Construction safety training using immersive virtual reality

Abstract

Construction workers’ ability to identify and assess risks is acquired through training and experience and is among the key factors that determine their behaviour and thus their safety. Yet researchers have questioned the effectiveness of conventional safety training. This research tested the hypotheses that safety training in a virtual reality (VR) construction site would be feasible and more effective, in terms of workers’ learning and recall in identifying and assessing construction safety risks, than would equivalent training using conventional methods. Sixty-six subjects were provided training in construction safety and their safety knowledge was tested prior to the training, immediately afterward, and one month later. Half of the subjects received traditional classroom training with visual aids; the other half were trained using a 3D immersive VR power-wall. Significant advantage was found for VR training for stone cladding work and for cast-in-situ concrete work, but not for general site safety. VR training was more effective in terms of maintaining trainees’ attention and concentration. Training with VR was more effective over time, especially in the context of cast-in-situ concrete works. Given the need for improved training and the advantages of training using VR, incorporation of VR in construction safety training is strongly recommended.

Full text

Construction safety training using immersive virtual
reality
RAFAEL SACKS
1
*, AMOTZ PERLMAN
2
and RONEN BARAK
1
1
Faculty of Civil and Environmental Engineering, Technion – Israel Institute of Technology, Rabin Building,
Technion Campus, Haifa 32000, Israel
2
Department of Management, Bar Ilan University, Management, Ramat Gan, Israel
Received 30 August 2012; accepted 22 July 2013
Construction workers’ ability to identify and assess risks is acquired through training and experience and is
among the key factors that determine their behaviour and thus their safety. Yet researchers have questioned
the effectiveness of conventional safety training. This research tested the hypotheses that safety training in a
virtual reality (VR) construction site would be feasible and more effective, in terms of workers’ learning and
recall in identifying and assessing construction safety risks, than would equivalent training using conventional
methods. Sixty-six subjects were provided training in construction safety and their safety knowledge was
tested prior to the training, immediately afterward, and one month later. Half of the subjects received tradi-
tional classroom training with visual aids; the other half were trained using a 3D immersive VR power-wall.
Significant advantage was found for VR training for stone cladding work and for cast-in-situ concrete work,
but not for general site safety. VR training was more effective in terms of maintaining trainees’ attention and
concentration. Training with VR was more effective over time, especially in the context of cast-in-situ con-
crete works. Given the need for improved training and the advantages of training using VR, incorporation of
VR in construction safety training is strongly recommended.
Keywords: Construction safety, immersive virtual reality, safety training.
Introduction
Making construction as safe as possible requires
concerted effort on the part of all involved: owners,
designers, construction companies at all levels of
management, construction workers, regulators and
educators (Hinze, 1997; Holt, 2001). Construction
workers themselves play a major role in determining
their own safety through their behaviour, either taking
or rejecting risks. Even when a company takes all rea-
sonable precautions in terms of preparing the site and
the work, enforcing compliance through disciplinary
measures, and providing training and personal protec-
tive equipment (PPE), workers can still make deci-
sions to work in ways that endanger them. Their
ability to identify risks, and their subjective analysis of
the magnitude of those risks, are among the factors
that determine their behaviour and thus their safety.
These skills are determined in part by safety
training. Training with low engagement (by lec-
tures, videos or demonstrations) which is common
in construction, has consistently been shown to be
minimally effective when compared with more
engaging forms of instruction (Burke et al., 2006).
Wilkins (2011) suggested that ‘there may be a
case for delivering less training in an artificial
classroom environment and more in the workplace,
where the practical ramifications of a failure to
adhere to safety regulations are more immediately
apparent’ (p. 1023). Given that one cannot know-
ingly expose construction workers to hazards, even
for purposes of instruction, the research reported
here explored the feasibility and the impact of
conducting safety training using a virtual construc-
tion site presented with a 3D immersive virtual
reality power-wall.
*Author for correspondence. E-mail: cvsacks@techunix.technion.ac.il
Construction Management and Economics, 2013
Vol. 31, No. 9, 1005–1017, http://dx.doi.org/10.1080/01446193.2013.828844
Ó2013 Taylor & Francis

