Integration of safety risk data.doc
Esmaeili, B., & Hallowell, M. (2013). Integration of safety risk data with highway construction schedules.Construction Management & Economics, 31(6), 528-541. doi:10.1080/01446193.2012.739288
The construction industry is characterized by a relatively high injury and illness rate compared to other industries. Within theconstruction industry, the highway construction and maintenance sector is one of the most dangerous. To improve safety in this sector, proactive methods of safety improvement and reliable risk data are needed. The safety risk quantification is the first step towards integrating safety data into design and planning. To enhance the current preconstruction safety practices,safety risks of highway construction and maintenance tasks were quantified and a decision support system was developed and tested that integrates safety risk data into the project schedules. Relative safety risks were quantified for 25 common highway construction tasks using the Delphi method. To ensure valid and reliable results, experts were selected according to rigorous requirements and multiple controls were employed to decrease cognitive biases. The data were incorporated into a decision support system called Scheduled-based Safety Risk Assessment and Management (SSRAM) that facilitates integration of safety risk data with project schedules. The resulting data-driven system produces predictive plots of safetyrisk over time based on the temporal and spatial interactions among concurrent activities. To test the utility of the decision support system and the validity of the underlying risk data, the system was tested on 11 active case study projects in the US. It was found that the database and associated decision support tool produce accurate and reliable risk forecasts that increase the viability of existing safety preconstruction activities. [ABSTRACT FROM AUTHOR
Intergration of safety risk data with highway construction schedules.pdf
Integration of safety risk data with highway construction schedules
BEHZAD ESMAEILI* and MATTHEW HALLOWELL
Department of Civil, Environmental, and Architectural Engineering, University of Colorado at Boulder, 428 UCB,
1111 Engineering Drive, Boulder, CO 80303, USA
Received 29 March 2012; accepted 9 October 2012
The construction industry is characterized by a relatively high injury and illness rate compared to other
industries. Within the construction industry, the highway construction and maintenance sector is one of the
most dangerous. To improve safety in this sector, proactive methods of safety improvement and reliable risk
data are needed. The safety risk quantification is the first step towards integrating safety data into design
and planning. To enhance the current preconstruction safety practices, safety risks of highway construction
and maintenance tasks were quantified and a decision support system was developed and tested that inte-
grates safety risk data into the project schedules. Relative safety risks were quantified for 25 common
highway construction tasks using the Delphi method. To ensure valid and reliable results, experts were
selected according to rigorous requirements and multiple controls were employed to decrease cognitive
biases. The data were incorporated into a decision support system called Scheduled-based Safety Risk
Assessment and Management (SSRAM) that facilitates integration of safety risk data with project schedules.
The resulting data-driven system produces predictive plots of safety risk over time based on the temporal
and spatial interactions among concurrent activities. To test the utility of the decision support system and
the validity of the underlying risk data, the system was tested on 11 active case study projects in the US. It
was found that the database and associated decision support tool produce accurate and reliable risk forecasts
that increase the viability of existing safety preconstruction activities.
Keywords: Decision support systems, occupational health and safety, risk management, scheduling.
Typically, safety management activities take place
during the construction phase (e.g. job hazard analy-
ses and site audits). In recent years, new safety man-
agement strategies have been introduced that help the
project team to identify and control hazards during
design and preconstruction. However, according to
Szymberski (1997), the potential to influence site
safety and health conditions decreases exponentially
as the project commences. Recent research has con-
firmed these findings and indicates that the most
effective safety programme elements occur during the
programming and preconstruction phases (Rajendran
and Gambatese, 2009). Unfortunately, the current
methods for considering safety and health in these
early phases are inconsistent, informal, and based
primarily on intuition and judgment (Hallowell,
2008). Thus, there is clearly a need to enhance pre-
construction safety management strategies, to create
user-friendly tools, and to increase their use in all
sectors of the industry.
One of the preconstruction methods that has been
shown to be highly effective is the integration of the
safety aspect into project schedules using risk data
(Yi and Langford, 2006). Unfortunately, integration
is limited because of a lack of data for specific con-
struction work tasks and the lack of reliable tools that
interface with existing scheduling software. The cur-
rent study aims to test the theory that loading safety
risk data into the project schedule is practical and will
improve predictions of high risk work periods. The
objectives are to (1) quantify relative safety risk values
for common highway construction activities; (2)
*Author for correspondence. E-mail: firstname.lastname@example.org
Construction Management and Economics, 2013 Vol. 31, No. 6, 528–541, http://dx.doi.org/10.1080/01446193.2012.739288
� 2013 Taylor & Francis
integrate these risk data into project schedules using a
novel decision support system; and (3) validate the
analytical procedure on case study projects.
This study focuses on risk quantification and risk
modelling for highway construction because the high-
way construction sector is one of the most dangerous
in the industry (Bureau of Labor Statistics, 2012). In
2005, this sector accounted for approximately 469
vehicle- and mobile heavy equipment-related deaths,
279 of which (59%) occurred in traffic work zones
(Center for Construction Research and Training,
2008). Furthermore, the Federal Highway Adminis-
tration (2004) estimates that a work zone fatality
occurs once every 10 hours and a work zone injury
occurs every 13 minutes. The presence of high-speed
traffic near work zones, prevalence of night-time
work, use of heavy equipment, exposure to weather,
and highly repetitive work tasks contribute to this rel-
atively high number of injuries (Bryden and Andrew,
1999; Arditi et al., 2005). Thus research is needed to
help practitioners to identify, analyse and respond to
high risk periods on highway construction projects.
