
time), which is a choice that could confound the outcomes of the
study. Although these studies manifest the promise of applying eye
tracking to construction safety, there remain ongoing gaps in
knowledge about the variables that impact the attentional alloca-
tion of construction workers. Identifying such factors will facilitate
identifying workers who are at greater risk and will thereby further
improve construction safety.
To respond to these past limitations, it is vital to understand the
three major group factors that affect attention, (1) task parameters,
(2) environmental variables, and (3) individual subject characteris-
tics. The first group includes several variables that define the cogni-
tive load required to implement a task, such as difficulty, detectabil-
ity, complexity, and administrative conditions. Task parameters can
also interact with other groups of factors to impact attentional allo-
cation, because environmental and subject characteristics will often
influence the worker’s experience of the task. The second group
includes environmental factors that can affect attentional allocation
that are either extraneous to the task itself (such as noise, tempera-
ture, and situational variables) or intrinsic to the situation (such as
crowded conditions and time of day). It is important to note that the
effects of noise and other environmental factors on attention vary
across studies due to their respective contexts (Ballard 1996); this
reality marks the need for isolating factors and subfactors as much
as possible when studying attention in safety contexts. Finally, the
third group of factors includes subject characteristics, such as demo-
graphics (e.g., gender, work experience, intelligence variations,
past experience), personality traits, and arousal level (e.g., medica-
tion, fatigue, stress) (Ballard 1996;McBride and Cutting 2015).
These subject traits will often impact workers’safety across days
and times, which make them valuable discussion points in the con-
text of construction safety. Combined, these different factors will
interact to impact construction workers’attentional allocation.
Consequently, to reduce the complexity of studying attention in the
context of construction safety, it is important to isolate specific
processes of interest while controlling for others.
In the context of this paper, the focus is on one of the individual
subject characteristics that plays a crucial role in detecting a hazard,
assessing a hazard’s risk, and selecting proper action: safety knowl-
edge (Chua and Goh 2004). Knowledge manifests in the cognition
and/or behaviors of individuals (Argote and Miron-Spektor 2011)
and has been divided into two categories: tacit and explicit (Zhang
and He 2016). Tacit safety knowledge can be acquired by learning
on the job, gaining experience, and being exposed to an injury
(Koskinen et al. 2003;Podgórski 2010;Hallowell 2012). These past
experiences and exposures will be stored as patterns of knowledge
in the brain, and when a worker is subsequently exposed to related
situations he or she will have a memory-based intuition to respond
correctly (Zhang and He 2016). On the other hand, explicit knowl-
edge can be delivered through safety training (Aboagye-Nimo et al.
2012) or captured from existing theories in books, safety records,
and guidelines (Hadikusumo and Rowlinson 2004). Regardless of
the type of safety knowledge, there is a consensus in the literature
that safety knowledge can be acquired via training, work
experience, and exposure (Podgórski 2010;Argote and Miron-
Spektor 2011;Aboagye-Nimo et al. 2012). Therefore, it is likely to
play a crucial role in guiding worker attention, and it underpins the
hazard recognition and responsiveness of workers.
To determine the importance of safety knowledge on con-
struction workers’safety, this paper investigates the relation-
ship between safety knowledge, attention, and hazard recogni-
tion. More specifically, the main objective of this study is to
track workers’eye movements and visual search strategies to
assess the impact of safety knowledge acquired from training,
work experience, and past injury exposure on construction
workers’attentional allocation toward hazards. To control for
other influences on attention (i.e., task parameters and environ-
mental variables), the research team conducted a laboratory-
based eye-tracking experiment using images taken from real-
life scenarios that manifested a variety of construction hazards.
Consequently, this study empirically establishes the relationship
between worker’s subjective safety knowledge characteristics,
their attention, and their safe behavior, filling an ongoing gap in
knowledge about the impact of safety knowledge on workers’
cognitive processes, and, specifically, attention.
Attention
The cognitive process of attention is a vital component of this study
and one that has been advanced by the expansion of eye-tracking
research. Unfortunately, a complete review of eye-tracking research
is beyond the scope of the present paper because eye tracking (1) has
been used in numerous domains, (2) has been used in a variety of
ways, and (3) affords the researcher the ability to choose from doz-
ens of different measures dependent on their critical question of in-
terest. Because of the variety of concepts available when examining
attention via eye tracking, it is important to provide an inclusive
overview of these topics so that the reader has the appropriate con-
text needed to understand this paper’s approach and conclusions.
Cognitive processing comprises how the mind receives, stores,
and uses information. The process of understanding and using in-
formation is highly related to many cognitive processes, includ-
ing attention, perception, memory, language, imagination, and
decision making (Anderson 2005). Before the twentieth century,
the roots of this process traced back to philosophy and physiol-
ogy. However, in the early- to mid-twentieth century, with the de-
velopment of information-processing approaches, the study of
cognition extended beyond psychology and behaviorist perspec-
tives. The findings of Chomsky’sstudy(Chomsky 1959) revealed
that understanding cognitive processes is critical to understand-
ing and assessing behavior. Cognitive processes select which
pieces of information to attend and which to ignore, and this, in
turn, manifests in human behavior. Because the main objective of
this research is to assess hazard detection and to study the unsafe
attentional distribution of construction workers with various
training and experience levels, it was necessary to measure work-
ers’attention as a primary cognitive process.
Attention, an important element in many cognitive tasks, has
been studied for over a century in many different domains. Defining
attention is difficult because it overlaps with many other cognitive
processes (McBride and Cutting 2015). One early proposed defini-
tion of attention is the focus of consciousness to a particular stimu-
lus while ignoring other distracting objects in the environment
(James 1913). Because humans are finite beings, their capacity for
information processing is limited, which influences their attentional
abilities. Moreover, emotional arousal, task difficulty, and an
observer’s interest/motivation in a task can significantly affect the
© ASCE 04017024-2 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
capacity of mental resources and attention abilities (Duchowski
2007). Because of the limited capacity of human information proc-
essing as well as the various environmental and individual charac-
teristics at play, this study examines attention in construction safety
according to how workers select to attend to or ignore different
types of hazards while working in complex and dynamic construc-
tion sites.
The authors conducted a comprehensive literature review of
studies related to cognition and attention and found that attention,
as a prominent process of cognition, can be classified into different
levels and that there are various approaches to studying attention.
Salient results of this review are provided in the following sections.
Levels of Attention
Sohlberg and Mateer (1989) identified four levels of attention: sus-
tained, selective, divided, and alternating. Sustained attention, or
vigilance, is the ability to devote one’s whole attention to a single
but complex stimulus set over extended periods, without being dis-
tracted, to detect infrequent targets (Parasuraman et al. 1998).
Sustained attention represents a basic attentional function that speci-
fies the efficacy of one’s information-processing capacity and the
higher functions of attention (e.g., selective and divided attention).
Because the eyes need to retrieve informationfrom the visual field to
perceive it, selective attention is considered the gateway to con-
sciousness (Theeuwes et al. 2009). The first step in the selective pro-
cess is to choose what to attend. Eyes and minds do not have enough
capacity to focus on every object in the field; therefore, they must
select what is most important to process. Observers concentrate on a
stimulus using selective attention (focus on only one stimulus) or di-
vided attention (splitting attention between two or more tasks/
objects) (Rehder and Hoffman 2005). Finally, alternating attention
represents one’scognitiveflexibility, or the ability to shift attention
among different tasks having different cognitive requirements
sequentially (Sohlberg and Mateer 1989). Originally, the intent of
this study was to assess workers’selective attention, but because
construction is such a complex environment with simultaneous
ongoing activities, the focus extended to assess both selective and di-
vided attention across various hazards in a scene.
Modeling Attention
Generally, visual attention plays a crucial role in the control of eye
movements, perception, learning, memory, and other interactions
with the visual world, although people are not always aware of it
(Bisley 2011). A major distinction that has guided all research in
the area of cognition and information processing is whether atten-
tion is goal driven (top down) or stimulus driven (bottom up). In
goal-driven visual attention, observers form deliberate strategies
and intentions for controlling their attention. The top-down model
is used when the task has a strong influence on where the observer
attends and looks (Yantis 1998). In contrast, stimulus-driven atten-
tion occurs when salient items in the environment capture attention
independent of the observers’intent. For example, people have
been shown to initially fixate on salient attributes of the image
related to color or contrast, and the initial scanning process is not
necessarily relevant to the observer’s perceptual goals (Todd and
Kramer 1993). More generally, top-down and bottom-up attentional
models interact with each other, and it is assumed that any observer
viewing a scene has a complex combination of expectations and
goals related to processing that scene (Chun and Wolfe 2001).
Although current conceptualizations of visual attention acknowl-
edge the strength of bottom-up processing (Torralba et al. 2006), it
is important to consider top-down influences in estimating attention
allocation, because in the real world, attention tends to be more task
relevant and goal driven than just free viewing of scenes (Jie and
Clark 2008). Moreover, a recent study by Vecera et al. (2014) has
demonstrated that goal-driven attentional control emerges with
increased experience with a task, whereas with relatively little expe-
rience attentional control tends to be more stimulus driven. In this
study, the research team focused on top-down influences by design-
ing an experiment in which workers were asked to purposefully
search for hazards in each image. It was anticipated that workers
would attend to different parts of the image with a level of intention-
ality that corresponded to their safety knowledge, so this type of
approach allowed for the study of the impacts of their safety knowl-
edge on their attentional modeling.
Attention and Eye Tracking
At any given time an environment contains much more perceptual
information than can be processed. The previously mentioned
evidence illustrates that what a person sees determines to what
that person is going to attend. If a hazard grabs a worker’s atten-
tion, the worker becomes an active seeker and processor of infor-
mation that is able to interact intelligently with the environment
and consider protective measures. A growing number of psycho-
logical and neuropsychological studies have demonstrated the
close relationship between attention and eye movements (Sun
et al. 2008). In one of the seminal studies, Yarbus (1967)showed
that records of observers’eye movements reflect human attention
and thought processes. The simple assumption in most attention
studies using eye tracking is that by tracking someone’seye
movements, one is able to follow the observer’s general path of
attention. This process will reveal both what simply drew the
observer’s attention and what fully captures their attention, which
in turn will give researchers clues about how the person perceived
the scene. Such insights help reveal perhaps one of the most im-
portant functions of attention—to guide fixations toward events
that are relevant to ongoing behavior (Duchowski 2007).
In the last decade, the use of eye tracking has been flourishing
in studies related to human-computer interaction and usability
research (Jacob and Karn 2003). On a somewhat smaller scale, eye-
tracking research also has been applied to transportation (Suh et al.
2006), driving (Palinko et al. 2010), aviation (Lavine et al. 2002),
marketing (Wedel and Pieters 2008), nuclear power control rooms
(Ha and Seong 2009), medicine (Zheng et al. 2011), and petrochem-
ical control rooms (Ikuma et al. 2014). Eye tracking in these fields
has been used mostly to estimate cognitive load, analyze user
behavior, reveal differences in aptitude and expertise, diagnose neu-
rological disorders, and analyze gaze control.