Role and content of training for
construction safety
The relatively dangerous nature of the construction
industry, as reflected in accident statistics, is well doc-
umented. Recent publications that report statistics
from around the world include Gangolells et al.
(2010), Hu et al. (2011) and Zhou et al. (2012). The
loss of life, the costs and the liabilities associated with
construction accidents have led through legislation to
construction safety programmes, which include train-
ing, inspection and enforcement, and this approach
appears to have reduced accident rates where it has
been applied (Mitropoulos et al., 2005).
Many researchers have used empirical methods to
evaluate the effectiveness of training in improving
safety. A regression analysis of the safety strategies
and the site safety records of 45 Hong Kong con-
struction companies identified safety training as one
of the four most effective components of a safety pro-
gramme (Tam and Fung, 1998). A similar analysis of
70 Thai construction projects found that safety induc-
tions were effective in reducing unsafe conditions
(Aksorn and Hadikusumo, 2008). Based on a survey
of design and construction firms in Pennsylvania with
105 responses, Toole (2002) identified lack of
training as one of eight root causes of construction
accidents.
However, standards for construction safety training
are low. In the UK, Australia and Hong Kong, one
day or less of safety training is sufficient for workers
to obtain the necessary certification as construction
labourers (Li et al., 2012). Wilkins (2011) raised seri-
ous questions concerning the effectiveness and the
content of existing modes of construction safety train-
ing in the US. His survey of 105 construction person-
nel who had taken the OSHA 10-hour ‘Construction
Safety Training Course’ revealed dissatisfaction with
the ways in which the courses were taught. Wilkins
highlighted the needs for training to cover content
covered relevant to the lives of the trainees, for pre-
sentation by a trainer knowledgeable about the sub-
ject, and for supplementing training with tangible
materials that are understandable. In the broader
domain of safety training in general, a meta-analysis
study of research from 1971 to 2003 that compared
the effectiveness of safety training methods has shown
that high-engagement training (involving behavioural
modification, e.g. hands-on practice in a realistic
setting) was on average three times as effective as
low-engagement training (e.g. using videos or written
material only) (Burke et al., 2006). A similar
meta-study, covering the years 1996–2005, also found
(inter alia) evidence for the superiority of
high-engagement over low-engagement training,
although the difference was small (Robson et al.,
2010).
One of the assumptions of the ‘safety programme’
approach is that a worker’s role in an accident can be
thought of as that of a link in a chain of causation, as
typified by accident causation models like the classical
‘domino theory’ (Heinrich, 1936). In this view, train-
ing workers to identify hazards, and thereby avoid or
prevent them, is one of the defences that can help
break the chain. Abdelhamid and Everett (2000)
found that lack of awareness of hazards on the part of
construction workers was a significant contributing
factor to the occurrence of accidents. Mitropoulos
et al. (2005) convincingly argue, however, that the
‘safety programme’ approach has a limited view of
accident causality ‘as it ignores the work system fac-
tors and their interactions that generate the hazardous
situations and shape the work behaviours’ (p. 816).
They maintain that the uncertainty that is typical of
construction, coupled with the unstructured nature of
work tasks and interactions between crews, leads to
unpredictable work conditions that contribute to acci-
dent causation. In this view, two additional actions
are needed to increase safety: increase the reliability
of production planning and improve error manage-
ment capabilities. The latter refers to increasing work-
ers’ ability to avoid hazards and mitigate errors, which
can be enhanced by training workers to recognize
hazards through situational awareness, rather than
through training workers in prescriptive performance
of standardized work procedures.
Hazards that arise on construction sites may be
either predictable, as part of planned activities, or
emergent (i.e. arising in dynamic ways and resulting in
unpredictable situations). Cognitive and psychomotor
processes including decision-making capacity, atten-
tion, reaction time, contrast sensitivity, and visual pur-
suit are integral aspects of hazard perception skills
(Horswill et al., 2008; Su
¨mer, 2011). Accordingly, haz-
ard perception is a multi-component cognitive skill and
this skill can improve with experience (Deery, 1999).
Training can improve a person’s ability to diagnose the
potential hazard and risk correctly (Duffy, 2003) and
experience can have a positive impact on risk-taking
perception and behaviours particularly in the percep-
tion of the risk of injury (Knight et al., 2012).
In his seminal work defining the ‘perceived boundary
of acceptable performance’ model of worker behaviour
in relation to safety, Rasmussen (1997, p. 183) stated
that:
In spite of a widely used ‘defence-in-depth’ design
strategy, most recent major accidents in large scale
1006 Sacks et al.

industrial systems appear to be caused by operation of
the systems outside the predicted preconditions for
safe operation during critical periods with excessive
pressure from a competitive environment.
The model explains the opposing pressures on work-
ers. The dual pressures to increase productivity while
investing minimal personal effort are countered by the
restraining pressure from ‘campaigns for safety cul-
ture’. Safety training is central to the safety culture.
According to this model, improving the effectiveness
of training will contribute to inducing safety conscious
behaviour because the perceived boundary of accept-
able performance will lie within the boundary of func-
tionally safe performance.
For all of these reasons, the investigation of safety
training reported here focuses on improving the ability
of construction personnel to identify and respond to
hazards, rather than on prescriptive instruction of
standardized methods to perform specific tasks.
Safety training with virtual reality
Virtual reality (VR) is a technology that uses comput-
ers, software and peripheral hardware to generate a
simulated environment for its user. The environment
may simulate a real or an imaginary environment. An
immersive virtual environment (IVE) is a computer-gen-
erated environment that gives a person a sense of
being within it by engaging the person’s senses and
reducing or removing their perception of the real
environment. While there is a wide range of technical
implementations, an IVE will typically have the fol-
lowing features: it will surround its user, obscuring
cues from the physical environment and increasing
the sense of ‘presence’ within the IVE; provide a
three-dimensional visual representation of the virtual
environment; track the user’s location and orientation
and update the virtual scene to match the user’s
movements; and give the user some degree of control
over the objects in it (Bailenson et al., 2008). A cen-
tral notion is that of ‘presence’ in the virtual environ-
ment, i.e., that subjects’ responses in the virtual
environment are similar to their responses in a real
environment (Sanchez-Vives and Slater, 2005).
The use of VR in training workers of various kinds
is common. Flight simulators are a well-known
example and VR has been shown to be effective for
road-safety training (Thomson et al., 2005). VR train-
ing simulations for surgical procedures are common
because they avoid the danger inherent in training on
humans or animals (Sutherland et al., 2006).
The hazardous nature of construction sites in and
of itself makes onsite training difficult, and certainly
prevents training through experience of failure. One
solution is to construct purpose-built training facilities
that physically simulate the construction site (de Vries
et al., 2004); such facilities exist in the Netherlands,
the UK and in Australia. VR systems and IVE offer
the opportunity to expose workers to hazardous situa-
tions, and indeed to accidents, as part of training.
Workers can assess a situation, decide on a course of
action, implement the action and immediately observe
the results. This results in cognitive information pro-
cessing, which leaves its mark in long-term memory
(Lucas et al., 2008).
Lucas et al.’s (2008) research in safety training for
operators of construction equipment is one of the very
few examples of the use of VR for construction safety
training. Another example is a generic industrial
safety VR training game, developed for the California
State Compensation Insurance Fund, which contains
a module that simulates material handling for con-
struction of a timber roof on a house (ForgeFX,
2012). Ku and Mahabaleshwarkar proposed the pos-
sibility of construction safety training using ‘Second
Life’, highlighting numerous challenges (Ku and
Mahabaleshwarkar, 2011). Li et al. (2012) used a 3D
game engine to generate a ‘virtual safety assessment
system’ (VSAS) which uses VR scenes projected on a
PC to assess construction workers’ competence with
regard to safety.
As can be seen by the dearth of examples, the use
of immersive virtual reality for training interventions
in construction to date is rare and the knowledge
about their use and effectiveness is severely limited.
Some of the practical challenges for researchers have
been the significant investment required to prepare
the immersive VR content, including 3D scenery, ani-
mations and the pedagogical aspects, and the need to
bring workers to a fixed facility.
Aims and objectives
Despite the importance of training as a component of
safety programmes, the construction industry around
the world has yet to adopt a sufficiently sophisticated
approach. The criticisms of traditional modes of train-
ing, summarized in part in the literature review above,
have led various researchers to propose, and some to
test, the use of VR simulations for safety training. Yet
there is still much to be learned, both because the
effectiveness of IVE systems for construction has not
been tested rigorously and because the successful use
of IVE or VR safety training systems in other
industries cannot prove their effectiveness in the
construction industry.
VR safety training 1007