This study was guided by a large body of literature.
In particular, literature that focused on the safety-
schedule integration and construction engineering and
management (CEM) decision support systems (DSSs)
proved to be most helpful. This body of literature was
used to guide the risk quantification process and the
development of a framework for integrating safety risk
into project schedules. A review of the salient findings
from relevant literature is provided below.
Safety schedule integration
Integrating safety planning and management in early
phases of construction projects is essential to effective
injury prevention and the development of a culture of
safety (Tarrants, 1980; Sawacha et al., 1999). Coble
and Elliott (2000) argued that integration of safety
aspects into planning starts with considering safety
during the scheduling of a construction project. There
have been a multitude of studies that attempt to inte-
grate various forms of safety information with project
schedules. These studies can be divided into two gen-
eral categories: those that attempted to cover safety
planning, injury prevention and regulatory informa-
tion, and those that integrated risk data. The majority
of studies focused on the former because these safety
data, such as regulatory information, are readily
available and not difficult to obtain.
Safety-schedule integration began with the work of
Kartam (1997) who designed a framework for
integrating extensive safety knowledge (e.g. Occupa-
tional Safety and Health Administration (OSHA)
regulations) into critical path method (CPM) sched-
ules using Microsoft Project, Primavera P6, Primavera
Suretrack and Timeline. According to Hinze et al.
(2005) the major weakness of this initial effort was
that there was never any success in making a link
between the safety elements and the electronic sche-
dule. In response to this shortcoming, Hinze et al.
(2005) built upon this research effort by developing
SalusLink, a tool that allows project managers to
access textual safety data contained in databases
managed by Primavera P6 and Suretrack. Though a
working prototype was produced, the software is not
commercially available. Saurin et al. (2004) and
Cagno et al. (2001) took a different approach by
developing safety planning and control models that
attached injury prevention strategies and methods of
safety planning to scheduled activities.
In the past five years, researchers have attempted
to integrate risk data into project schedules as a
means to identify high risk work periods and
leverage scheduling controls to prevent periods of
excessive risk. For example, Wang et al. (2006)
developed a simulation-based model (SimSAFE) that
integrates expected injury cost data for each activity
in a network schedule. This stand-alone software
system allows safety managers to identify work zones
that are associated with relatively high risk as mea-
sured by cumulative potential accident costs. Yi and
Langford (2006) took risk integration a step further
by developing a robust framework for ‘safety
resource scheduling’ using patterns which are similar
to resource levelling. Although Yi and Langford
(2006) offered a strong framework for the integration
of safety risk data with project schedules, there were
no robust risk data as the database only included
fatalities that occurred as a result of falls from
height. Furthermore, Navon and Kolton (2006,
2007) created an automated monitoring and control
model that is capable of identifying fall hazards and
their location. The major limitations of this body of
literature are that there is not a robust safety risk
database and the interactions (i.e. compatibility and
incompatibility) among tasks were ignored.
Researchers have begun to model the interactions
among risk factors and create frameworks that inte-
grate detailed user-provided data into preconstruction
planning tools. For example, a series of studies mod-
elled the spatial and temporal interactions of concur-
rent work tasks by using information available in 4D
geographic models and user-provided data for ‘loss-of-
control events’ (Rozenfeld et al., 2009, 2010; Sacks
et al., 2009). The major limitation of these models is
that hazards related to each task must be identified
Safety risk data 529
and quantified by the user, which can be time inten-
sive, laborious and unrealistic in practice (Rozenfeld
et al., 2009). In another study, Hallowell et al. (2011)
adapted Yi and Langford’s (2006) model and sug-
gested a new framework to integrate safety risk data
into project schedules. In addition to integrating base-
level risks for individual tasks, this framework also
considered robust task interactions obtained through
the Delphi process. The limitation of this work was
that they did not test the applicability of the framework
on actual projects and base-level task risks were not
quantified. Thus, the current research aims to address
the limitations of the previous studies by quantifying
highway construction safety risks for common work
tasks and testing the efficacy of the framework
presented by Hallowell et al. (2011) on active projects.