The existing literature on hazard detection and attentional alloca-
tion using eye-tracking technology has mainly focused on automo-
tive driving. These studies use visual behavior as an index of hazard-
detection skill when drivers are exposed to potential hazards using
static pictures or scenarios (Underwood et al. 2005), a movie-based
simulator (Underwood et al. 2011;Borowsky et al. 2012;Mackenzie
and Harris 2015), or field experiment (Sun et al. 2016). Eye tracking
has also been used to study the impact of age (Romoser et al. 2005),
experience (Underwood 2007), and cognitive load (such as cell
phone conversations) (Muttart et al. 2007) on drivers’safety per-
formance. Furthermore, to improve the hazard-detection skills of
drivers, eye-tracking technology has been used to evaluate the effec-
tiveness of interactive training processes (Underwood 2007;Fisher
et al. 2007;Pradhan et al. 2009). Although these studies show the
potential relationship between eye movement, hazard detection, and
safety knowledge, the application of this technology remains
© ASCE 04017024-3 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
unexplored in the field of construction safety. Thus, the study of
attention using eye-tracking technology may elucidate the safety be-
havioral paradigms of workers in dangerous situations on construc-
tion sites.
Eye-Tracking Metrics
Each individual research project requires the choice of eye-tracking
metrics that are relevant to the tasks and appropriate for attentional
allocation and cognitive process analysis. The process used for
identifying fixations and monitoring saccades (the rapid movements
of the eye between fixation points) in an eye-tracking protocol is the
most essential part of eye-movement data analysis (Salvucci and
Goldberg 2000;Duchowski 2007). Minimizing the complexity of
eye-tracking data by identifying fixations and removing raw sac-
cade data can improve researchers’understanding of attentional
allocation and visual processing behavior, because little visual proc-
essing is achieved during saccades, and findings related to saccades
have not been used in many research applications (Salvucci and
Goldberg 2000). To explore the determinants of ocular behavior,
many derived metrics stemming from these basic measures (fixation
and saccade) have been incorporated in different eye-tracking stud-
ies as dependent variables, including fixation-derived and saccade-
derived metrics.
Areas of Interest
The first step in assessing eye-movement parameters and establish-
ing dependent measures is to define an area of interest (AOI) in
each image used in the study. AOIs are objects and locations of in-
terest within a scene as defined by the research team (Jacob and
Karn 2003). For example, in marketing-related studies, specialists
might be interested in knowing the total time it takes each observer
to view the desired target (brand logo) on a company’s home page
(Goldberg et al. 2002). Further, in the study of Weibel et al. (2012),
to understand dynamic allocation of pilot attention, the AOIs were
flight instruments in the cockpits of commercial airplanes. In the
present study, the AOIs are defined as the different types of hazards
that appeared in the construction scenario images, including fall
hazards, struck-by hazards, and housekeeping hazards.
Fixation-Derived Metrics
Fixation is defined as a relatively stationary eye position with a rela-
tively short minimum duration (often 100–200 ms, although longer
fixations can also be observed as a function of an individual’s proc-
essing goals). Although fixations can be interpreted differently in
various studies depending on their context, fixation-count and gaze-
duration (consecutive fixations) metrics have generally been con-
sidered to assess the depth of cognitive processing and the distribu-
tion of attention (Zhao et al. 2014). For example, Hauland (2003)
indicated that a longer gaze that falls into the specific AOI before an
incident happens can be used as a situation awareness anticipation
measure. Three commonly used fixation-derived metrics in eye-
tracking studies (Bhoir et al. 2015) are (1) first fixation time (the
amount of time, i.e., in milliseconds, that passes following the
image’sfirst appearance on the screen until the observer first fixates
on an AOI), (2) dwell percentage (relative to the amount of time
spent viewing an image, the proportion of time in which gaze was
fixated on an AOI), and (3) run count (the average number of times
that each participant returns their attention to an AOI).
Saccade-Derived Metrics
Saccades, defined as quick eye movements from one location to
another, vary in duration but often take about 25–150 ms, depend-
ing on the amplitude of the saccade. In most studies, researchers
remove the raw saccade data (Salvucci and Goldberg 2000),
whereas in other studies they measure saccade-derived metrics to
understand the observer’s searching duration and to evaluate the
design of the interface (Fuchs 1971;Pan et al. 2004). For example,
in computer interface and usability studies, larger saccade ampli-
tude indicates a well-designed interface with sufficient cues to ena-
ble users to rapidly find the desired targets (Goldberg et al. 2002).
Another study showed that regressive saccades during reading are
associated with comprehension difficulties (Just and Carpenter
1980).
Visualization of Eye Movement Data
Eye-tracking provides a large amount of data, and the visualization
of eye-movement patterns can provide insight through a compre-
hensive statistical analysis. Visualization techniques commonly
used for representing eye-tracking data are heat maps and scan
paths (Raschke et al. 2014). A heat map is a two-dimensional visu-
alization in which all fixation values that were analyzed are repre-
sented in colors (Bojko 2009). A heat map can be created for an
individual or for a group of people. In this study, the fixation count
heat map was used to define AOIs and to better compare visual
attention of construction workers across the scene. On the other
hand, a scan path is a compelling visualization of eye movements
defined as a spatial arrangement of a sequence of saccade-fixation-
saccade (Poole and Ball 2006). It has been widely argued in previ-
ous studies that scan paths reveal considerable information about
visual attention and other underlying cognitive processes involved
in eye movements (Laeng and Teodorescu 2002;Foerster and
Schneider 2013). In this study, in addition to statistical analysis out-
puts, the research team used heat maps and scan paths to study vis-
ual search strategies and cognitive processes of workers in each
scene.
Point of Departure
Construction workers need attention to not only complete their
activities, but also to enrich their conscious awareness of the
entire dynamic environment at a construction site. Training can
make a person phenomenally conscious of certain hazards, but
attention is also critical to detect, perceive, and react to hazards
appropriately. Because attending to something elevates conscious
perception of it (De Brigard 2012), workers’lack of attention
may expose them to hazardous situations more frequently.
Therefore, investigating variables that might impact a worker’s
attention can be of great importance.
This study departs from the current body of knowledge because
it is one of the earliest studies to empirically measure construction
workers’attention using eye-tracking technology. Although some
previous studies have been conducted using eye-movement data to
measure the situation awareness of air-traffic controllers (Hauland
2003), a limited number of studies have measured construction
workers’attention empirically using eye-tracking technology
(Bhoir et al. 2015;Dzeng et al. 2016). Additionally, this project
focused on controlling for other attentional factors (i.e., task param-
eters and environmental variables) to more thoroughly examine the
chosen factor, specifically safety knowledge acquired from training,
work experience, and past injury exposure. A primary objective of
© ASCE 04017024-4 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
this study is to investigate how workers with different levels of
work experience, training, and injury exposure (independent varia-
bles) attend to a complex construction scene. Thus, to achieve this
overarching objective, four null hypotheses were defined and tested
(Table 1).
Research Methods
The research hypotheses were tested by collecting information on
two types of variables: independent variables (safety knowledge,
i.e., training, work experience, injury exposure) and dependent
variables (eye-tracking metrics). The independent variable data
were collected via a questionnaire, which, in addition to general
questions related to the participants’background and years of ex-
perience in the construction industry, asked respondents whether
they had received the Occupational Safety and Health Administra-
tion (OSHA) 10-h training (formal) or any on-site safety training
(informal), and whether they had previous injury exposure. The de-
pendent variable data were collected in a laboratory experiment using
eye-tracking equipment, described in the following sections.
Apparatus
An EyeLink II (manufactured by SR Research Ltd., Kanata, ON,
Canada), with a high spatial resolution and a sampling rate of 500
Hz, tracked and recorded the participants’eye movements to deter-
mine where they attended. The EyeLink II is a video-based eye-
tracking system that uses cameras mounted on the headset to docu-
ment the path of a viewer’s focus. Participants completed the
experiment seated approximately 45 cm from the computer screen
on which they observed images (the general procedure can be seen
in the left half of Fig. 1).
Participants
Participants were all construction workers who were invited to the
study in one of the following three ways: (1) an invitation flyer was
posted at construction sites in Lincoln and Omaha, Nebraska; (2)
researchers extended invitations by stopping by construction compa-
nies’main offices and contacting facility managers at the University
of Nebraska-Lincoln; and (3) a flyer with a one-page summary of
the research project was sent to Associated Builders and Contractors
members and department advisory boards. As a result, a total of 31
Table 1. Tested Null Hypotheses
Hypothesis Tested null hypothesis
Gaining more experience improves workers’attentiveness to
hazards
Null hypothesis 1 (H
1
): Workers’work experience (years of experience) has no impact on
their attentiveness to hazards on a construction site
Receiving training (formal or informal) improves workers’
attentiveness to hazards
Null hypothesis 2 (H
2
): Workers’safety training (formal or informal) has no impact on their
attentiveness to hazards on a construction site
Exposure to injury impacts workers’attentiveness to hazards Null hypothesis 3 (H
3
): Workers’past injury exposure (either personal injury or seeing
somebody else injured) has no impact on their attentiveness to hazards on a construction site
By keeping training as a control variable, gaining more
experience improves trained workers’attentiveness to hazards
Null hypothesis 4 (H
4
): Work experience (years of experience) of trained workers has no
impact on their attentiveness to hazards on a construction site
•Years of experience
•Safety training
•Past injury exposure
•First fixation time
•Dwell percent
•Fixation count
•Run count
ET Metrics
•Fall hazards
•Ladder hazards
•Struck-by hazards
•Housekeeping
hazards
Areas of Interest
•Heatmap (attentional
distribution)
•Scanpath (cognitive
process)
•Real-time map of
worker eye
movement
Visualization
Statistical analyses
and permutation
simulations
Pool of construction
scenario images
Eye-Tracking Lab Experiment
Professional
safety
managers
focus group
Overt attentional
shift
Controlling
exogenous factors
Goal-driven
attentional
modeling
Experimental Design
Analyses and Evaluation Implications
Data Collection
Real-time
tracking of eye
movements
45cm
Quantitative Analyses
Qualitative Analyses
Contribution to Academia
Contribution to Practice
Develop advanced
safety training
Measure effectiveness
of the training
Identifying hidden
hazards and at-risk
workers
Identify limitations and
potential future
research areas
Characterize variation
in construction
workers’ attention
Verify the role of
cognitive process in
construction safety
Setting up and
calibrating
apparatus
(remote
eye-tracker)
Question:
Can you
identify any
hazard?
Recruit
construction workers
Demographic Survey
35 scenario images
Divided
level of attention
Independent Variables
Fig. 1. Research framework (images courtesyof David Ausmus, with permission)
© ASCE 04017024-5 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
construction laborers (30 males, 1 female) were recruited for the
study. Because eye tracking is experimental research, the sample
size cannot be as large as it would be for survey-style studies.
However, the size of the sample met the parameters of Pernice and
Nielsen (2009), who stipulated that sample sizes for eye-tracking
studies vary greatly,ranging from 6 for a qualitative study to 30 fora
quantitative study. Given the number of subjects used in this study
and the number of trials performed by each, the sample is robust
enough to match the size recommended for eye-tracking studies.