Therefore, we sought in this work to build a virtual
construction site using IVE technology, to compile a
set of safety training scenarios, and to test the setup
using a set of experiments with a between-group
design. The first hypothesis was that trainees would
perceive the virtual construction site environment to
be a sufficiently authentic simulation of a construction
site to facilitate learning. The second hypothesis was
that safety training in the virtual construction site
would be more effective, in terms of workers’ atten-
tion, learning and recall, and their ultimate success in
identifying and assessing construction safety risks,
than would training using conventional methods. The
latter essentially tests Burke et al. (2006)’s conclusions
concerning ‘high-engagement’ versus ‘low-engage-
ment’ training in the construction context. An addi-
tional specific goal was to gain practical knowledge
about the virtual construction site and its use: to learn
from the implementation and from the conduct of the
training sessions.
Method
The research compared safety training using
conventional methods vs. training, using immersive
virtual reality with a ‘between-participant’ experimen-
tal design. The experimental setup exploited the
controllable nature of virtual environments to allow
study of training interventions within a complex and
realistic environment. We adopted a behavioural
experiment research approach rather than using sur-
vey or other methods because experiments can better
determine cause-and-effect relationships (Campbell
et al., 1963). Three groups of 20 to 25 subjects each
took part in three replications of the experiment (an
earlier pilot experiment was performed with five sub-
jects). The procedure consisted of five steps:
(1) An individual safety knowledge test, to set the
baseline for measurement. The test scope and
technique is described in detail below.
(2) Safety training: at this stage each group was
randomly divided into two sub-groups of 10–
12 subjects each. Each sub-group received two
hours of training on the same construction
safety topics, but one sub-group received con-
ventional classroom instruction supported with
slides (photos, graphics and text), while the
other sub-group received the same instruction
in a virtual construction site.
(3) A second individual safety knowledge test was
applied immediately after training, using the
same test instrument.
(4) Subjects were asked to fill out an individual
experience questionnaire immediately after the
second safety test.
(5) A third safety knowledge test was given approx-
imately one month after training in order to
measure the subjects’ degree of recall. This test
is called the ‘short-term’ test, in accordance
with the definitions provided by Robson et al.
(2010, p. 19). Once again, the same instrument
was used.
In the research design, a number of steps were
taken to maximize reliability and validity: the groups
were divided randomly; a full-scale pilot experiment
was conducted to test the equipment and prepare the
experimenters; the experiment was repeated three
times; the subjects were tested for their baseline safety
knowledge before starting the experiments; and the
pre-experiment baseline results were compared across
the randomly divided sub-groups to confirm that
there was no significant difference in their initial
safety knowledge and risk perception skills. In the
latter case, a T-test comparison was made for risk
perception, prevention and risk evaluation results for
each of the three training topics. All nine results were
well outside the 10% limit of significance (six
approached the maximum 50% result for perfectly
matched populations), indicating that the groups
could not be distinguished from one another.
Population
A total of 71 subjects took part in this experiment in
four distinct groups, as shown in Table 1. After the
initial pilot phase, which served for testing and refin-
ing the method, three groups with a total of 66 sub-
jects were tested. Groups 1 and 3 were comprised of
construction workers from a cast-in-situ concrete
vocation training course. These subjects had limited
work experience, but they were familiar with con-
struction site environments. Group 2 was composed
of third year BSc civil engineering students. These
subjects also had limited work experience, and in this
group, most had little or no prior experience of the
construction site environment.
Immersive virtual environment
The virtual site consisted of a construction site
simulation displayed on an immersive virtual reality
power-wall. The power-wall setup consisted of three
rear-projection screens, each 2.4m wide and 1.8m
high, arranged in a ‘theatre’ with a 150˚ angle
1008 Sacks et al.

between each screen, as shown in Figure 1. This is an
open configuration of a three-sided EON ICube
CAVE (Cave Automatic Virtual Environment) that
uses 3D stereo projection with active glasses that have
a 120Hz frequency. The VR stereo software projec-
tion system was EON Studio V7.0. The instructor,
and in turns each of the trainees, used a head tracking
system and an XBOX controller that was also tracked
using eight cameras mounted on the tops of the
screens. Black curtains above the screens hid the ceil-
ing space beyond them and the 3D projectors were
the only source of light. The setup also has a stereo
sound system.
This setup fulfils many but not all of the conditions
for an ‘ideal’ IVE as outlined by Bailenson et al.
(2008) and others; it does not entirely exclude the
perception of the physical environment for all of the
users, and at any given time only the user wearing the
tracked glasses was at the vantage point of an actor in
the environment (the others were restricted to the
sense of accompanying him or her, having no control
of their own movement).
Training materials
The content for the training was selected based on
the frequency of accident types reported in Perlman
et al. (2012) as a guide. Within the more common
accident types, specific hazards were selected
according to their relevance to the local population
of construction workers, and their amenity to visual-
ization. The topics were organized in three
chapters:
(1) General site safety, including vehicle and
worker movement on site, working under
cranes, falls from heights and personal protec-
tive equipment (PPE).
(2) Safety in cast-in-situ concrete, including work
at heights, working with tools and equipment
(steel formwork, concrete pumps, reinforce-
ment).
(3) Safety during stone cladding work on facades,
including work on scaffolding, working with
electrical tools and winches.
Figure 1 Immersive VR power-wall setup in the virtual construction laboratory
Table 1 Experiment population
Group Description
Number of subjects in
training sessions
Number of completed competence
tests one month after training
Pilot 4th year BSC students, Faculty of Civil Engineering 5 –
1 Cast-in-situ concrete worker trainees (near the end
of their three-month course at the time)
20 2
2 3rd year BSc students, Faculty of Civil Engineering 25 15
3 Cast-in-situ concrete worker trainees (one month
into their three month course)
21 6
Totals for groups 1 to 3 66 23
VR safety training 1009