CEM decision support systems
As computing technologies have improved, increased
attention has been paid to the development of com-
puter applications that increase the speed and quality
of decision making. One category of these tools is
decision support systems (DSSs), which are defined
as, ‘an interactive IT-based system that helps decision
makers utilize data and models in making their deci-
sions’ (Carter et al., 1992, p. 3). Typically, the two
main objectives for using DSSs are: performing a
given task in the decision-making process more
quickly and with fewer resources (efficiency); and
improving the quality of the outcome of the decision-
making process (effectiveness). In addition, DSSs
help a manager to make more informed decisions,
consider a multitude of criteria and alternatives,
reduce the time needed to make an effective decision,
and focus attention on the most important elements
of a scenario. They also reduce complexity of the
problem to a manageable level and reduce uncertainty
(Carter et al., 1992). In CEM, DSSs have been uti-
lized in many areas such as: resource sharing (Perera,
1983); prequalifying subcontractors (Russell et al.,
1990); optimizing heavy lift planning (Lin and Hass,
1996); resource levelling (Leu et al., 2000); making
go/no-go decisions for international projects (Han and
Diekmann, 2001); selecting appropriate project deliv-
ery methods (Molenaar and Songer, 2001); schedul-
ing steel fabrication (Karumanasseri and AbouRizk,
2002); and providing guidance during dispute resolu-
tion (Palaneeswaran and Kumaraswamy, 2008).
In addition to the applications mentioned above,
some DSSs have been developed to enhance decision
making in the area of safety. For example, Kak et al.
(1995) developed a knowledge-based program to
facilitate access to the explicit safety knowledge on
construction sites. Their program searches applicable
safety regulations (e.g. OSHA) for a particular task
and provides suggestions to improve compliance.
Gambatese et al. (1997) presented a tool for incorpo-
rating safety-related issues in the design phase of a
project called ‘Design for Construction Safety Tool-
Box’, which has the ability to identify project-specific
hazards and provide design suggestions to mitigate
those hazards. Recently, Hadikusumo and Rowlinson
(2004) applied a visual reality concept to develop a
design for safety tool to capture tacit knowledge of
safety professionals. The results contribute to the
current arsenal of CEM and safety tools by providing
an applied DSS that integrates safety risk data into
Point of departure
To build upon safety risk management research, the
relative safety risk of common highway reconstruction
tasks was assessed and the efficacy of a DSS that inte-
grates safety risk data into project schedules was
tested. A thorough review of relevant literature
revealed no study that has directly quantified highway
reconstruction safety risks or attempted to assess tem-
poral models for safety risk integration using actual
data. It is expected that the findings presented will
aid project managers in their preconstruction safety
management activities and will be especially effective
for safety managers who are responsible for multiple
The research objectives were achieved in three distinct
phases. In the first phase, the Delphi method was
employed to quantify relative safety risks. In the
second phase, a graphical user interface was developed
in MATLAB, called Scheduled-based Safety Risk
Assessment and Management (SSRAM), that is capa-
ble of creating temporal safety risk profiles for highway
construction projects. Finally, the output of the system
was validated by employing a Multi-Attribute Utility
Assessment (MAUA) technique and conducting 11
case studies. The following sections discuss the details
of the research methods employed in these three
Phase I method: risk quantification
In order to develop an appropriate scope for data col-
lection, clear definitions of common highway con-
struction work tasks were needed. Therefore, the 25
highway tasks identified and described by Pandey
530 Esmaeili and Hallowell
(2009) and refined by Hallowell et al. (2011) were
used as a foundation. To quantify the relative risk val-
ues for these tasks, the Delphi method was selected.
The traditional paradigm in risk quantification
adopted by Brauer (1994) and Hallowell and
Gambatese (2009) was followed where frequency and
severity ratings for each task are solicited from an
expert panel through multiple rounds of surveys and
The Delphi method was chosen for obtaining
safety risk values for six main reasons. First, there
were no objective highway repair and maintenance
safety risk data available from government databases.
The common national databases such as Occupa-
tional Safety and Health Administration Integrated
Management Information System (OSHA IMIS,
n.d.) and National Institute of Occupational Safety
and Health Fatality Assessment and Control Reports
(NIOSH FACE, n.d.) include only high severity
injuries and do not provide enough information
regarding the task performed when the injury
occurred. Second, according to Gyi et al. (1999), the
validity of statistical data obtained from accident
reports is significantly compromised by underreport-
ing, especially for minor injuries. Third, Snashall
(1990) stated that accident report processes are not
consistent between and within companies
(e.g. definition of construction activities) such that
empirical data cannot be easily interpreted and com-
pared. Fourth, accidents happen in a complex sys-
tem created by interrelated worksite characteristics
that cannot be separated from the project context
(Mitropoulos et al., 2005). Fifth, according to Dijk-
sterhuis et al. (2006), intuitive decision processes like
Delphi that use heuristic principles lead to accurate
risk estimates in complex scenarios. Finally, Delphi
is a rigorous process that allows researchers to obtain
unbiased data using the judgment of qualified
experts, which has been used successfully for risk
quantification in similar studies (e.g. Hallowell and
Gambatese, 2009; Hallowell et al., 2011).
The Delphi method was developed by Rand
Corporation for the US Air Force in late 1940s to
elicit reliable and unbiased judgments from a group of
experts by conducting an iterative process and
providing controlled feedback (Helmer, 1967; Lin-
stone and Turoff, 1975). The Delphi method involves
assembling qualified experts, developing appropriate
questionnaires, and conducting multiple rounds of
surveys with controlled feedback between rounds to
achieve consensus (Cabaniss, 2001; Hallowell and
Gambatese, 2010). This method is applied under the
assumption that the collective expertise of the panel is
superior to the judgment of individuals (Hogarth,
1978; Boje and Murnighan, 1982; Hill, 1982).