The participants were mostly young to middle-aged (93% were
between the ages of 20 and 55), 35.5% of the participants had fewer
than 5 years of experience, 25.8% had between 5 and 10 years of ex-
perience, and 38.7% had more than 10 years of experience in the
construction field. Sixty percent of the participants had received the
OSHA 10-h training, 16% had received informal or on-site safety
training, and the remaining 24% had not received any safety train-
ing. A total of 69.9% of participants reported that they had been
exposed to an injury on the jobsite. All participants had normal or
corrected-to-normal vision. The experiment was conducted in a sin-
gle 30-min session for each worker. The data for four participants
were omitted from the analysis because acceptable levels of calibra-
tion on the eye tracker could not be achieved. The final analyses
were based on the data from the remaining 27 workers. All partici-
pants received gift cards as compensation.
Design and Procedure
The eye-tracking experiment consisted of the presentation of 35
construction site images ordered randomly, each of which showed
multiple hazards varying in safety risk (four examples of different
scenario images are shown in Fig. 2). Construction site images
belonged to different private residential and commercial projects
taken from real construction sites across the United States. Scenario
images covered different types of activities, including but not lim-
ited to site work, roofing, lifting materials, finishing, erecting struc-
tures, and painting. Each image appeared on the screen for a maxi-
mum of 20 s. The participants donned the eye-tracking headset,
which was calibrated before each session. Subjects were to scan
each image and look for potential hazards. Using a video-game
remote control, they then reported whether or not they found any
hazards by pushing Button A for Yes or Button B for No. The sys-
tem gathered eye-tracking data from the moment the image initially
appeared through to the moment the participant pushed A or B.
Once the participant made a selection, he or she was asked about the
number of hazards identified, and then the screen changed to reveal
the next image.
In addition to the eye-tracking experiment, participants took a
short survey to collect demographic information, including age,
gender, nationality, years of experience, obtained certifications,
training, and injury exposure. All procedures were approved by
the University of Nebraska-Lincoln Institutional Review Board.
Defining AOIs
To extract eye-tracking metrics, an AOI or an overlap or near-
overlap of the stimulus (a hazardous situation) and the fixation
points needed to be defined. To identify the AOIs, the research
team first studied participants’scan paths and heat maps for each
picture to understand where they were looking. In the second step,
a focus group consisting of five safety managers independently
reviewed and discussed the original pictures without the heat maps
to identify hazards in each scenario image. All safety mangers
were certified and had at least 10 years of experience in residential
and commercial projects. The focus group of safety managers was
also asked to express the risks they perceived in each scenario.
Thus, for each construction scenario image, they indicated the
expected frequency of each injury type (categorized based on
injury severity). Overlaying and comparing the results of Steps
One and Two enabled the research team to finalize the AOIs. In total,
177 hazardous situations (AOIs) in 35 pictures were identified.
Various types of hazards manifested in the scenario images: lad-
der related, fall to lower level related, fall protection system related,
housekeeping, struck by, caught-in/between, and electrical hazards.
However, due to sample and visibility limitations, the research team
had to remove caught-in/between and electrical hazards from the
analysis. First, there was only a single image in which a caught-in/
between hazard existed, which made the sample too small for this
type of hazard. Concurrently, because the main type of electrical
hazard in the images was powerlines, which are very thin objects,
the task of locating fixations on such small and narrow AOIs proved
very challenging. Accordingly, the research team decided to remove
electrical hazards from the analysis.
Therefore, the research team decided to focus on the following
AOIs: fall to lower level (i.e., a worker is in the proximity of an
unprotected building edge or roof, unguarded roof and floor open-
ings, scaffolding, skylights); fall-protection systems (i.e., misuse of
lanyard and other fall-protection systems); ladder related (i.e.,
improper use of ladders, such as inappropriate type and length of
ladder, ignoring ladder extension rules, unstabilized ladder, unse-
cured straight ladders, unsafe behavior of workers who are working
on the ladder), struck-by hazards (i.e., having the probability of
being struck by heavy equipment or falling objects like tools, or
collapsing masonry or concrete walls); and housekeeping related
(i.e., slippery conditions of working and walking surfaces, unsafe
material storage, unsanitary conditions of work environment).
These types of hazards are among the most typical safety risks that
lead to accidents.
Analysis Procedures
To test the research hypotheses, the research team separated the par-
ticipants into groups based on the independent variables (training,
work experience, and injury exposure). Inferential statistical
Fig. 2. Examples of scenario images (images courtesy of David
Ausmus, with permission)
© ASCE 04017024-6 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
analyses illustrated how different levels of these variables impacted
the eye-tracking metrics (dependent variables), which indicated the
attentional distribution of construction workers. To study the impact
of training on eye-tracking metrics, participants were grouped based
on whether they had previously received safety training. Regarding
work experience, 5 and 10 years of experience in the construction
industry were considered the cutoffs that determined low- and high-
experienced workers. In terms of injury exposure, people were clas-
sified based on whether they had been injured and/or had seen
someone else injured on a construction site.
The research team linked the eye movement data of participants
to individual AOIs for each hazard to enable analysis. The choice of
an appropriate eye-tracking metric depends on the research context.
According to the results of a 2015 pilot study conducted by Bhoir
et al. (2015) three fixation-related measures were preferred to serve
as indicators of gaze, because they provide unique insight into vis-
ual attention across AOIs. First, the time of first fixation on the
defined AOIs was calculated. This metric reveals which hazards
captured workers’attention more quickly than others (e.g., how
quickly was the AOI fixated). Second, because the time each partic-
ipant spent scanning each picture was different, the gaze (consecu-
tive fixations) on the target AOI was divided by the total duration of
all gazes, yielding the dwell-percentage ratio. The average percent-
age of all workers in the same group indicated that the dwell per-
centage on each AOI was a good indicator of which AOIs (or types
of hazard) captured workers’attention more than others. Finally,
the run-count metric revealed the mean number of times workers
returned their attention to each AOI to further investigate the situa-
tion, indicating the extent to which workers perceived the AOI to be
dangerous. The EyeLink Data Viewer was used to analyze the two-
dimensional eye-movement patterns of participants. After grouping
participants based on independent variables and extracting the eye-
tracking data, the research team faced the challenge of selecting an
appropriate statistical test.
Statistical Method
To compare the average of two or more groups, parametric tests,
such as the t-test and ANOVA, are the most commonly used techni-
ques. To use parametric tests, one needs to satisfy several require-
ments, such as testing a random sample from a population, normality,
and equality of variances (Anderson 2001;Field 2013). However, the
majority of laboratory experiments in psychology and physiology are
not based on a random sample from a population, have a small num-
ber of subjects, and often do not meet the requirement of distribu-
tional assumptions in the parametric test (Ludbrook 1994;LaFleur
and Greevy 2009). To overcome these challenges, statisticians have
suggested the use of randomization techniques (Edgington 1995;
Manly 1997;Ludbrook and Dudley 1998;Gleason 2013).
Randomization provides a reliable alternative test when sample
sizes are small, when the sample distributions are nonnormal (either
because of outliers or skewed data), when the data have mixed dis-
tributions, or when the data were not collected at random (Adams
and Anthony 1996;Eudey et al. 2010). Randomization statistical
methods (bootstrap and permutation simulation) are computer-
intensive techniques using reshuffling and resampling to build large
samples (e.g., 1,000 samples) from original data and obtain p-values
based on created distributions (Adams and Anthony 1996;
Anderson 2001). Therefore, randomization statistical techniques
can provide a higher power than other nonparametric techniques,
because randomization statistical techniques use the actual data
rather than ranks, which are used in nonparametric techniques
(Edgington 1995;Adams and Anthony 1996;Ludbrook and Dudley
1998;Drummond and Vowler 2012;Gleason 2013).
The permutation test is a randomization technique that was intro-
duced in the early twentieth century by R. A. Fisher, but it only
recently became more popular and practical due to decreases in
computational costs (LaFleur and Greevy 2009;Gleason 2013).
The fundamental idea behind permutation simulation is to generate
a reference distribution by recalculating data statistics using resam-
pling (Berger 2000). In other words, the permutation tests calculate
the probability of obtaining a value equal or more than the observed
value of a test statistic after randomly shuffling data several times
(Anderson 2001;Ernst 2004). Using permutation simulation omits
the randomized resampling bias found in the bootstrapping
approach, which means that, in most cases, the permutation simula-
tion of actual data is more powerful than the bootstrapping approach
(LaFleur and Greevy 2009).
As in other laboratory physiological experiments (Ludbrook
1994;LaFleur and Greevy 2009), this study faced some limitations
that inhibited the application of parametric statistical techniques.
First, it was impractical to randomly select a sample of participants
from a broad population because participants needed to be present
in a laboratory at the University of Nebraska’s campus; thus, the
construction workers who were selected for participation were from
a population of workers near Lincoln and Omaha. Second, the sam-
ple size of the study was small, which impacts the power of the
parametric techniques. Third, there was a mixed-normality across
group distributions. Because parametric statistical methods are sen-
sitive to the violation of assumptions, each of these characteristics
obviated the use of traditional parametric statistical techniques.
In response, the research team decided to use a randomization
technique. Among available randomization techniques, permutation
simulations were used (1) to provide higher power even given the
nonnormality and mixed-normality across group distributions and
the small sample size and (2) to obtain more robust outcomes in the
presence of outliers and missing data. Because the permutation sim-
ulations are based on permutations of actual data, the influence of
extreme data points is reduced compared with parametric counter-
parts, which are affected considerably by the presence of outliers.
To perform the permutation statistical simulations, the research
team used the Deducer package in Java Graphical User Interface
(GUI) for R 1.7-9 of the open-source statistical package R version
R2.15.0 (R Development Core Team). The permutation simulations
were performed at a level of 95% statistical significance (p<0.05)
and 90% moderately statistical significance (p<0.1). The results of
these statistical analyses appear in the following sections.
Results and Analysis
Table 2describes the various groupings of participants. After
extracting the eye-movement metrics (run count, dwell percentage,
and first fixation time) for each participant on AOIs, the research
team then performed the statistical analyses described earlier.
Because inferential statistics are important, permutation simulations
were run to compare groups’means in terms of eye-movement met-
rics across different types of hazards. Significant results at 0.05 and
0.1 alpha levels are discussed in more detail.
Work Experience and Worker Attentiveness
Regarding work experience, 38.7% of the respondents had more
than 10 years of experience, 25.8% had between 5 and 10 years of
experience, and 35.5% had fewer than 5 years of experience in the
construction field. To magnify the potential difference between the
groups, the research team compared the group with fewer than 5
years of work experience with the group possessing more than 10
© ASCE 04017024-7 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
years of experience. The descriptive statistics relating eye move-
ment metrics to work experience are provided in Table 3. Workers
with less experience (Group A
1
) had lower first fixation times on
hazardous areas; however, workers with more experience (Group
B
1
) had higher run counts, meaning they returned their attention to
hazardous areas in images more often. In addition, more experi-
enced workers (Group B
1
) needed less processing time for ladder
and fall-to-lower-level hazards, as demonstrated by their shorter
dwell time percentage; experienced workers (Group B
1
) also gazed
for a longer duration on fall-protection systems and housekeeping
hazards.