A half-hour lesson plan was prepared for each
chapter. The materials were collated from safety
codes and from the standard construction supervisors’
safety courses run by the Israel Institute for Occupa-
tional Safety and Hygiene (IIOSH). Visual material
was supplemented with numerous photographs taken
at three construction sites. A set of presentation slides
was prepared for each chapter (examples are shown in
Figure 2(a) and (b)). Once the presentations for the
classroom training were ready, the work of preparing
equivalent content for the virtual reality environment
began. The textual and theoretical slides were
unchanged; but all of the visual material was replaced
with VR scenarios, each with various possible safe
and unsafe starting situations and outcomes.
A detailed storyboard was prepared for each of the
21 VR scenarios. Figure 3 shows a typical example.
In this case, workers are tying reinforcement to form
a column cage. As can be seen in the photograph,
they are using two formwork shutter pieces as a tem-
porary work surface and face the danger of the shut-
ters falling on to them. Each storyboard included full
details of the mechanism of the accident scenario that
it described: which objects move, their motion paths,
rotation axes, what other objects they affect, etc.
All 21 scenarios were implemented within a virtual
construction site which was composed and tested in
earlier work (Perlman et al., 2012). The site incorpo-
rated a residential building under construction with
an entrance floor with a tall lobby space, four apart-
ments on each floor, elevator shafts and stairwells.
Figure 4 shows the entrance of the building and part
of the site. Cast-in-place concrete works were under-
way on its eighth floor, interior and finishing works
A person shall not stand under a load lifted by a
crane unless absolutely unavoidable, and if so, then
only for the shortest peroid of time necessary.
A person shall not approach nor handle a lifted load
unless the load is steady and it is no more than 1m
above the ground or the surface on which the person
is standing.
A load may only be lifted vertically with a crane, and
all necessary measures must be taken to prevent its
sway or rotation (e.g. direction ropes).
Working under a crane
(a)
Figure 2 Examples of slides from the general site safety chapter: (a) text instructions and (b) an illustration of the correct
way to receive a load
1010 Sacks et al.

were being performed in apartments on the third and
fourth floors, and stone cladding work was being done
on the facades. A work elevator, tower crane, concrete
pump and a variety of earthworks equipment and
trucks were modelled operating on the site. Static
equipment (scaffolding, formwork, shoring, various
interior work platforms, hand tools, reinforcing cages
and starter bars), materials (palettes of blocks, bags of
cement, floor tiling packages, etc.) and waste and
refuse were included to enhance the effect.
Background sounds of a construction site were used
for the audio track, together with the sounds of the
vehicles travelling.
Three software tools were used. The building itself
was modelled in REVIT; all other 3D geometry was
modelled using 3D Studio MAX, and the VR scenar-
ios were generated with EON Studio v6. The move-
ments of objects in the scenarios were implemented
in two ways: (1) ‘physics’ animation, in which objects
behave as rigid objects subject to the force of gravity
and move accordingly; and (2) ‘keyframe’ animation,
in which objects move along predefined paths. Physics
A ‘skydeck’
panel standing
on its wide side 3. The left hand
side panel starts
to collapse
because the
rebar cage is
pulling it
4. The rebar cage
falls down onto the
workers feet
2. The right hand side
stand starts to collapse
along the Axis of Rotation
1. The rebar cage
starts moving as
if connected to
the top face of
the stand
Workers
Axis of Rotation
Axis of Rotation
Figure 3 Storyboard example for reinforcement working table collapsing
Figure 4 A partial view of the virtual building under construction
VR safety training 1011