The Delphi process was conducted in two rounds
where expert panellists were asked to provide inde-
pendent frequency and severity ratings for each of the
25 highway construction tasks. In order to maintain
consistency, the authors have adopted an objective
risk scale created by Hallowell and Gambatese (2009)
that incorporates a complete spectrum of frequency
and severity scales (see Table 1). The severity scale
ranges from negligible injury to fatality and the fre-
quency scale ranges from one incident occurrence
every six minutes (0.1 w-h) to one incident occur-
rence every 100 million or more worker-hours (>100
million w-h). After the first round of surveys, the data
were aggregated and the level of consensus was mea-
sured and evaluated. In the second round, panellists
were asked to review the median responses from the
first round and provide final ratings. As will be dis-
cussed, a third round of data collection was not
needed because the target consensus was achieved in
the second round.
Selection of expert panellists
As the number of panellists in a Delphi study
increases, the accuracy of the results also tends to
increase (Murphy et al., 1998). In a review of past
Delphi studies, Rowe and Wright (1999) found that
the number of panellists has ranged from 3 to 80. As
noted by Linstone and Turoff (1975) factors such as
the expected volume of the data, time constraints,
and the number of experts available can affect the
appropriate number of panellists. A relatively large
panel was desired to quantify risks for the 25 highway
construction tasks that can be performed in a variety
of work environments.
Careful attention was paid to ensure that all
panellists were highly qualified. The expert panel was
Table 1 Frequency and severity scales (adapted from Hallowell and Gambatese, 2009)
Worker hours per incident Subjective level Score
>100 million Temporary discomfort 2
10–100 million Persistent discomfort 4
1–10 million Temporary pain 8
100 000–1 million Persistent pain 16
10 000–100 000 Minor first aid 32
1000–10 000 Major first aid 64
100–1000 Medical case 128
10–100 Lost work time 256
0.1–1 Fatality 26 214
Safety risk data 531
assembled using the 165 contacts provided
by the National Work Zone Safety Information Clear-
tacts/browse/all_experts). Because this website provides
no information regarding the qualification of any of the
contacts as ‘experts’, the research team independently
validated expert status with an introductory survey
using guidance provided by Hallowell and Gambatese
(2010). Of the 165 individuals contacted, 75 (45%)
responded, and 27 (36%) were qualified as experts.
According to Moser and Kalton (1971), this response
rate is acceptable for Delphi studies.
The resulting pool of individuals averaged over 25
years of highway construction safety experience. Over
80% of respondents had a Professional Engineering
(PE) licence, were a Certified Safety Professional
(CSP), or had at least a bachelor’s degree in a related
field and all respondents were upper-level managers
or executives (e.g. corporate safety manager, director
of research, and senior project manager). It should be
noted that, despite the relative large publication lists
of some participants, the panel was largely profes-
sional in nature. This was preferred as accurately
quantifying relative risks relies upon a wealth of
Number of iterations and feedback
One of the objectives of the Delphi process is to reach
consensus, which can be achieved by conducting mul-
tiple iterations of questionnaires and providing anony-
mous feedback between rounds. Two to seven rounds
have been used in the previous large-scale Delphi stud-
ies (Dalkey et al., 1970). According to Jolson and
Rossow (1971), iterations can be terminated when the
changes in variance are no longer significant. The
research team administered two rounds of surveys
because the size of the expert panel (27) exceeded the
minimum size recommended (eight) for traditional
Delphi studies (Brockhoff, 1975; Boje and Murnighan,
1982) and there was a high degree of consensus
among the experts after the second round.
In order to decrease the complexity of probability
assessment, many individuals use a limited number of
heuristic controls (Tversky and Kahneman, 1974).
However, relying on these heuristics may produce sys-
tematic errors in judgment known as cognitive biases.
Despite their importance, cognitive biases have not
received adequate attention in previous Delphi studies
(Hallowell and Gambatese, 2010). The following
eight biases were identified and controlled: collec-
tive unconscious, contrast effect, neglect of probabil-
ity, Von Restorff effect, myside bias, recency effect,
primacy effect and dominance (Hallowell and
To minimize the potential influence of cognitive
biases, several controls were implemented. First,
respondents were kept anonymous. Maintaining the
anonymity of respondents reduces the impact of
group dynamics, dominant personalities and the
bandwagon effect (Manoliadis et al., 2006). Second,
randomizing the question order of the surveys mini-
mizes the potential influence of primacy and contrast
effects (Hallowell and Gambatese, 2010). Third, the
median ratings from the previous rounds were
provided as feedback, which significantly reduces vari-
ability among panellists (Martino, 1970). Fourth,
experts were asked to rate frequency and severity lev-
els separately to avoid the neglect of probability bias.
Finally, to ensure internal validity and to enhance the
reliability of the results, all experts were provided with
consistent task names and descriptions.