In total, 18 (three eye movement metrics for six types of hazards)
permutation simulations were each run 1,000 times to compare eye-
tracking metrics between groups with different levels of experience
across different types of hazards (significant differences are indi-
cated on Table 3). Contrary to the null hypothesis H
1
, results indi-
cated that work experience has a significant impact on worker atten-
tiveness to hazards on a construction site. The results showed that
more experienced workers (Group B
1
) returned their attention more
frequently to hazardous areas related to fall-protection systems
(Welch’st=−2.862; p= 0.015 <0.05), ladders (Welch’st=
−2.336; p= 0.03 <0.05), struck-by hazards (Welch’st=−3.906;
p= 0.002 <0.05), and housekeeping hazards (Welch’st=−1.845;
p= 0.088 <0.1). In addition, the results indicated that first fixation
times for hazardous areas did not differ significantly among less
experienced workers (Group A
1
) and more experienced workers
(Group B
1
). Moreover, eye-movement patterns of experienced
workers (Group B
1
) showed that they tended to spend significantly
more time examining hazardous areas related to fall-protection sys-
tems in images (Welch’st=−1.740; p= 0.099 <0.1).
Training and Worker Attentiveness
Sixty percent of the participants claimed they had received the
OSHA 10-h training, 16% of the participants had informal or on-
site safety training, and the remaining 24% had no safety training at
all. To test whether having safety training could affect a worker’s
attentional allocation and hazard-detection skills, the participants
were divided into two groups: people who had received safety train-
ing, including the OSHA 10-hour training; informal or on-site
safety training (Group A
2
); and people who had not received any
safety training at all (Group B
2
). The descriptive statistics for these
groups’eye movement metrics across AOIs are summarized in
Table 4.
The descriptive analysis revealed that trained workers (Group
A
2
) were slower to fixate on hazardous areas compared with work-
ers who had not received any safety training (Group B
2
). The results
of the run-count descriptive analysis showed that these trained
workers (Group A
2
) also tended to return their attention to hazard-
ous areas (e.g., ladder, fall-protection systems, struck-by hazards,
housekeeping hazards) more often, excluding falls-to-lower-level
hazards. Moreover, trained workers (Group A
2
) generally dwelt
less on hazards comparedwith Group B
2
. However, trained workers
spent more time examining ladder, struck-by, and housekeeping
Table 2. Groupings of Workers Based on Background: Group A versus Group B
Null
hypothesis Group A Group B
H
1
(A
1
) Workers who had fewer than 5 years of work experience (B
1
) Workers who had more than 10 years of work experience
H
2
(A
2
) Workers who had received safety training (formal or informal) (B
2
) Workers who had not received any safety training (formal or
informal)
H
3
(A
3
) Workers who had been injured on the jobsite (B
3
) Workers who had not been injured on the jobsite
(A
4
) Workers who had seen someone injured on the jobsite (B
4
) Workers who had not seen someone injured on the jobsite
(A
5
) Workers who had been injured and/or had seen someone injured
on the jobsite
(B
5
) Workers who had not been injured and/or had not seen someone
injured on the jobsite
H
4
(A
6
) Workers who had trained and had more than 10 years of
experience
(B
6
) Workers who had trained and had fewer than 5 years of experience
Table 3. Impact of Work Experience on Workers’Attentional Allocation across Different Types of Hazards
Hazards (AOIs) Group
First fixation time (ms) Dwell percentage Run count
NMean SD MD Mean SD MD Mean SD MD
General A
1
9 3,196.344 631.912 3,211.060 0.104 0.009 0.105 1.833
a
0.331 1.805
B
1
11 3,671.524 1,082.945 3,767.006 0.010 0.011 0.099 2.235
a
0.469 2.229
Ladder A
1
9 3,225.971 775.104 3,326.000 0.107 0.024 0.095 2.111
a
0.503 1.889
B
1
11 3,607.659 1,319.006 3,305.647 0.105 0.018 0.096 2.606
a
0.695 2.667
Fall to lower level A
1
9 1,181.896 306.796 1,101.116 0.266 0.047 0.276 3.788 0.886 3.535
B
1
11 1,412.959 586.633 1,125.949 0.240 0.050 0.250 4.402 1.006 4.326
Fall-protection system A
1
9 3,620.371 824.743 3,434.400 0.079
b
0.008 0.079 1.539
a
0.353 1.532
B
1
11 4,087.963 1,085.257 3,629.250 0.086
b
0.010 0.085 2.054
a
0.452 2.170
Struck by A
1
9 4,890.770 964.985 5,262.824 0.059 0.020 0.066 1.269
a
0.294 1.269
B
1
11 5,189.599 1,965.198 4,920.750 0.071 0.014 0.069 1.969
a
0.497 1.846
Housekeeping A
1
9 3,549.130 779.484 3,562.800 0.078 0.015 0.082 1.496
b
0.327 1.400
B
1
11 3,798.890 1,437.093 3,777.000 0.080 0.024 0.086 2.012
b
0.637 2.200
Note: N= number of subjects in each group; SD = standard deviation in each group; MD = median in each group.
a
p≤0.05.
b
p≤0.1.
© ASCE 04017024-8 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
hazardous areas. The reasons behind these findings, to be discussed
later, may relate to different affective factors, such as workers’
background and work experience.
To investigate the differences between these groups, 18 series
of permutation simulations were each run 1,000 times. The results
of these simulations meant that the authors could not reject the
null hypothesis H
2
: safety training (formal or informal) has no
impact on worker attentiveness to hazards on a construction site.
With the resulting p>0.05, it was found that training alone has a
minimal (nonsignificant) impact on workers’attentiveness to haz-
ards. The only significant effect of informal or on-site safety
trainingwasonidentifyingstruck-byhazardsinimages(Welch’s
t= 2.693; p-value = 0.032).
Past Injury Exposure and Worker Attentiveness
Among the 27 participants who had reliable eye-tracking data,
some of them did not answer the survey questions about their past
injury exposure (i.e., four participants did not answer whether they
had been injured and three participants did not answer whether they
had seen someone injured on a jobsite). Of the remaining 23 work-
ers, 13 had been injured and the other 10 stated that they had never
been exposed to any kind of injuries at a jobsite. To better under-
stand the impact of injury exposure on attentional allocation of
workers, the research team divided the participants into the groups
identified in Table 2. Descriptive statistics of eye movement metrics
across AOIs are presented in Tables 5–7. For further investigation,
72 permutation simulations comparing the previously mentioned
groups were each run 1,000 times.
The descriptive statistics presented in Table 5demonstrate the
impact on workers’attentional allocation and hazard-detection
capabilities of having been or not having been injured (Group A
3
versus Group B
3
). Although workers in Group B
3
fixated more
quickly on most hazardous areas, detailed analysis of the first fix-
ation time did not provide any meaningful discrimination
between groups (Group A
3
versus Group B
3
). Interestingly, while
run-count analyses showed that workers in Group A
3
returned
their attention to hazardous areas (all types of hazards) more fre-
quently, they tended to dwell longer than those in Group B
3
only
on ladder, fall-protection system, and struck-by hazards in the
images.
Table 6depicts differences in eye-movement metrics of workers
who had (Group A
4
) or had not (Group B
4
) seen someone else
injured on a jobsite. Workers in Group B
4
fixated more quickly on
hazardous stimuli than workers in Group A
4
. However, further
investigation into their eye-movement patterns revealed that
although Group A
4
workers returned their attention to hazardous
areas more frequently, they did not dwell much on those areas;
Table 4. Impact of Safety Training on Workers’Attentional Allocation across Different Types of Hazards
Hazards (AOIs) Group
First fixation time (ms) Dwell percentage Run count
NMean SD MD Mean SD MD Mean SD MD
General A
2
19 3,707.500 896.162 3,636.453 0.101 0.009 0.101 2.201 0.459 2.217
B
2
5 3,233.559 621.273 2,997.412 0.106 0.010 0.106 2.149 0.401 2.117
Ladder A
2
19 3,498.433 1,044.360 3,326.000 0.108 0.022 0.096 2.608 0.673 2.667
B
2
5 3,305.928 675.298 3,666.250 0.106 0.017 0.106 2.400 0.666 2.722
Fall to lower level A
2
19 1,385.279 476.604 1,273.861 0.247 0.045 0.250 4.337 1.003 4.326
B
2
5 1,087.521 465.293 939.429 0.285 0.049 0.297 4.791 1.314 4.209
Fall-protection system A
2
19 4,170.635 1,254.556 3,629.250 0.082 0.009 0.082 1.933 0.470 2.064
B
2
5 4,065.010 939.264 3,874.703 0.082 0.010 0.079 1.838 0.351 1.936
Struck by A
2
19 5,606.011 1,829.265 5,328.000 0.068 0.015 0.068 1.767 0.497 1.731
B
2
5 4,686.430 805.299 5,149.000 0.056 0.018 0.064 1.315 0.275 1.269
Housekeeping A
2
19 4,049.956 1,197.663 4,144.000 0.078 0.019 0.079 1.912 0.529 2.000
B
2
5 3,669.257 902.123 4,104.800 0.071 0.022 0.059 1.707 0.393 1.733
Note: N= number of subjects in each group; SD = standard deviation in each group; MD = median in each group.
Table 5. Impact of Being Injured on Workers’Attentional Allocation across Different Types of Hazards
Hazards (AOIs) Group
First fixation time (ms) Dwell percentage Run count
NMean SD MD Mean SD MD Mean SD MD
General A
3
13 3,760.634 797.876 3,876.145 0.101 0.007 0.101 2.469 0.465 2.515
B
3
10 3,514.459 841.847 3,310.964 0.101 0.013 0.104 1.989 0.399 1.862
Ladder A
3
13 3,412.060 1,150.025 3,240 0.107 0.015 0.103 2.944 0.650 3.000
B
3
10 3,560.817 681.234 3,425.756 0.103 0.027 0.091 2.267 0.668 1.944
Fall to lower level A
3
13 1,346.532 446.439 1,289.581 0.245 0.045 0.244 4.911 1.040 4.488
B
3
10 1,271.499 520.490 1,107.387 0.254 0.060 0.279 3.928 0.872 3.977
Fall-protection system A
3
13 4,315.336 1,214.908 4,205.511 0.087 0.011 0.084 2.244 0.579 2.277
B
3
10 3,961.475 985.744 3,553.077 0.080 0.008 0.082 1.747 0.432 1.638
Struck by A
3
13 5,473.008 1,278.817 5,060.909 0.071 0.010 0.069 2.027 0.575 1.769
B
3
10 5,575.305 2,127.791 5,344.615 0.060 0.019 0.061 1.446 0.397 1.462
Housekeeping A
3
13 4,235.941 1,260.845 4,424 0.070 0.018 0.069 2.051 0.527 2.067
B
3
10 3,661.517 973.167 3,760.8 0.087 0.016 0.089 1.760 0.546 1.633
Note: N= number of subjects in each group; SD = standard deviation in each group; MD = median in each group.
© ASCE 04017024-9 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
instead, they distributed their attention across images, excluding
more complex scenes that included struck-by or fall-protection sys-
tem hazards. This result might imply that this group tended to main-
tain situational awareness of the environment by distributing their
attention across scene.
Notwithstanding, the results showed that Group B
5
(who had
neither been injured themselves nor had seen anyone else injured on
a jobsite) fixated more quickly on hazardous areas than Group A
5
(who had both been injured and/or seen someone else injured at a
jobsite), but these groups (A
5
versus B
5
) did not differ significantly
in first fixation time on hazards. Of note, Group A
5
spent less time
examining hazardous areas in images and tended to distribute their
attention across the scene; instead, Group B
5
tended to dwell on spe-
cific hazardous areas with the potential cost of failing to identify
other hazards (Table 7).