animation was used wherever possible, but in some
circumstances the inability of the physics engine to
deform objects under impacts generates highly unreal-
istic behaviour, resulting in the need to carefully pro-
gramme object motion paths.
The requirement for training of groups led to selec-
tion of a ‘third person’ view, in which the camera was
positioned so that all of the subjects see an avatar
who is the accident victim. An alternative configura-
tion uses a ‘first person’ view, in which the subject is
the victim, experiencing, for example, his or her own
fall. This option is suitable for a single trainee, but
not for groups. The default views for each scenario
were therefore positioned to make the scene tangible
and enhanced the experience of danger. Of course,
using the controller, the trainer was able to navigate
to any alternative point of view and to repeat the
virtual accident as needed.
Safety test
A safety knowledge test was applied to each subject
immediately before the training, immediately after the
training, and again one month later. The third test
was applied remotely, with the result that it was not
possible to maintain full participation. Of the 52 sub-
jects who began the third test, 23 completed it.
The same test was used at all three testing steps.
The correct answers were not revealed at any point.
The test had three chapters, one for each training
topic: general site safety, cast-in-situ concrete work
and stone cladding work. Each chapter had three
question types:
(1) Open questions, where the subject had to
identify different potential hazards apparent in
a set of photographs from construction sites,
recommend ways to eliminate them, and assess
their risk level (on a scale of 1 to 5).
(2) Behaviour questions, where subjects were
presented with an image of a construction
scene with a specific hazard and asked how
they would behave in the situation depicted:
(a) work as usual; (b) inform the foreman and
continue working as usual; (c) fix the hazard-
ous situation themselves; or (d) refuse to work
in that situation.
(3) Knowledge questions, where the subjects had
to identify appropriate safety equipment, safety
instructions and terminology.
To score the open question, responses of all of the
subjects were scanned and a list of all the hazards
identified was composed. Individual scores were then
computed by evaluating what proportion of the full
list of hazards each subject had identified. Both the
total number of identifications and the number of
unique identifications were used (in many cases, sub-
jects identified the same hazard more than once in
their responses). Similarly, the score for eliminating
hazards was the number of unique and correct meth-
ods identified for eliminating each hazard. To score
the risk level assessments, a panel of three IIOSH
construction safety instructors was asked to set a nor-
mative value for each hazard’s probability, severity
and risk level. The score for risk level for each subject
was computed as the standard deviation of the set of
the absolute differences between the subject’s answers
and the safety experts’ answers.
Training experience questionnaire
Immediately after the training, the subjects were
asked to complete a questionnaire to assess their
experience. The questionnaire included questions
about knowledge, emotions and attitudes. The goal
was to compare the experience of subjects in the VR
training with that of those who had the traditional
training. The questions posed are detailed in Table 5
below together with the results.
Results
The safety test scores were compared in three
different ways, to determine immediate effectiveness
of the training, short-term effectiveness, and the
degree of recall. The results of the learning experience
questionnaire were used to triangulate the three safety
test score comparisons and to reveal other aspects of
the training’s effect on the subjects.
The first set of results compares the subjects’ safety
test scores from the tests taken just prior to the train-
ing with the scores from the tests taken immediately
after the training. The differences between the
‘before’ and ‘after’ scores were computed and the sig-
nificance of the difference between the two popula-
tions was evaluated using T-tests. The degree of
significance (95%, 90% or none) is noted in the
penultimate column of Table 2.
The results demonstrate effectiveness in immediate
learning of hazard identification and prevention skills
for both the VR and the traditional training groups
(Table 2). Over all three of the safety training topics,
the differences between before and after scores for
risk identification (t = –5.77, p < 0.05) and prevention
(t = –4.92, p < 0.05) were significant. Within the
topics, significant differences were found for the
1012 Sacks et al.

topics ‘general site safety’ in risk identification scores
(t = –4.1, p < 0.05) and in prevention scores (t=–
2.74, p < 0.05); similarly, for ‘reinforced concrete
works’ significant differences between before and after
scores were found for risk identification (t = –1.76, p <
0.1); for ‘stone cladding works’, significant differences
between before and after scores were also found for
risk identification (t = –4.42, p < 0.05) and prevention
(t = –5.56, p < 0.05). Interestingly, those who had the
traditional training rated the risk levels higher after the
training; the VR training groups, on the other hand,
reduced their risk level assessments after training.
The results were also used to assess any possible
advantage of VR over traditional training. For this,
the margin of improvement of the test scores for each
type of training was compared. Here, the results were
different for the different sub-topics of the safety
training. VR was significantly better than traditional
training for hazard identification for reinforced con-
crete works (t = 1.08, p < 0.1) and for stone cladding
works (t = 1.76, p < 0.05); it was also significantly
better for learning prevention knowledge for the cast-
in-situ concrete works (t = 3.29, p < 0.05). Although
the results indicate an advantage for the general site
safety topic as well, this cannot be asserted with confi-
dence.
Short-term effectiveness of the training was
investigated by comparing scores for the safety tests
administered before training and one month after
training (see results in Table 3). In all cases, scores on
identification task and on the prevention task were
significantly improved after training. In particular we
found training to be effective in scores on risk
identification for reinforced concrete works (t = –2.57,
p < 0.05) and stone cladding works (t = 2.28, p < 0.1),
and in scores for prevention for general site safety
(t = –3.85, p < 0.1) and stone cladding works (t = 1.76,
p < 0.05). Due in part to the small sample size, it was
not possible to establish a clear-cut advantage for the
virtual reality training, with the exception of cast-in-situ
concrete works, where the superiority of the VR train-
ing proved to be statistically significant for scores on
identification task (t = 2.58, p < 0.05) and on the pre-
vention task (t = 1.85, p < 0.05).
We also investigated recall after training by com-
paring scores after training to scores one month later
(see results in Table 4). In most cases the differ-
ences between the two times were not significant,
indicating that recall was good. However we did find
some differences in scores for risk identification for
general site safety (t = 2.19, p < 0.1) and stone clad-
ding works (t = 3.14, p < 0.05) and for prevention
for general site safety (t = 2.18, p < 0.1) and stone
cladding works (t = 4.40, p < 0.05). However, with
the exception of risk level assessment for stone
cladding works (t = 2.42,p < 0.05), we could not
demonstrate significant differences between VR and
traditional training.
Finally, participants were all asked to complete a
questionnaire in which they evaluated various aspects
Table 2 Safety testing data to examine immediate effectiveness (n = 66)
Skill
Traditional
groups test results
Virtual reality
groups test results
Training effect (sig.) Advantage of VR (sig.)Before After Before After
Total over all safety training chapters
Identification 9.77 11.17 9.67 13.08
⁄⁄ ⁄⁄
Prevention 6.47 7.57 7.69 11.06
⁄⁄ ⁄⁄
Risk level assessment 1.55 1.98 2.11 1.83 NS NS
General site safety
Identification 4.03 4.93 3.86 4.89
⁄⁄
NS
Prevention 2.80 3.23 3.06 4.17
⁄⁄
NS
Risk level assessment 0.64 0.72 0.73 0.67 NS NS
Reinforced concrete works
Identification 1.80 1.60 1.86 2.58
⁄⁄
Prevention 1.33 0.97 1.75 2.44 NS
⁄⁄
Risk level assessment 0.52 0.68 0.70 0.63 NS
⁄⁄
Stone cladding works
Identification 3.93 4.63 3.94 5.61
⁄⁄ ⁄⁄
Prevention 2.33 3.37 2.89 4.44
⁄⁄
NS
Risk level assessment 0.52 0.54 0.63 0.53 NS NS
Notes: Safety tests applied before training and immediately after training. T-test results:
⁄⁄
p < 0.05,
⁄
p < 0.1, NS = no significance.
VR safety training 1013