Phase II method: decision support system
One of the structured design methods to develop a
decision support system (DSS) is prototyping (Andr-
iole, 1989). The prototyping principles established
by Boar (1984) were followed where the develop-
ment of a DSS involves input from perspective users
and is refined with professional feedback in an itera-
tive process. Following the guidance provided by
Andriole (1989), the first step in designing the DSS
involved identifying the tasks that the system must
perform and the requirements of the user. According
to Boar (1984), 20 to 40% of DSS’s problems can
be attributed to the design process. Well-defined
requirements will make a link between users, tasks
and organizational needs (Andriole, 1989). Here, a
quick prototype was made and its features were
modified by receiving feedback from the users in an
iterative process. In the second step of the DSS
development, the safety risk data were mathemati-
cally integrated with activity sequences. The data
from Hallowell et al. (2011) and those established
through the Delphi process were used to populate
the theoretical model shown in Equation 1. In the
subsequent phase, the research team tested this
model with active construction projects in the US.
½SFTask�1�n ¼ ½RIndividual�1�25 � ð½RInteraction�25�25 � ½XSchedule�25�nÞ ð1Þ
[RIndividual] is a matrix that includes safety risk values
for individual tasks; [RInteraction] is a matrix that includes
the safety risk interactions among tasks from Hallowell
532 Esmaeili and Hallowell
et al. (2011); [XSchedule] is a matrix that includes 0’s and
1’s depending on whether or not particular activities
are scheduled for a given time period. If in time t,
activity i is being performed, then Xit=1, otherwise Xit = 0. [SFTask] is the resulting safety risk matrix that
includes the resulting risk for each time period.
Phase III method: risk data and DSS validation
One of the methods that has been used extensively to
decompose the general measure of effectiveness of a
DSS is Multi-Attribute Utility Assessment (MAUA)
(Adelman and Donnell, 1986; Sage, 1991). MUAU is
a formal structure that maps different measures of
effectiveness against one another and is defined as,
‘scoring and weighting procedures to evaluate the
overall utility of a knowledge-based system to users
and sponsors’ (Adelman and Riedel, 1997, p. 37).
This method has been used to evaluate similar DSSs
in several studies in the past (e.g. Adelman and
Ulvila, 1991). The total measure of effectiveness is
the weighted sum of all the utility scores, shown as
Equation 2 (Adelman, 1992):
UðiÞ ¼ W1: uðxi1Þ þ W2: uðxi2Þ þ � � � þ WJ: uðxijÞ ð2Þ
where: U(i) is the overall utility for alternative i; wj is
the cumulative relative weight on attribute j; u(xij) is
the utility scale value for alternative i on attribute j.
Figure 1 presents the hierarchy of effectiveness
criteria that was created from existing literature and
discussions with potential users. The three main
evaluation criteria were: usability, applicability and
reliability. Usability was defined as the system’s ease
of use, response time, ease of training and graphic
displays; applicability was defined as the extent that
the program and its output can be used by a
construction firm to enhance decision making and
resource allocation; and reliability was defined as the
predictive accuracy of the system. It is notable that
predictive accuracy of the framework and developed
DSS relies heavily on the reliability of the safety risk
database. In other words, the reliability scores
obtained from the MAUA process are an indicator of
the validity of quantified safety risks.
In order to determine the total utility of the system,
the research team used a case study approach where
relative weights of the criteria and scores for the sys-
tem were obtained through interviews with prospec-
tive users. Case studies were chosen because the
sample size and randomization requirements of a true
experiment were not feasible (Adelman, 1992) and
case studies are appropriate for studying new strate-
gies in context (Yin, 2003). The main units of analy-
sis were active or recently completed projects.
To obtain a representative sample of US highway
construction projects, highway construction firms that
were members of the Associated General Contractors
(AGC) or the Colorado Asphalt Pavement Associa-
tion were asked to participate. Of the 39 contractors
that were contacted, a total of five firms agreed to
provide project data and participate in a series of
interviews. The revenue of the companies ranged
from $50 million to $2.5 billion with the average of
$583 million. The companies, on average, had more
than 700 workers and had been in the highway con-
struction business for over 50 years.
Extent of use
Usefulness of output
Impact on the current procedures
General ease of use
Ease of training
Ease of data entry
Figure 1 Hierarchy of measures of effectiveness
Safety risk data 533
According to Yin (2003), the number of cases com-
pleted and the quality of pattern matching have signifi-
cant impact on the validity and reliability of the
results. Literature suggests that four to 10 cases will
provide valid and reliable data as long as pattern
matching is strong and data are collected consistently
among cases (Eisenhardt, 1989; Yin, 2003). To ensure
adequate data, a total of 11 case studies were con-
ducted. The demographics of these cases are summa-
rized in Table 2. As shown in Table 2, a diverse set of
projects is included ranging from large scope and long
duration to small scope and short duration. Also, a
higher number of projects were located in Colorado,
which limits the external validity of the results.