Next, to determine whether the effect of injury exposure on
eye movements was greater for certain hazard types, the authors
examined the extent to which the characteristics of each group
predicted first fixation time, dwell percentage, and run count on
each hazard type. The results of the permutation simulations are
shown in Table 8. Workers who had previous injury exposure dis-
tributed their attention significantly differently in all hazardous
situations. Contrary to null hypothesis H
3
, injury exposure has a
significant impact on workers’attentiveness to hazards on a con-
struction site. Workers who had injury exposure tended to return
their attention more frequently to hazardous areas (Welch’st=
−2.846; p=0.015<0.05). Also, workers who personally had
been injured at a jobsite distributed their attention significantly
differently across scenes that included struck-by hazards. The
run-count and dwell percentage analyses verified that these work-
ers more frequently returned their attention to potential or active
struck-by hazards in images (Welch’st=−2.859; p=
0.013 <0.05) and gazed at these hazards for a longer duration
(Welch’st=−1.816; p=0.083<0.1). Comparing Groups A
3
and
B
3
revealed that workers in Group A
3
spent significantly less time
on housekeeping hazards (Welch’st= 2.302; p=0.036<0.05).
Significantly, workers in Group A
4
deployed frequent short fixa-
tions on housekeeping hazardous situations to detect active or
potential hazards without dwelling on one source. The follow-up
investigation will be discussed later.
Work Experience and Worker Attentiveness while
Controlling for Training
Because the results demonstrate that the standard OSHA 10-h train-
ing has a minimal impact (not significant) on search strategy, atten-
tional allocation, and the hazard-identification skills of construction
workers, it can be concluded that standard OSHA 10-h training
alone cannot be considered a comprehensive tool for improving
workers’attention. Moreover, this study’s eye-movement data
Table 7. Impact of Being Injured and/or Seeing Someone Else Injured on Workers’Attentional Allocation across Different Types of Hazards
Hazards (AOIs) Group
First fixation time (ms) Dwell percentage Run count
NMean SD MD Mean SD MD Mean SD MD
General A
5
9 3,910.608 870.891 3,997.135 0.100 0.005 0.101 2.576 0.450 2.593
B
5
6 3,405.179 714.482 3,310.964 0.103 0.008 0.104 1.899 0.453 1.748
Ladder A
5
9 3,631.806 1,304.424 3,240.000 0.105 0.015 0.103 3.062 0.563 3.000
B
5
6 3,387.897 620.099 3,359.556 0.113 0.032 0.100 2.222 0.771 1.917
Fall to lower level A
5
9 1,316.858 487.587 1,243.721 0.244 0.045 0.244 5.207 1.046 5.628
B
5
6 1,103.080 184.127 1,081.177 0.263 0.044 0.280 3.791 1.033 3.523
Fall-protection system A
5
9 4,418.435 1,325.734 4,296.690 0.086 0.012 0.084 2.355 0.631 2.298
B
5
6 4,008.249 1,182.101 3,592.171 0.079 0.010 0.078 1.628 0.424 1.532
Struck by A
5
9 5,731.014 1,363.010 5,289.913 0.070 0.009 0.065 2.060 0.557 1.731
B
5
6 5,687.108 1,713.703 5,392.161 0.058 0.015 0.061 1.346 0.340 1.270
Housekeeping A
5
9 4,473.882 1,319.159 4,829.600 0.066 0.016 0.066 2.067 0.435 2.067
B
5
6 3,647.123 944.816 3,577.714 0.081 0.016 0.084 1.522 0.472 1.433
Note: N= number of subjects in each group; SD = standard deviation in each group; MD = median in each group.
Table 6. Impact of Seeing Someone Else Injured on Workers’Attentional Allocation across Different Types of Hazards
Hazards (AOIs) Group
First fixation time (ms) Dwell percentage Run count
NMean SD MD Mean SD MD Mean SD MD
General A
4
13 3,708.054 999.684 3,756.299 0.099 0.010 0.100 2.437 0.454 2.455
B
4
11 3,381.456 725.299 3,310.964 0.104 0.008 0.105 1.982 0.466 1.860
Ladder A
4
13 3,689.773 1,137.103 3,458.4 0.100 0.015 0.095 2.838 0.647 2.944
B
4
11 3,095.264 670.915 3,305.647 0.114 0.025 0.106 2.354 0.768 1.944
Fall to lower level A
4
13 1,380.634 567.458 1,187.905 0.242 0.056 0.244 4.877 1.048 4.488
B
4
11 1,199.450 308.056 1,101.116 0.259 0.044 0.275 3.922 0.881 3.767
Fall-protection system A
4
13 4,256.243 1,174.578 4,205.511 0.085 0.010 0.084 2.223 0.596 2.277
B
4
11 3,897.974 1,118.337 3,434.4 0.081 0.010 0.079 1.716 0.456 1.681
Struck by A
4
13 5,631.502 1,849.055 5,289.913 0.068 0.015 0.065 1.917 0.561 1.731
B
4
11 5,160.357 1,543.208 5,149 0.067 0.016 0.068 1.552 0.553 1.269
Housekeeping A
4
13 4,230.567 1,281.604 4,593.5 0.075 0.021 0.076 2.082 0.432 2.200
B
4
11 3,658.894 893.291 3,562.8 0.080 0.017 0.084 1.691 0.600 1.533
Note: N= number of subjects in each group; SD = standard deviation in each group; MD = median in each group.
© ASCE 04017024-10 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
results supported the idea that in addition to training, years of expe-
rience seem to be a prominent variable in shaping the safety percep-
tions of workers. To verify this finding, training was considered a
controlling factor, and workers were grouped based on their years
of experience (less than 5 years or more than 10 years). The descrip-
tive statistics relating eye-movement metrics to work experience
while controlling for training are provided in Table 9. Workers with
less experience (Group A
6
) generally had lower first fixation times
on hazardous areas; however, workers with more experience
(Group B
6
) had higher run counts, meaning they returned their
attention to hazardous areas in images more often.
To investigate the differences between these groups, 18 series of
permutation simulations were each run 1,000 times. Among trained
workers, workers who had more experience (Group B
6
) distributed
their attention significantly differently across scenes and had signifi-
cantly higher run counts across all types of hazards: ladder related
(Welch’st= 4.136; p= 0.003 <0.05), fall to lower level (Welch’st=
2.804; p= 0.023 <0.05), fall-protection system (Welch’st= 4.997;
p=0.002<0.05), struck by (Welch’st= 3.392; p= 0.004 <0.05),
and housekeeping (Welch’st=5.109;p= 0.001 <0.05).
Contrary to null hypothesis H
4
, the result of the permutation sim-
ulations confirmed that training has minimal influence on the
attentional allocation and hazard detection of groups and what
differentiates trained workers in safety perception is their years of
experience in the field.
Findings and Discussion
Work Experience
The first null hypothesis, which aimed to test whether years of work
experience had an impact on workers’attentiveness to hazards on a
construction site, was rejected. The findings indicated that more
experienced workers (>10 years) returned their attention to hazard-
ous areas more often than less experienced workers. These hazards
may be comprehended by experienced workers as more dangerous
than others or perhaps extensive experience is required with certain
stimuli before one classifies it as a potential hazard. Previous litera-
ture showed that hazard-detection skill can be impacted by safety
experience, i.e., one of the indicators of tacit knowledge (Zhang
et al. 2014;Hadikusumo and Rowlinson 2004), so the outcomes
confirm existing theory.
The results presented in Table 3revealed that when the con-
struction environment includes potential or active sources of haz-
ards, more experienced workers tended to maintain a balance
between processing and searching the scene by spending less
time exploring hazardous areas and by more frequently tracking
back to those hazardous areas. The results of previous studies
(Choudhry and Fang 2008;Chietal.2005) indicated that both
young and less experienced workers are more prone to accidents,
and when these workers gain more experience, they become more
Table 8. Permutation Simulation Results for Workers Who Had Injury Exposure
Hazard Eye-tracking metrics
Group A
3
versus
Group B
3
Group A
4
versus
Group B
4
Group A
5
versus
Group B
5
Welch’stp-value Welch’stp-value Welch’stp-value
General Run count −2.656 0.019
a
−2.409 0.032
a
−2.846 0.015
a
Ladder Run count −2.441 0.029
a
−1.652 0.103 −2.292 0.047
a
Fall to lower level Run count −2.463 0.028
a
−2.425 0.026
a
−2.588 0.035
a
Fall-protection system Run count −2.357 0.025
a
−2.357 0.021
a
−2.667 0.014
a
Struck by Run count −2.859 0.013
a
−1.600 0.130 −3.080 0.009
a
Dwell percentage −1.816 0.083
b
−0.069 0.936 −1.752 0.096
b
Housekeeping Run count −1.288 0.210 −1.802 0.086
b
−2.258 0.053
b
Dwell percentage 2.302 0.036
a
0.702 0.482 1.769 0.113
a
p≤0.05.
b
p≤0.1.
Table 9. Impacts of Work Experience on Workers’Attentional Allocation across Different Types of Hazards, with Training as a Controlling Factor
Hazards (AOIs) Group
First fixation time (ms) Dwell percentage Run count
NMean SD MD Mean SD MD Mean SD MD
General A
6
9 3,931.732 992.628 3,997.135 0.097 0.010 0.098 2.379
a
0.359 2.320
B
6
6 3,044.915 603.285 3,161.181 0.105 0.009 0.106 1.701
a
0.205 1.748
Ladder A
6
9 3,933.498 1,219.092 3,442.889 0.100 0.016 0.095 2.852
a
0.479 2.944
B
6
6 3,132.274 866.041 3,057.133 0.108 0.028 0.094 1.917
a
0.392 1.833
Fall to lower level A
6
9 1,541.430 570.365 1,403.628 0.228 0.044 0.232 4.579
a
0.989 4.395
B
6
6 1,126.322 ,199.763 1,124.571 0.268 0.050 0.276 3.477
a
0.523 3.477
Fall-protection system A
6
9 4,250.217 1144.259 3,958.000 0.086 0.011 0.084 2.175
a
0.373 2.277
B
6
6 3,318.001 425.332 3,422.200 0.079 0.005 0.080 1.411
a
0.218 1.457
Struck by A
6
9 5,442.838 2,043.338 4,920.750 0.072 0.014 0.069 2.009
a
0.459 1.846
B
6
6 4,707.313 1,168.006 5,141.417 0.065 0.019 0.070 1.308
a
0.341 1.231
Housekeeping A
6
9 3,991.903 1,339.133 3,777.000 0.087 0.019 0.088 2.244
a
0.408 2.200
B
6
6 3,407.124 830.206 3,205.900 0.075 0.013 0.079 1.356
a
0.266 1.367
Note: N= number of subjects in each group; SD = standard deviation in each group; MD = median in each group.
a
p≤0.05.
b
p≤0.1.
© ASCE 04017024-11 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
aware of safety requirements. Moreover, a study conducted by
Sawacha et al. (1999) showed that there is a strong relationship
between the experience of operatives and their level of safety per-
formance; however, the Sawacha study was based on a question-
naire; thus, the present study takes this investigation a step further
by supporting this claim using empirical data.