of the learning experience. The results are shown in
Table 5. Here we found a significant advantage for
VR over traditional training (t = –5.65,p < 0.05) for
a variety of parameters. The advantages of VR were
measured in question 2 (‘To what extent was your
feeling strong that the demonstrations represent real
situations in construction sites?’) and question 6
(‘During the training, did you feel discomfort due to
Table 3 Safety testing data to examine the short-term training effectiveness (n = 23)
Skill
Traditional groups test
results
Virtual reality groups test
results
Training effect (sig.) Advantage of VR (sig.)Before After one month Before After one month
Total over all safety training chapters
Identification 7.50 8.67 8.88 11.06
⁄⁄
NS
Prevention 5.50 6.17 9.59 11.06
⁄
NS
Risk level assessment 1.57 1.86 2.10 1.68 NS NS
General site safety
Identification 3.17 3.50 3.53 4.18 NS NS
Prevention 2.50 2.50 3.88 4.06
⁄
NS
Risk level assessment 0.52 0.90 0.88 0.55 NS NS
Reinforced concrete works
Identification 1.67 1.67 2.06 3.00
⁄⁄ ⁄⁄
Prevention 1.33 1.00 2.35 2.24 NS
⁄⁄
Risk level assessment 0.66 0.67 0.58 0.44 NS NS
Stone cladding works
Identification 2.67 3.50 3.29 3.88
⁄
NS
Prevention 1.67 2.67 3.35 3.76
⁄⁄
NS
Risk level assessment 0.93 0.29 0.64 0.69 NS NS
Notes: Safety tests applied before training and one month after training. T-test results:
⁄⁄
p < 0.05,
⁄
p < 0.1, NS = no significance.
Table 4 Safety testing data to examine recall (n = 23)
Skill
Traditional groups test
results
Virtual reality groups test
results
Recall effect (sig.) Advantage of VR (sig.)After After one month After After one month
Total over all safety training chapters
Identification 10.00 8.67 13.35 11.06
⁄
NS
Prevention 10.00 6.17 13.47 11.06
⁄
NS
Risk level assessment 2.15 1.86 1.61 1.68 NS NS
General site safety
Identification 4.33 3.50 5.12 4.18
⁄
NS
Prevention 4.17 2.50 5.12 4.06
⁄
NS
Risk level assessment 0.62 0.90 0.65 0.55 NS NS
Reinforced concrete works
Identification 1.17 1.67 2.88 3.00 NS NS
Prevention 1.33 1.00 3.00 3.24 NS NS
Risk level assessment 0.84 0.67 0.48 0.44 NS NS
Stone cladding works
Identification 4.50 3.50 5.35 3.88
⁄⁄
NS
Prevention 4.50 2.67 5.35 3.76
⁄⁄
NS
Risk level assessment 0.66 0.29 0.49 0.69 NS
⁄⁄
Notes: Safety tests applied immediately after training and one month after training. T-test results:
⁄⁄
p < 0.05,
⁄
p < 0.1, NS = no signifi-
cance.
1014 Sacks et al.

the thought of a possible accident at work?’). Both
had significant results (D= 0.4, p = 0.084 and
D= 0.4, p = 0.096 respectively).
The results also appear to reflect an observation
made during the training concerning the degrees of
attention and concentration of the trainees. The
researcher who observed the traditional training ses-
sions noted that trainees tended to lose concentration
after about 40 minutes; they asked to leave the room
to freshen up, they began to use their mobile phones,
and their attention was diverted from the material.
The instructors allowed them a break. In contrast, the
virtual reality trainees were observed to maintain full
focus for the hour and a half of the training session.
The results for question 8 (‘Did you feel you were
concentrating in class?’; D= 0.8, p = 0.0002) and
question 9 (‘To what extent was the learning a pleas-
ant experience?’; D= 0.5, p = 0.01) show this differ-
ence. While these results do not necessarily reflect or
measure the sense of presence as discussed above,
they are in line with the literature; for example, tests
have shown that children more easily pay attention in
a virtual classroom environment and that VR is effec-
tive in sustaining one’s attention (Cho et al., 2002).
Discussion and conclusions
This research has shown that it is possible to build a
virtual construction site using IVE technology that
incorporates safety scenarios and that it can be used
effectively for training. The first hypothesis, that
trainees would perceive the virtual construction site
environment to be a sufficiently authentic simulation
Table 5 Results of the learning experience questionnaire (n = 66)
Question
Traditional
(n = 31)
VR
(n = 35)
T-test
significance
1. Did you feel that the dangers were demonstrated realistically? 4.0 4.0 NS
2. To what extent was your feeling strong that the demonstrations
represent real situations in construction sites?
3.8 4.2
⁄
3. How strongly will the training affect your learning about safety? 3.8 4.2
⁄⁄
4. To what extent will you remember what you’ve learned a year from now? 3.4 3.8
⁄
5. To what extent will training affect your behaviour on a construction site? 3.8 4.4
⁄⁄
6. During the training, did you feel discomfort
due to the thought of a possible accident at work?
3.2 3.6
⁄
7. Could the training illustrate realistic situations in the field? 3.9 4.0 NS
8. Did you feel you were concentrating in class? 3.4 4.2
⁄⁄
9. To what extent was the learning a pleasant experience? 3.7 4.2
⁄⁄
10. Will you recommend similar training to your friends? 3.6 4.2
⁄⁄
11. Do you want to have similar training in the future? 3.8 4.2
⁄
12. Will the training help you avoid accidents on the site? 4.1 4.5
⁄⁄
13. To what extent was the time of the training a worthwhile investment? 4.1 4.5
⁄⁄
14. Overall, was the training a pleasant learning experience? 4.1 4.6
⁄⁄
15. Will the training influence your attitudes to safety? 4.3 4.5
⁄
16. Rate the quality of training on general site safety. 4.2 4.4
⁄
17. Rate the quality of training on safety in reinforced concrete works. 4.2 4.2 NS
18. Rate the quality of training on safety in stone cladding works. 4.3 4.3 NS
19. During the training, what was the strength of feeling of
discomfort from each of the following potential accidents:
a. Man falling from the scaffold to the ground 4.0 4.0 NS
b. Injury to a worker from exposed reinforcing bars 3.3 3.8
⁄
c. Run over by a truck 3.5 3.5 NS
d. Scaffolding collapse on people 3.7 3.8 NS
e. Worker crushed by a concrete beam lifted by a crane 3.5 3.9 NS
f. Worker crushed by steel shuttering/formwork for concrete 3.7 3.7 NS
g. Worker injured by falling objects 3.8 3.6 NS
All 3.7 4.0
⁄⁄
Notes: Scores range from 1 = ‘not at all’ to 5 = ‘very much’ or ‘very strongly’. T-test results:
⁄⁄
p < 0.05,
⁄
p < 0.1, NS = no significance.
VR safety training 1015