In order to increase the reliability and internal
validity of the study, a specific case study protocol
was implemented. The following four steps were
conducted for every case study:
(1) Interviews were conducted with the construc-
tion project manager or safety managers to
quantify the relative weights of the attributes by
conducting pairwise comparisons between crite-
ria. The interviewees were asked to use a pro-
vided comparison scale that was based on
previously successful studies described by Saaty
(1980). A consistency ratio was then used to
ensure that each respondent’s ratings were inter-
nally consistent. As suggested by Shapira and
Goldenberg (2005), participants were asked to
repeat the rating process if their internal consis-
tency ratio exceeded 0.1. In other words, if an
individual’s pairwise comparisons among crite-
ria resulted in 10% or greater internal inconsis-
tency, they were asked to repeat the process
until their ratings were in agreement. An
acceptable internal consistency ratio indicates
that there is no intolerable conflict in the com-
parisons of a participant’s response (Shapira and
(2) After the weights had been found, interviews
were conducted to determine the DSS’s scores
for different criteria. In order to gather opin-
ions about the usability and applicability of the
DSS, the operation of the system was demon-
strated to the participants. Immediately follow-
ing the demonstration, the users were asked to
complete an 18-question survey (two questions
for each attribute) that addressed all criteria
shown in Figure 1. The participants were asked
to rate the system’s performance on a scale
from 0% (very poor performance) to 100%
(very strongly performance), with 50% being
(3) The project schedule was then obtained and
the project manager was interviewed to ensure
that the research team had an accurate under-
standing of the actual activities that were per-
formed on the project. With the project
manager, the tasks and durations were matched
with the tasks described in Hallowell et al.
(2011). This mapping process was required
because the DSS was built around the data
from previous research and consistency of task
names was required for the system to operate
effectively. Once the tasks were mapped, the
schedule integration function of the DSS was
used to produce a safety risk profile. The con-
struction project manager or safety manager
was then asked to compare the risk profile
Table 2 Projects demographic information
1 150 50 88 Bid/Build Unit price 2 California
2 110 30 65 Design/
Lump sum 0 Utah
3 48 36 94 Bid/Build Monthly
4 5.5 4 100 – Pay
5 4.5 6 100 Bid/Build Unit price 0 Washington
6 0.32 1.5 99 Bid/Build Unit price 0 Colorado
7 0.38 3 90 Bid/Build – 0 Colorado
8 0.66 5 100 – – 0 Colorado
9 0.37 10 100 Bid/Build Unit price 0 Colorado
10 0.49 1.5 100 Bid/Build – 0 Colorado
11 1.5 3 100 – Pay
534 Esmaeili and Hallowell
created by the DSS with the actual level of risk
and provide an approximate percentage agree-
ment with the system. The interviewees aimed
to compare the pattern of the risk profile with
near misses and the actual hazards that existed
during the work. In fact, the current study did
not aim to predict injuries in the jobsite.
Rather, the focus was on predicting high risk
work periods where the potential for injury is
relatively high. It is important to distinguish
the difference between hazards and accidents.
For example, if a worker was exposed to adja-
cent traffic, there were significant hazards even
though no injury was realized.
(4) The final step of the case study involved a fol-
low-up questionnaire that included open-ended
questions that gave the participants an oppor-
tunity to share their thoughts on the perceived
strengths and weaknesses of the system.
Phase I results: risk quantification
All 27 expert panellists provided complete responses
to the Delphi surveys in the first round and the abso-
lute variance of responses for frequency and severity
were 0.733 and 0.838, respectively. Once the data
were aggregated and summarized for the panel, sec-
ond rounds of surveys were administered. In the sec-
ond round, the median responses from the previous
round were provided to the panellists and they were
given the option to agree with the other group’s col-
lective assessment or provide a new rating. Of the 27
surveys that were sent in the second round, 24 sur-
veys were returned resulting in 89% response rate.
Absolute variances of responses in the second round
were 0.198 and 0.191 for frequency and severity,
respectively. Because the established consensus was
achieved in the second round, there was no need for
a third round. Additionally, the median ratings did
not change between rounds, which is evidence of
strong internal validity.
To facilitate calculations, the frequency ratings
were converted from a range of values with units of
worker-hours per incident to a single point value with
units of incidents per worker-hour. The mean value
was selected as a point value and inverted to obtain a
number with appropriate units. For example, if the
Delphi panel rated the average frequency as 10–100
w-h/incident, the mean value, 55 w-h/ incident, was
inverted (0.018 incidents/w-h) to determine the fre-
quency value for that particular risk and activity.
Severity values were not changed from the severity
scale in Table 1.
The frequency ratings ranged from 1.8E-8 to 0.018
incidents per worker-hour and the severity ratings
ranged from 4 to 256 units on the severity scale. Unit
risk scores were calculated by multiplying the average
frequency scores by the average severity scores. The
resulting data for the 25 work tasks are provided in
descending order of relative risk in Table 3. In this
table, risk is described in terms of units of severity per
worker-hour (S/w-h). The task ‘construction zone
traffic control’ has the highest unit risk (0.047 S/w-h)
while ‘watering and dust palliatives’ (1.8 � 10-8 S/w-h) and ‘install field facilities’ (1.8 � 10-8 S/w-h) have the lowest unit risk.