First time of fixation was not significant for any of the AOIs,
which means that more experienced workers were not faster to fix-
ate on potential hazards; however, these workers were more likely
to return to the AOIs multiple times. Generally, more experienced
workers spent less processing time than less experienced workers,
which is an outcome that agrees with the study by Dzeng et al.
(2016). Similarly, the duration of time that more and less experi-
enced workers spent dwelling on fall-protection system hazards
was significantly (moderately) different (p-value = 0.099). The
results presented in Table 3revealed that when the construction
environment includes potential or active sources of fall-protection
system hazards, more experienced workers tended to maintain a
balance between processing and searching the scene by spending
more time exploring fall-protection system hazardous areas and fre-
quently tracked back tothose areas.
In a previous study that used eye tracking to compare the hazard
identification of workers, Dzeng et al. (2016) noted that experienced
workers did not perform significantly better than novices in identi-
fying hazards. Although their findings are in contrast with the
results of the current study, it is critical to note the points of depar-
ture between the two studies. First, the authors of the current study
have defined experienced workers as those with more than 10 years
of experience and novices as those with less than 5 years of experi-
ence. Dzeng et al. (2016) qualified experienced workers as those
with a minimum of 5 years of experience and, more critically, their
novice sample consisted of students, not construction workers, who
had no work experience and never received safety training of any
substantive nature. In other words, Dzeng et al. (2016) compared
workers with experience with students without any experience,
which significantly impacted the results of their study.
Comparatively, the central objective in the current study is to study
whether gaining more experience on a construction site improves
workers’hazard-identification skills (as manifested by attention and
eye movements); thus, all participants in this study were required to
have experience working on a construction site. All of the partici-
pants in the current study are professional construction workers,
and the authors took rigorous steps to divide them into two distinct
groups.
Second, whereas the Dzeng et al. (2016) stimuli consisted of a
limited number (only four) of virtual images of hypothetical con-
struction scenarios, the current study’s 35 images were taken from
real construction sites and represented a wide range of common
hazards. In addition, there were at least 20 AOIs for each type of
hazard across the images, which helped to obtain more reliable out-
comes by measuring the average performance of workers in identi-
fying different types of hazards. As such, although the study Dzeng
et al. (2016) conducted is an important step in applying eye-tracking
technology to a construction setting, the present study considerably
expands the scope of what can be studied via eye tracking in con-
struction safety.
The findings of this study can further be compared with the
results of other previous eye-tracking studies. Underwood (2007)
suggested that the efficiency of visual search strategies relates to the
changes in driving skills that mark the transition from novice to
experienced drivers. Many other studies have provided support for
the notion that driving experience is a key predictor of crash rates,
with young novice drivers being at a greater risk (Chapman and
Underwood 1998;Konstantopoulos et al. 2010). Similarly, learning
by doing and gaining experience can be one of the best training
options for construction workers (Choudhry and Fang 2008). This
study shows that as construction workers gain experience, their
hazard-detection skills improve, enabling them to search and
examine scenes more intuitively. Thus, it can be concluded that
gaining experience in construction is strongly related to effective
and efficient visual search and attentional distribution.
Safety Training
The second null hypothesis of the study stated that past safety train-
ing (formal or informal) would have minimal (nonsignificant)
impact on workers’attentiveness to hazards on a construction site.
The hypothesis could not be rejected for formal training. The results
showed no significant differences in the search strategies or atten-
tional patterns of workers with or without the OSHA 10-h certificate
training when they were exposed to hazardous situations. The find-
ings of this eye-tracking experiment can be alarming for construc-
tion safety professionals, because it implies that the most common
and basic training (i.e., OSHA 10-h certificate) may not consider-
ably improve hazard-detection skills. Although the results do not
state that the OSHA 10-h certificate is ineffective, they do suggest
that developing more innovative and interactive training techniques
can improve workers’hazard-detection skills [e.g., using OSHA
visual inspection training, which is interactive and game-based
training; using a three-sided virtual reality, such as cave automatic
virtual environment Perlman et al. (2014)].
Regarding informal training, however, the second null hypothe-
sis was rejected for run counton struck-by hazards. The results indi-
cated that the attentional pattern of workers who have had informal
training (i.e., through on-site training or participation in safety
meetings) was significantly different for struck-by hazardous situa-
tions (Welch’st= 2.693; p-value = 0.032). To study this finding fur-
ther, attentional distributions (heat maps) of two workers, one who
had informal training and the other who had no safety training, were
created. As the heat map comparison shows (Fig. 3), the worker
with informal training appears more attentive to struck-by hazards
and fixates on areas that show potential for struck-by accidents.
These workers distributed their attention across the scene and
scanned areas close to a potential hazard, checking the proximity of
equipment to workers and other equipment in the scene.
Past Injury Exposure
The third null hypothesis, which tested whether workers’past injury
exposure affected their attentional patterns, was also rejected. The
results of the analysis demonstrated that injury exposure signifi-
cantly impacts the cognitive processes of those who had previously
been exposed to specific hazards (Table 8). The test subjects in this
group behaved more conservatively and tended to check the sur-
rounding environment frequently with short fixations, seeking
potential and active sources of hazards. This finding conforms with
some previous studies (Törner and Pousette 2009;Shin et al. 2014),
which found that past injury exposure increases workers’risk
awareness. It was interesting that workers who had been injured
before had no significant differences in attentional allocation to
housekeeping hazardscompared with those who had no injury expo-
sure. The reason behind this finding may correlate with the type(s)
of hazards they had been exposed to before. According to workers’
self-reports, none of the workers had experienced injury with house-
keeping or a cluttered environment as sources of hazard.
Workers who had past injury exposure became more conserva-
tive and careful regarding struck-by hazards in a dynamic
© ASCE 04017024-12 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
construction environment. Jacob and Karn (2003) showed that a
higher dwell percentage on particular elements is a sign of that ele-
ment’s importance for that group, leading to the conclusion that
workers with past injury exposure would dwell longer and return
their attention more often to comprehend the complexity of a
dynamic hazardous situation (struck-by hazard) than workers with
no injury exposure. The findings also indicated that workers who
had been injured or had witnessed another’s injury tended to not
dwell on specific hazards and distributed their attention over the
scene while frequently returning their attention to better compre-
hend all dimensions of a hazardous area. On the other hand, the
mean dwell time percentages for workers without any injury expo-
sure experience were longer, and they tended to focus their attention
on specific hazards, possibly misidentifying other active hazards or
failing to anticipate potential hazards on the construction site.
Furthermore, grouping workers based on their past experience
of seeing another worker injured provided additional insight into
the cognitive processes of construction workers. Having seen
another worker injured impacted the attentional allocation of the
current participants. These workers appeared to become more con-
cerned about their safety and health, and when they faced hazardous
situations, they tended to maintain situational awareness. Although
the workers who had witnessed someone else injured frequently
checked AOIs related to housekeeping (p-value = 0.086) and ladder
hazards (p-value = 0.103) (moderately significant), their run count
to fall hazards was more significant (p-value = 0.026). Self-reports
of workers revealed that approximately 70% of them had experi-
enced witnessing another worker injured due to fall incidents, such
as falling from scaffold, falling from an unprotected edge, or falling
from height due to broken cable. Scan paths for a representative
worker of each group verified the statistical findings that workers
who saw another worker falling from height exhibited differential
visual behaviors and followed a different search strategy (Fig. 4).
Moreover, the run count associated with struck-by–related hazards
was not significantly different between these groups. Returning to
the workers’self-reports of their injury experience also confirmed
that none of the workers acknowledged seeing another worker being
injured due to struck-by accidents.
Years of Experience while Controlling for
Training Impacts
The fourth null hypothesis, which aimed to test whether years of
work experience of trained workers had an impact on their atten-
tiveness to hazards on a construction site, was rejected. The results
suggested that as trained workers gain more experience in the con-
struction field, they will be more aware of safety requirements on
jobsites and of the full extent of all potentially hazardous items/sit-
uations in the environment. This outcome is consistent with previ-
ous studies in which workers become better at detecting hazards
and behaving safely by gaining experience and being exposed to
hazardous situations (tacit knowledge) rather than by receiving for-
mal training with traditional educational methods (explicit knowl-
edge) (Podgórski 2010).
For example, as seen in Figs. 5(a–c), the less experienced worker
is more stimulus driven and concentrated on imminent hazards,
Fig. 3. Attentional distributions (heat maps) (images courtesy of David Ausmus, with permission): (a) original picture; (b) worker who had informal
safety training; (c) worker who had no safetytraining
Fig. 4. Difference in cognitive process (search strategy) (images courtesy of David Ausmus, with permission): (a) original picture; (b) worker who
had not seen someone injured due to fall hazards; (c) worker who hadseen someone injured due to fall hazards
© ASCE 04017024-13 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
such as the danger to the man standing on the scaffold, but the more
experienced workers showed a balance in focusing and dividing
their attention across the scene and are more goal driven. To better
understand the difference between the search strategies and cogni-
tive processes of these groups, scan paths were obtained from one
worker in each group. Scan paths suggest that experienced workers
optimized their attention by distributing it across the scene and scru-
tinized more areas to maintain their awareness about potential or
active sources of hazards in scene. The study of Aboagye-Nimo et
al. (2012) points out that “explicit knowledge […] will not be suffi-
cient in practice in order to prevent accidents”(p. 419). They high-
lighted the important role of tacit knowledge, namely work experi-
ence, in learning safe work practices. In sum, these findings align
with the study of McBride and Cutting (2015), who verified that as
a result of years of experience and training, the controlled atten-
tional allocation for hazard detection in less experienced workers
turned into an automatic attentional allocation in more experienced
workers.
Conclusions
One of the root causes of accidents is human error (Abdelhamid and
Everett 2000), i.e., the lack of attention and the failure of workers to
identify hazards. If workers are distracted or lack attentiveness,
then they cannot identify and properly respond to a hazard. Because
there is a direct link between visual cues and attentional allocation,
one of the scientific methods of studying attention is tracking the
eye movements of people. Although eye tracking has been exten-
sively used to study attention in other domains of science
(Duchowski 2007), its applications remain relatively unexplored in
the field of construction safety.
To address this knowledge gap, the research team used eye-track-
ing technology to measure the impact of safety knowledge (in terms
of training, work experience, and injury exposure) on construction
workers’attentional allocation. It was found that although work ex-
perience and injury exposure significantly impact visual search strat-
egies and attentional allocation, the difference between workers with
and without the OSHA 10-h certificate is not significant. One should
note that the results do not state that the OSHA 10-h certificate is
ineffective, but it calls for developing more innovative training tech-
niques, such as interactive and game-based training, which can
improve workers’hazard-detection skills. The results indicate that
by integrating both tacit knowledge (work experience and injury ex-
posure) and explicit knowledge (e.g., interactive training), future
endeavors to enhance worker safety knowledge will achieve supe-
rior outcomes in terms of worker safety awareness.