of a construction site to facilitate learning, is
supported both by the VR learning outcomes and by
the responses to the post-experiment evaluations. In
general, we found that instruction using virtual reality
was more effective than safety training with traditional
classroom training methods using slide presentations,
thus confirming the second hypothesis.
Virtual reality training was the more effective learn-
ing experience. Specifically, we found a distinct
advantage for virtual reality training for stone cladding
work and for cast-in-situ concrete work. We did not
find a clear advantage for site safety in general. Train-
ees maintained a high level of alertness for the entire
period in the virtual reality training. By contrast, in
normal training, trainees were unable to maintain
concentration beyond the first hour. Finally, training
with virtual reality was more effective over time than
traditional training, especially in cast-in-situ concrete
works. It was not possible to determine unambigu-
ously the advantage of learning effectiveness of VR in
the long run for the other types of work. Those
trained with traditional lessons rated risk levels higher
after training, whereas those trained with VR lowered
their risk assessments. These results are consistent
with Burke et al.’s (2006) findings that the higher the
level of engagement in safety training, the more effec-
tive it is. Given the average age of the trainee subjects
(almost all in their early 20s), the benefit of comput-
erized learning for people accustomed to computer-
ized environments identified by Font (2004) may also
explain some of the advantages.
Although subjects rated the VR scenarios highly in
terms of the realism of their depiction of situations
in construction sites, the advantages of the VR train-
ing may have been underestimated. This is because
the VR scenarios built for the experiments were lim-
ited in their degree of sophistication and reality.
Preparation of VR scenarios is an intensely time-con-
suming task, requiring attention to fine details and
with a steep learning curve. Therefore, whereas the
traditional training was based on mature materials,
in a commercial setting the VR training materials
could likely be improved significantly beyond the
level achieved. This may have increased the likeli-
hood of identifying significant differences between
the two approaches.
A basic conflict found in the course of the research
is the apparent need to reduce the degree of intensity
of the learning experience as the size of the trainee
group increases. Maximum effect is obtained for an
individual in immersive 3D when he or she experi-
ences the ‘virtual’ hazard first hand, i.e. when they
are navigating the model themselves. However, with
groups of trainees (10–12 in each group in this case),
time constraints for training prevent use of the
controller for each scenario by each and every trainee.
The nature of the virtual reality scenarios themselves
is dependent on the degree to which the trainees will
be observers rather than direct participants. With
fewer participants, individual subjects can be given
more opportunities to control the environment, thus
maximizing the effectiveness of the learning (Bailen-
son et al., 2008). Further research is needed to find
the correct balance between the degree of first hand
control that is needed, on the one hand, and the abil-
ity to train large groups of construction workers.
The results and their discussion highlight important
advantages of virtual reality environments for con-
struction safety training. First, VR has been shown to
be suitable for presenting trainees with hazards
directly and realistically without compromising their
safety. Second, the research has shown that safety
training with VR holds the attention of trainees better
than conventional classroom training does. Thirdly,
VR can be used to give trainees a measure of control
over the environment, thus reinforcing learning. The
primary disadvantage lies in the cost of developing
training materials and virtual construction sites.
Against the background of the dissatisfaction with
existing modes of training expressed by Li et al.
(2012) and Wilkins (2011), among others, there is a
clear need for improved training for construction
safety. The results of this research have shown that
the use of training that incorporates VR is feasible
and has distinct advantages over traditional methods.
Its use with other interactive methods is therefore
strongly recommended.
Acknowledgements
The authors are indebted to the construction safety
instructors of the Israel Institute for Occupational
Safety and Hygiene (IIOSH) for their assistance in
preparing the training materials; to the subjects from
the Construction Industry Labour Incentive Training
Program and the Holon Vocational Training College;
and to the subjects from the Technion. This research
was funded in part by a grant from the Israel Ministry
of Industry, Trade and Employment, Office of the
Chief Works Inspector.
References
Abdelhamid, T.S. and Everett, J.G. (2000) Identifying root
causes of construction accidents. ASCE Journal of
Construction Engineering and Management,126(1), 52–60.
Aksorn, T. and Hadikusumo, B.H.W. (2008) Measuring
effectiveness of safety programmes in the Thai construc-
tion industry. Construction Management and Economics,26
(4), 409–21.
1016 Sacks et al.