Phase II results: DSS development
In order to provide a user friendly environment to
integrate safety risk data and project schedules, a
graphical user interface (GUI) was developed in
MATLAB called Scheduled-based Safety Risk Assess-
ment and Management (SSRAM). MATLAB was
chosen for two main reasons: it is a strong program-
ming language to develop graphical user interfaces
and it allows the research team to make an active con-
nection between standard project management soft-
ware and safety risk databases. An applied DSS for
integrating safety risk data into project schedules
should include two main capabilities: (1) receiving the
schedule from the user; and (2) creating the safety
risk profile. These capabilities were considered during
the development. Although the DSS can be used to
manually add tasks, start dates and end dates to build
the schedule, the researchers built a bridge between
Primavera 6, MS Project and the DSS using MS
Excel as medium to increase efficiency. After entering
projects to the program, the user can save the sched-
ules as M-file ( ⁄ .m or
⁄ .matt). The resulting DSS
(SSRAM) is a knowledge-based system with a
schedule integration engine.
The conceptual formulation and computational
process of the SSRAM is shown in Figure 2. The
safety risk database includes the base-level safety risk
(obtained from Delphi panel in the first phase) and
the safety risk interactions (a 25 � 25 matrix from Hallowell et al., 2011). The user can insert the
schedule manually or import it from scheduling
software (e.g. Primavera 6). Once the schedule is
entered, the SSRAM loads safety risk data from the
database to the imported schedule using Equation 1
and subsequently plots the risk profile. There are sev-
eral practical applications of the safety risk profiles.
For example, the risk profiles can be used to identify
high risk periods during the project, the safety risk
can be levelled utilizing float of activities, or the pro-
ject manager can allocate safety resources according
Safety risk data 535
to the risk profile. In addition, the program is able to
create safety risk profiles for multiple projects or port-
folios of a company. This is important because safety
managers for highway construction companies must
often manage multiple concurrent projects. Using the
SSRAM helps them to strategically allocate their time
and safety resources.
Phase III results: risk data and DSS validation
The results of pairwise comparisons made by the
user group are shown in Table 4. The users
believed that the reliability, the general ease of use
of the program, and the usefulness of the output
are the most important attributes. Because the
usability and applicability have a subset of
Table 3 Safety risk data for common highway reconstruction work tasks
Task name Frequency score (incident/w-h ⁄ E-5) Severity score Unit risk scores (S/w-h
Construction zone traffic control 1800 256 4700
Install traffic control devices 18 64 4700
Installing flexible pavement/patching 18 64 12
Pavement marking 18 64 12
Seal joints and cracks 18 64 12
Excavation 18 64 12
Install culverts, drains, sewers 18 64 12
Install culvert pipe and water lines 18 64 12
Reset structures 18 64 12
Heat and scarifying 18 64 12
Survey 1.8 32 5.8
Clear and grub 0.18 16 2.9
Recycle cold bituminous pavement 0.18 16 2.9
Install curb and gutters 0.18 16 2.9
Install rigid pavement (concrete) 0.18 16 2.9
Install cribbing 0.18 16 2.9
Recondition bases (compaction) 0.18 16 2.9
Install water control devices 0.18 16 2.9
Lay aggregate base course 0.18 16 2.9
Mobilization/demobilization 0.18 16 2.9
Prime, coat, rejuvenate pavement 0.18 16 2.9
Demolition of existing pavement 0.18 16 2.9
Landscape 0.18 16 0.029
Install field facilities 0.0018 4 0.0073
Watering and dust palliatives 0.0018 4 0.0073
] 1×n =[RIndividual]1×25 ×([RInteraction]25×25 ×[XSchedule]25×n)
1. Safety risk profiles
1. Inserting manually
2. Importing from scheduling software
2. Risk threshold
Safety risk database
1. Base level safety risks (Table 2)
2. Safety risk interactions (Hallowell et al., 2011)
Figure 2 SSRAM’S framework
536 Esmaeili and Hallowell
attributes, the relative weights for the higher tier
attributes were computed by finding the products of
the subsets (see Table 4). For example, the relative
weight of workload (0.17) was multiplied by the rel-
ative weight assigned to usability (0.15) to reach
the total ‘workload’ weight of (0.03). Once the
weights of the criteria were found, they were multi-
plied against the scores and the resulting products
were summed to compute the global utility factor
(0.67). This number can be interpreted as the value
that the SSRAM adds to the current safety manage-
ment practice, which ranges from a score of 1 that
corresponds to a revolutionary product that
completely changes current industry practice and is
perfectly executed to a score of 0 where no value is
added. One should note that the interviewees, who
rated the output of the program, were not informed
of the analytical procedure that resulted in the out-
put. Therefore, any inconsistencies among the inter-
viewees’ opinions and the Delphi experts’ judgment
decrease the actual utility of the system as a whole.
In fact, the proposed validation methodology tested
the ability of the SSRAM to forecast hazardous
conditions. Considering the complex and dynamic
nature of the construction projects, reaching 100%
accuracy was not realistic.
Although the research team was satisfied with the
results, there are no similar DSS validation studies
to compare against. Fortunately, the follow-up inter-
view questions validated the SSRAM global utility
score because respondents indicated that the pro-
gram is easy to use and greatly improves precon-
struction safety management at the project and
Though the results have the potential to impact posi-
tively on preconstruction safety management, there
are several limitations of the research.