This study’s results increase the field’s understanding of the
variables that impact attentional allocation and provide a novel
approach for improving construction site safety by using eye-
tracking technologies that have been widely accepted as the most
direct and continuous measure of attention. The results of the study
reported in this paper make a significant contribution to the existing
literature. First, the eye-tracking metrics that characterize the varia-
tion in construction workers’attention while they search for hazards
were identified. Second, given the established link between eye
Fig. 5. Differences in cognitive process and attentional allocation (images courtesy of David Ausmus, with permission): (a and d) original picture;
(b) attentional distribution for the group of workers who had received safety training and fewer than 5 years of experience; (c) attentional distribution
for the group of workers who had received safety training and more than 10 years of experience; (e) visual search strategy for a worker who had
received safety training and fewer than 5 years of experience; (f) attentional distribution and visual search strategy for the group of workers who had
received safety training and more than 10 years of experience
© ASCE 04017024-14 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
movements and cognitive processes, the findings of this study lay
the foundation for using eye-tracking technologies to further study
the role of cognition inconstruction safety. The previously described
approach can be used by other researchers to advance knowledge
regarding the way that attentional allocation of construction workers
impacts the likelihood of accidents and can subsequently lead to the
development of new accident-causation theories.
The results also make a significant contribution to practice.
Because of these results, one can conclude that traditional teaching
methods (e.g., lectures, media) for construction worker education
and training do not improve workers’visual search skills suffi-
ciently to identify hazards. More effective interactive methods
should be applied instead of conventional teacher-student training.
For instance, safety knowledge could be more effectively acquired
using interactive and experiential learning methods. The need to
find more effective training methods becomes more urgent in the
face of the fact that 185,000 new workers are predicted to join con-
struction in this decade (NCCER 2013). Eye-tracking technology
can provide a viable solution for this challenge by providing real-
time tracking of the eye movements of construction workers to mea-
sure the effectiveness of their training (via pretest and posttest mon-
itoring). Eye tracking can further be used for innumerable purposes
inclined toward achieving safer working conditions, such as identi-
fying hidden hazards, measuring the situation awareness of work-
ers, and improving the effectiveness of safety-training programs.
This study has some limitations and areas that future work can
address. First, the research team examined the gaze exhibited toward
static images of construction sites rather than dynamic real-world
construction sites. One might argue that since the images are merely
pictures ona screen,they do not translateto construction sites; work-
ers may act differently in interactions with real-world construction
sites. To completely address this issue, future research may examine
the workers’gazes and assess their hazard-detection skills on real-
world construction sites using a mobile eye tracker. Another limita-
tion of this study relates to sampling and the number of participants.
Because of practical limitations, the research team only recruited
workers located near Lincoln and Omaha. In addition, although the
sample of the present study was relatively large for an eye-tracking
study, the practical limitations of recruiting construction workers to
participate in the experiment in a laboratory setting restricted the
number of participants. Future studies might address this limitation
by replicating the study in other geographical regions and also
increasing the number of participants.
Acknowledgments
The research team thanks the University of Nebraska-Lincoln for
supporting the research reported in this paper through a Research
Council Faculty Seed Grant. The research team also thanks
workers who participated in this study. Appreciation is extended
to Mr. David W. Ausmus (image photographer), Mr. Ronald D.
Allen, and other professional safety managers for their
considerable and vital support in this study. Any opinions,
findings, and conclusions or recommendations expressed in this
material are those of the writers and do not necessarily reflect the
views of the University of Nebraska-Lincoln.
References
Abdelhamid, T. S., and Everett, J. G. (2000). “Identifying root causes
of construction accidents.”J. Constr. Eng. Manage.,10.1061
/(ASCE)0733-9364(2000)126:1(52),52–60.
Aboagye-Nimo, E., Raiden, A., Tietze, S., and King, A. (2012). “The use of
experience and situated knowledge in ensuring safety among workers of
small construction firms.”Proc., 28th Annual ARCOM Conf.,S.D.
Smith, ed., Association of Researchers in Construction Management,
Reading, U.K., 413–422.
Adams, D. C., and Anthony, C. D. (1996). “Using randomization techniques
to analyse behavioural data.”Animal Behav, 51(4), 733–738.
Anderson, J. R. (2005). Cognitive psychology and its implications,
Macmillan, New York.
Anderson, M. J. (2001). “Permutation tests for univariate or multivariate
analysis of variance and regression.”Can. J. Fish. Aquat. Sci., 58(3),
626–639.
Argote, L., and Miron-Spektor, E. (2011). “Organizational learning: From
experience to knowledge.”Org. Sci., 22(5), 1123–1137.
Ballard, J. C. (1996). “Computerized assessment of sustained attention: A
review of factors affecting vigilance performance.”J. Clin. Exp.
Neuropsychol., 18(6), 843–863.
Berger, V. W. (2000). “Pros and cons of permutation tests in clinical trials.”
Stat. Med., 19(10), 1319–1328.
Bhoir, S. A., Hasanzadeh, S., Esmaeili, B., Dodd, M. D., and Fardhosseini,
M. S. (2015). “Measuring construction workers’attention using eye-
tracking technology.”Proc., ICSC15: The Canadian Society for Civil
Engineering 5th Int./11th Construction Specialty Conf., Canadian
Society for Civil Engineering, Montreal, QC, Canada.
Bisley, J. W. (2011). “The neural basis of visual attention.”J. Physiol.,
589(1), 49–57.
Bojko, A. A. (2009). “Informative or misleading? Heat maps decon-
structed.”Human-computer interaction. New trends, J. A. Jacko, ed.,
Springer, Berlin, 30–39.
Borowsky, A., Oron-Gilad, T., Meir, A., and Parmet, Y. (2012). “Drivers’
perception of vulnerable road users: A hazard perception approach.”
Accid. Anal, Prev., 44(1), 160–166.
Chapman, P. R., and Underwood, G. (1998). “Visual search of driving situa-
tions: Danger and experience.”Perception, 27(8), 951–964.
Chi, C. F., Chang, T. C., and Ting, H. I. (2005). “Accident patterns and pre-
vention measures for fatal occupational falls in the construction indus-
try.”Appl. Ergon., 36(4), 391–400.
Chomsky, N. (1959). “Review: Verbal behavior by B. F. Skinner.”
Language, 35(1), 26–58.
Choudhry, R. M., and Fang, D. (2008). “Why operatives engage in unsafe
work behavior: Investigating factors on construction sites.”Saf. Sci.,
46(4), 566–584.
Chua, D., and Goh, Y. (2004). “Incident causation model for improving
feedback of safety knowledge.”J. Constr. Eng. Manage.,10.1061
/(ASCE)0733-9364(2004)130:4(542), 542–551.
Chun, M. M., and Wolfe, J. M. (2001). “Visual attention.”Blackwell hand-
book of sensation and perception, E. B. Goldstein, ed., Blackwell,
Malden, MA.
De Brigard, F. (2012). “The role of attention in conscious recollection.”
Front. Psychol.,3,1–10.
Drummond, G. B., and Vowler, S. L. (2012). “Different tests for a differ-
ence: How do we do research?”Adv. Physiol. Educ., 36(1), 3–5.
Duchowski, A. (2007). Eye tracking methodology: Theory and practice,
Springer Science and Business Media, Dordrecht, Netherlands.
Dzeng, R. J., Lin, C. T., and Fang, Y. C. (2016). “Using eye-tracker to com-
pare search patterns between experienced and novice workers for site
hazard identification.”Saf. Sci., 82(Feb), 56–67.
Edgington, E. S. (1995). Randomization tests, 3rd Ed., Marcel-Dekker,
New York.
Ernst, M. D. (2004). “Permutation methods: A basis for exact inference.”
Stat. Sci., 19(4), 676–685.
Eudey, T., Kerr, J., and Trumbo, B. (2010). “Using R to simulate permuta-
tion distributions for some elementary experimental designs.”J. Stat.
Educ., 18(1), 1–30.
Field, A. (2013). Discovering statistics using IBM SPSS statistics, 4th Ed.,
Sage, London.
Fisher, D., Pradhan, A., Pollatsek, A., and Knodler Jr., M. (2007).
“Empirical evaluation of hazard anticipation behaviors in the field and
on driving simulator using eye tracker.”Transportation Research
Record, 2018, 80–86.
© ASCE 04017024-15 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
Foerster, R. M., and Schneider, W. X. (2013). “Functionally sequenced
scanpath similarity method (FuncSim): Comparing and evaluating scan-
path similarity based on a task’s inherent sequence of functional (action)
units.”J. Eye Mov. Res., 6(5), 1–22.
Fuchs, A. F. (1971). “The saccadic system.”The control of eye movements,
Academic, New York, 343–362.
Garrett, J., and Teizer, J. (2009). “Human factors analysis classification
system relating to human error awareness taxonomy in construction
safety.”J. Constr. Eng. Manage.,10.1061/(ASCE)CO.1943-7862
.0000034, 754–763.
Gleason, J. (2013). “Comparative power of the ANOVA, randomization
ANOVA, and Kruskal-Wallis test.”Ph.D. dissertation, Wayne State
Univ., Detroit.
Goldberg, J. H., Stimson, M. J., Lewenstein, M., Scott, N., and Wichansky,
A. M. (2002). “Eye tracking in web search tasks: Design implications.”
Proc., 2002 Symp. on Eye Tracking Research and Applications,
Association forComputing Machinery, New York, 51–58.
Ha, J. S., and Seong, P. H. (2009). “A human–machine interface evaluation
method: A difficulty evaluation method in information searching
(DEMIS).”Reliab. Eng. Syst. Saf., 94(10), 1557–1567.
Hadikusumo, B., and Rowlinson, S. (2004). “Capturing safety knowl-
edge using design-for-safety-process tool.”J. Constr. Eng. Manage.,
10.1061/(ASCE)0733-9364(2004)130:2(281), 281–289.
Hallett, P. (1986). “Eye movements (and human visual perception).”
Handbook of perception and human performance,Vol.1,Wiley,
New York.
Hallowell, M. R. (2012). “Safety-knowledge management in American
construction organizations.”J. Manage. Eng.,10.1061/(ASCE)ME
.1943-5479.0000067, 28(2), 203–211.
Hasanzadeh, S., Esmaeili, B., and Dodd, M. (2016). “Measuring construc-
tion workers’real-time situation awareness using mobile eye-tracking.”
Construction Research Congress 2016, ASCE, Reston, VA, 2894–
2904.
Hauland, G. (2003). “Measuring team situation awareness by means of eye
movement data.”Proc., HCI Int., J. Jacko and C. Stephanidis, eds.,
Lawrence Erlbaum Associates, Mahwah, NJ, 230–234.
Hoffman, J. E., and Subramaniam, B. (1995). “The role of visual attention
in saccadic eye movements.”Percept. Psychophys., 57(6), 787–795.
Ikuma, L. H., Harvey, C., Taylor, C. F.,and Handal, C. (2014). “A guide for
assessing control room operator performance using speed and accuracy,
perceived workload, situation awareness, and eye tracking.”J. Loss
Prev. Process Ind., 32(Nov), 454–465.
Jacob, R. J., and Karn, K. S. (2003). “Eye tracking in human-computer
interaction and usability research: Ready to deliver the promises.”Mind,
2(3), 4.
James, W. (1913). The principles of psychology, Vol. II, Dover, New York.
Java Graphical User Interface (GUI) for R 1.7-9 [Computer software].
Deducer package, RoSuDa, Univ. of Augsburg, Augsburg, Germany.
Jie, L., and Clark, J. J. (2008). “Video game design using an eye-movement-
dependent model of visual attention.”ACM Trans. Multimedia Comput.
Commun. Appl., 4(3), 22.