Bailenson, J.N., Yee, N., Blascovich, J., Beall, A.C.,
Lundblad, N. and Jin, M. (2008) The use of immersive
virtual reality in the learning sciences: digital transforma-
tions of teachers, students, and social context. Journal of
the Learning Sciences,17(1), 102–41.
Burke, M.J., Sarpy, S.A., Smith-Crowe, K., Chan-Serafin,
S., Salvador, R.O. and Islam, G. (2006) Relative
effectiveness of worker safety and health training methods.
American Journal of Public Health,96(2), 315–24.
Campbell, D.T., Stanley, J.C. and Gage, N.L. (1963)
Experimental and Quasi-Experimental Designs for Research,
Houghton Mifflin, Boston.
Cho, B.-H., Ku, J., Jang, D.P., Kim, S., Lee, Y.H., Kim, I.
Y., Lee, J.H. and Kim, S.I. (2002) The effect of virtual
reality cognitive training for attention enhancement.
CyberPsychology & Behavior,5(2), 129–37.
de Vries, B., Verhagen, S. and Jessurun, A.J. (2004)
Building management simulation centre. Automation in
Construction,13(5), 679–87.
Deery, H.A. (1999) Hazard and risk perception among
young novice drivers. Journal of Safety Research,30(4),
225–36.
Duffy, V.G. (2003) Effects of training and experience on
perception of hazard and risk. Ergonomics,46(1–3),
114–25.
Font, C.M. (2004) The virtual construction of the mind:
the role of educational psychology. Interactive Educational
Multimedia,9, 32–47.
ForgeFX (2012) State Compensation Insurance Fund,
Safety Training Game, available at http://bit.ly/18hLkjM
(accessed 1 August 2012).
Gangolells, M., Casals, M., Forcada, N., Roca, X. and
Fuertes, A. (2010) Mitigating construction safety risks
using prevention through design. Journal of Safety
Research,41(2), 107–22.
Heinrich, H.W. (1936) Industrial Accident Prevention,
McGraw-Hill, New York.
Hinze, J.W. (1997) Construction Safety, Prentice Hall, Upper
Saddle River, NJ.
Holt, A.S.J. (2001) Principles of Construction Safety, Black-
well Science, London.
Horswill, M.S., Marrington, S.A., McCullough, C.M.,
Wood, J., Pachana, N.A., McWilliam, J. and Raikos, M.
K. (2008) The hazard perception ability of older drivers.
Journals of Gerontology Series B: Psychological Sciences and
Social Sciences,63(4), 212–18.
Hu, K., Rahmandad, H., Smith-Jackson, T. and
Winchester, W. (2011) Factors influencing the risk of falls
in the construction industry: a review of the evidence.
Construction Management and Economics,29(4), 397–416.
Knight, P.J., Iverson, D. and Harris, M.F. (2012) Early driv-
ing experience and influence on risk perception in young
rural people. Accident Analysis & Prevention,45, 775–81.
Ku, K. and Mahabaleshwarkar, P.S. (2011) Building
interactive modeling for construction education in virtual
worlds. Journal of Information Technology in Construction
(ITcon),16, 189–208.
Li, H., Chan, G. and Skitmore, M. (2012) Visualizing
safety assessment by integrating the use of game technol-
ogy. Automation in Construction,22, 498–505.
Lucas, J., Thabet, W. and Worlikar, P. (2008) A VR-based
training program for conveyor belt safety. Journal of
Information Technology in Construction (ITcon),13,
381–407.
Mitropoulos, P., Abdelhamid, T. and Howell, G. (2005)
Systems model of construction accident causation. Jour-
nal of Construction Engineering and Management,131(7),
816–25.
Perlman, A., Sacks, R. and Barak, R. (2012) Hazard recog-
nition and risk perception in construction. Safety Science
(forthcoming).
Rasmussen, J. (1997) Risk management in a dynamic
society: a modelling problem. Safety Science,27(2–3),
183–213.
Robson, L., Stephenson, C., Schulte, P., Amick, B., Chan,
S., Bielecky, A., Wang, A., Heidotting, T., Irvin, E., Egg-
erth, D., Peters, R., Clarke, J., Cullen, K., Boldt, L.,
Rotunda, C. and Grubb, P. (2010) A Systematic Review
of the Effectiveness of Training & Education for the Protection
of Workers, Institute for Work & Health and National
Institute for Occupational Safety and Health, Toronto
and Cincinnati, OH.
Sanchez-Vives, M.V. and Slater, M. (2005) From presence
to consciousness through virtual reality. Nature Reviews
Neuroscience,6(4), 332–9.
Su
¨mer, N. (2011) Cognitive and psychomotor correlates of
hazard perception ability and risky driving. Paper pre-
sented at the Sixth International Driving Symposium on
Human Factors in Driver Assessment, Training and Vehicle
Design, Olympic Valley – Lake Tahoe, CA, 27-30 June,
Transportation Research Board (TRB), Washington DC.
Sutherland, L.M., Middleton, P.F., Anthony, A., Hamdorf,
J., Cregan, P., Scott, D. and Maddern, G.J. (2006) Sur-
gical simulation: a systematic review. Annals of Surgery,
243(3), 291–300.
Tam, C.M. and Fung, I.W.H. (1998) Effectiveness of
safety management strategies on safety performance in
Hong Kong. Construction Management and Economics,
16(1), 49–55.
Thomson, J.A., Tolmie, A.K., Foot, H.C., Sarvary, P.,
Whelan, K.M. and Morrison, S. (2005) Influence of vir-
tual reality training on the roadside crossing judgments of
child pedestrians. Journal of Experimental Psychology,11
(3), 175–86.
Toole, T. (2002) Construction site safety roles. Journal of
Construction Engineering and Management,128(3), 203–10.
Wilkins, J.R. (2011) Construction workers’ perceptions of
health and safety training programmes. Construction
Management and Economics,29(10), 1017–26.
Zhou, W., Whyte, J. and Sacks, R. (2012) Construction
safety and digital design: a review. Automation in
Construction,22, 102–11.
VR safety training 1017