Just, M. A., and Carpenter, P. A. (1980). “A theory of reading: From eye fix-
ations to comprehension.”Psychol. Rev., 87(4), 329.
Konstantopoulos, P., Chapman, P., and Crundall, D. (2010). “Driver’s vis-
ual attention as a function of driving experience and visibility. Using a
driving simulator to explore drivers’eye movements in day, night and
rain driving.”Accid. Anal. Prev., 42(3), 827–834.
Koskinen, K. U., Pihlanto, P., and Vanharanta, H. (2003). “Tacit knowledge
acquisition and sharing in a project work context.”Int. J. Project
Manage., 21(4), 281–290.
Laeng, B., and Teodorescu, D. S. (2002). “Eye scanpaths during visual im-
agery reenact those of perception of the same visual scene.”Cognit. Sci.,
26(2), 207–231.
LaFleur, B. J., and Greevy, R. A. (2009). “Introduction to permutation and
resampling-based hypothesis tests.”J. Clin. Child Adolesc. Psychol.,
38(2), 286–294.
Lavine, R. A., Sibert, J. L., Gokturk, M., and Dickens, B. (2002). “Eye-
tracking measures and human performance in a vigilance task.”Aviat.
Space Environ. Med., 73(4), 367–372.
Lopez, R., Love, P. E., Edwards, D. J., and Davis, P. R. (2010). “Design
error classification, causation, and prevention in construction engi-
neering.”J. Perform. Constr. Facil.,10.1061/(ASCE)CF.1943-5509
.0000116, 399–408.
Ludbrook, J. (1994). “Advantages of permutation (randomization) tests in
clinical and experimental pharmacology and physiology.”Clin. Exp.
Pharmacol. Physiol., 21(9), 673–686.
Ludbrook, J., and Dudley, H. (1998). “Why permutation tests are superior
to t and F tests in biomedical research.”Am. Stat., 52(2), 127–132.
Manly, B. F. J. (1997). Randomization and Monte Carlo methods in biol-
ogy, Chapman &Hall, New York.
Mackenzie, A. K., and Harris, J. M. (2015). “Eye movements and hazard
perception in active andpassive driving.”Vis. Cognit., 23(6), 736–757.
McBride, D. M., and Cutting, J. C. (2015). Cognitive psychology: Theory,
process, and methodology, Sage, Thousand Oaks, CA.
Muttart, J., Fisher, D., Knodler, M., and Pollatsek, A. (2007). “Driving with-
out a clue: Evaluation of driver simulator performance during hands-
free cell phone operation in a work zone.”Transportation Research
Record, 2018, 9–14.
NCCER (National Center for Construction Education and Research).
(2013). “Craft workforce development 2013 and beyond.”hhttps://goo
.gl/qf09NUi(Apr. 25, 2016).
Palinko, O., Kun, A. L., Shyrokov, A., and Heeman, P. (2010). “Estimating
cognitive load using remote eye tracking in a driving simulator.”Proc.,
2010 Symp. on Eye-Tracking Research and Applications, Association
for Computing Machinery, New York, 141–144.
Pan, B., Hembrooke, H. A., Gay, G. K., Granka, L. A., Feusner, M. K., and
Newman, J. K. (2004). “The determinants of web page viewing behav-
ior: An eye-tracking study.”Proc., 2004 Symp. on Eye Tracking
Research and Applications, Association for Computing Machinery,
New York, 147–154.
Parasuraman, R., Warm, J. S., and See, J. E. (1998). “Brain systems of vigi-
lance.”The attentive brain, MIT Press, Cambridge, MA, 221–256.
Perlman, A., Sacks, R., and Barak, R. (2014). “Hazard recognition and risk
perception in construction.”Saf. Sci., 64(Apr), 22–31.
Pernice, K., and Nielsen, J. (2009). “How to conduct eye-tracking studies.”
Nielsen Norman Group, Fremont, CA.
Podgórski, D. (2010). “The use of tacit knowledge in occupational safety
and health management systems.”Int. J. Occup. Saf. Ergon., 16(3),
283–310.
Poole, A., and Ball, L. J. (2006). “Eye tracking in HCI and usability
research.”Encycl. Hum. Comput. Interact., 1, 211–219.
Pradhan, A., et al. (2009). “Attention maintenance in novice drivers:
Assessment and training.”Proc., Int. Driving Symp. on Human Factors
in Driver Assessment, Training, and Vehicle Design, Univ. of Iowa
Public Policy Center, Iowa City, IA.
R Development Core Team. (2012). “R: A language and environment for
statistical computing, version R2.15.0.”Report prepared for the R
Foundation for Statistical Computing, Vienna, Austria.
Raschke, M., Blascheck, T., Richter, M., Agapkin, T., and Ertl, T. (2014).
“Visual analysis of perceptual and cognitive processes.”2014 Int. Conf.
on Information Visualization Theory and Applications (IVAPP), IEEE,
New York, 284–291.
Rehder, B., and Hoffman, A. B. (2005). “Eye tracking and selective atten-
tion in category learning.”Cognit. Psychol., 51(1), 1–41.
Richardson, D. C., and Spivey, M. J. (2004). “Eye tracking: Characteristics
and methods.”Encycl. Biomater. Biomed. Eng., 3, 1028–1042.
Romoser, M. R., Fisher, D. L., Mourant, R., Wachtel, J., and Sizov, K.
(2005). “The use of a driving simulator to assess senior driver perform-
ance: Increasing situational awareness through post-drive one-on-one
advisement.”Proc., Third Int. Driving Symp. on Human Factors in
Driver Assessment, Training and Vehicle Design, Univ. of Iowa Public
Policy Center, Iowa City, IA, 456–463.
Rozenfeld, O., Sacks, R., Rosenfeld, Y., and Baum, H. (2010). “Construction
jobsafetyanalysis.”Saf. Sci., 48(4), 491–498.
Salvucci, D. D., and Goldberg, J. H. (2000). “Identifying fixations and sac-
cades in eye-tracking protocols.”Proc., 2000 Symp. on Eye Tracking
Research and Applications, Association of Computing Machinery, New
York, 71–78.
© ASCE 04017024-16 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.
Sanders, M. S., and McCormick, E. J. (1998). “Human factors in engineer-
ing and design.”Ind. Robot, 25(2), 153–153.
Sawacha, E., Naoum, S., and Fong, D. (1999). “Factors affecting safety per-
formance on construction sites.”Int. J. Project Manage., 17(5),
309–315.
Shepherd, M., Findlay, J. M., and Hockey, R. J. (1986). “The relationship
between eye movements and spatial attention.”Q. J. Exp. Psychol. Sect.
A, 38(3), 475–491.
Shin, M., Lee, H. S., Park, M., Moon, M., and Han, S. (2014). “A system dy-
namics approach for modeling construction workers’safety attitudes
and behaviors.”Accid. Anal. Prev., 68(Jul), 95–105.
Sohlberg, M. M., and Mateer, C. A. (1989). Introduction to cognitive reha-
bilitation: Theory and practice, Guilford Press, New York.
Suh, W., Park, P., Park, C., and Chon, K. (2006). “Relationship between
speed, lateral placement, and drivers’eye movement at two-lane rural
highways.”J. Transp. Eng.,10.1061/(ASCE)0733-947X(2006)132:
8(649), 649–653.
Sun, Q., Xia, J., Nadarajah, N., Falkmer, T., Foster, J., and Lee, H. (2016).
“Assessing drivers’visual-motor coordination using eye tracking,
GNSS and GIS: A spatial turn in driving psychology.”J. Spatial Sci.,
61(2), 1–18.
Sun, Y., Fisher, R., Wang, F., and Gomes, H. M. (2008). “A computer
vision model for visual-object-based attention and eye movements.”
Comput. Vision Image Understanding, 112(2), 126–142.
Theeuwes, J., Belopolsky, A., and Olivers, C. N. (2009). “Interactions
between working memory, attention and eye movements.”Acta Psychol.,
132(2), 106–114.
Todd,S.,andKramer,A.F.(1993).“Attentional guidance in visual
attention.”Proc., Human Factors and Ergonomics Society 37th
Annual Meeting, Human Factors &Ergonomics Society, Santa
Monica, CA.
Törner, M., and Pousette, A. (2009). “Safety in construction—A compre-
hensive descriptionof the characteristics of high safety standards in con-
struction work, from the combined perspective of supervisors and
experienced workers.”J. Saf. Res., 40(6), 399–409.
Torralba, A., Oliva, A., Castelhano, M. S., and Henderson, J. M. (2006).
“Contextual guidance of eye movements and attention in real-world
scenes: The role of global features in object search.”Psychol. Rev.,
113(4), 766.
Underwood, G. (2007). “Visual attention and the transition from novice to
advanced driver.”Ergonomics, 50(8), 1235–1249.
Underwood, G., Crundall, D., and Chapman, P. (2011). “Driving simulator
validation with hazard perception.”Transp. Res. F: Traffic Psychol.
Behav., 14(6), 435–446.
Underwood, G., Phelps, N., Wright, C., Van Loon, E., and Galpin, A.
(2005). “Eye fixation scanpaths of younger and older drivers in a hazard
perception task.”Ophthal. Physiol. Opt., 25(4), 346–356.
Vecera, S. P., Cosman, J. D., Vatterott, D. B., and Roper, Z. J. (2014). “The
control of visual attention: Toward a unified account.”Psychol. Learn.
Motivation, 60, 303–347.
Wedel, M., and Pieters, R. (2008). “A review of eye-tracking research in
marketing.”Review of marketing research, N. K. Malhotra (ed.), Vol. 4,
Emerald Group, Bingley, U.K., 123–147.
Weibel, N., Fouse, A., Emmenegger, C., Kimmich, S., and Hutchins, E.
(2012). “Let's look at the cockpit: Exploring mobile eye-tracking for
observational research on the flight deck.”Proc., Symp. on Eye Tracking
Research and Applications, Association for Computing Machinery, New
York, 107–114.
Yantis, S. (1998). “Objects, attention, and perceptual experience.”Visual
attention, R. D. Wright (ed.), Oxford University Press, New York, 187–
214.
Yarbus, A. L. (1967). Eye movements and vision, Plenum, New York.
Zhang, L., and He, J. (2016). “Critical factors affecting tacit-knowledge
sharing within the integrated project team.”J. Manage. Eng.,10.1061
/(ASCE)ME.1943-5479.0000402, 04015045.
Zhang, S., Boukamp, F., and Teizer, J. (2014). “Ontology-based semantic
modeling of safety management knowledge.”Proc., 2014 Int. Conf. on
Computing in Civil and Building Engineering, R. Issa and I. Flood, eds.,
ASCE, Reston, VA, 2254–2262.
Zhao, F., Schnotz, W., Wagner, I., and Gaschler, R. (2014). “Eye tracking
indicators of reading approaches in text-picture comprehension.”
Frontline Learn. Res., 2(4), 46–66.
Zheng, B., et al. (2011). “Surgeon’s vigilance in the operating room.”Am. J.
Surg., 201(5), 673–677.
© ASCE 04017024-17 J. Manage. Eng.
J. Manage. Eng., 2017, 33(5): 04017024
Downloaded from ascelibrary.org by VIRGINIA TECH UNIVERSTIY on 03/06/18. Copyright ASCE. For personal use only; all rights reserved.