(1) The risk quantification portion of the study
required that the Delphi panel assume typical
conditions in their ratings. Consequently, the
data are limited by the fact that actual conditions
such as weather, crew safety culture and fatigue
affect the true risk values (Manu et al., 2010).
Because there are large numbers of external risk
factors that can affect the base-level risk of activ-
ities, considering their effects was unrealistic.
The influence of external factors may explain
some of the variation between the predicted
values and actual values during the case studies.
(2) The risk values were estimated as a single point
estimate for each task. The average risk values
may not capture all characteristics of risk or, as
Kaplan and Garrick (1981) stated, a single
number cannot communicate risk effectively
due to the great loss of information.
(3) Although several measures were employed to
decrease cognitive biases in the Delphi process,
there are still several limitations to the frequency
and severity values provided by experts. One of
the common limitations is related to accidents
with low probability of occurrence and high
impacts. Taleb (2007), one of the prominent
researchers in this area, called these extreme
events, ‘Black Swans’. He stated that it is almost
impossible to predict extreme events because
they do not have predecessor events (Taleb,
Table 4 Weights of measures of effectiveness and their consistency ratio
Measures of effectiveness Relative weights Total weights Scores Utility
Usability 0.15 – – –
Applicability 0.25 – – –
Reliability 0.60 0.60 0.66 0.40
Usability General ease of use 0.31 0.05 0.69 0.03
Ease of training 0.26 0.04 0.75 0.03
Ease of data entry 0.17 0.03 0.66 0.02
Workload 0.17 0.03 0.60 0.02
Graphical features 0.07 0.01 0.68 0.01
Applicability Extent of use 0.23 0.06 0.58 0.03
Usefulness of output 0.36 0.09 0.73 0.06
Impact on the current procedures 0.29 0.07 0.61 0.04
Performance 0.13 0.03 0.70 0.02
Safety risk data 537
2007). In fact, predicting these low-probability,
high-impact events is extremely difficult and
more attention should be paid to reduce the vul-
nerability of the system towards their conse-
quences than anticipating them (Taleb, 2004).
(4) The safety risks were quantified for only 25
tasks. In order to add a new task to the schedule,
its base-level safety risk and the interactions with
other tasks must be quantified separately.
(5) The external validity is limited because the
data and DSS were validated on projects in
Colorado, Oregon, California, Utah and Wash-
ington, with a higher number in Colorado.
Although it is expected that projects are repre-
sentative of the US, the scope of inference is
theoretically limited only to these states.
Despite these limitations, the resulting data and
SSRAM significantly furthered knowledge and were
accurate and useful enough to gain favourable
responses from industry users.
Conclusions and recommendations
According to Esmaeili and Hallowell (2012a), the
construction industry is saturated with respect to
traditional injury prevention strategies and new safety
innovations are needed. Previous research has estab-
lished that the potential to prevent construction inju-
ries is at its highest during the preconstruction phase
and decreases exponentially as a project progresses
(Gambatese et al., 1997; Szymberski, 1997). Tradi-
tionally, preconstruction safety improvement tech-
niques such as designing for safety have faced
significant barriers that stem from the fact that they
are largely designer-controlled (Hinze and Wiegand,
1992). One of the contractor-controlled practices that
can be used to overcome this shortcoming is safety-
Although there have been attempts to integrate
safety risk data into project schedules (e.g. Yi and
Langford, 2006; Sacks et al., 2009), these attempts
were not successful because of the absence of a valid
and reliable safety risk database. This limitation was
addressed by quantifying the relative risk of 25 com-
mon highway construction tasks and combining this
information with risk interaction data obtained by
Hallowell et al. (2011) to populate a new risk integra-
tion framework. To facilitate implementation, validate
the risk data, and test the utility of the underlying
framework, the research team created a safety DSS
(SSRAM) using MATLAB. The resulting tool inter-
faced with Primavera P6 to produce plots of safety risk
over time for single or multiple concurrent projects.
The tool was then tested on 11 case study projects.
This validation effort revealed that the data are valid
and reliable despite recognized limitations and the
SSRAM has the potential to improve safety resource
optimization and preconstruction safety management.
To summarize, the authors tested the validity of the
risk database as an input and reliability of the risk pro-
files generated by integrating base-level risk data and
risk interaction data with highway project schedules.
The traditional safety management approach
involves investing safety resources such as time and
money at a uniform rate throughout the lifespan of a
project (Rozenfeld et al., 2009). However, because
physical conditions change rapidly, safety risk levels
may also fluctuate. Consequently, uniform resource
allocation to safety within a project and among
projects may not be the optimum strategy. One of the
viable solutions for this problem is to apply lean
thinking to the construction process (Womack and
Jones, 2003) so that injury prevention practices can
be treated as production control activities (Rozenfeld
et al., 2010). According to Rozenfeld et al. (2010),
the ability to predict fluctuating safety risk levels is
essential to practical lean-based safety management.
The findings can be used to predict safety risk levels
and use schedule float to distribute or concentrate
risk. In addition, risk profiles enable thoughtful safety
planning and effective allocation of safety resources in
a single project or multiple projects.
In addition, the data and framework presented can
be used by project managers to enhance preconstruc-
tion safety management by identifying high risk peri-
ods. In response, safety managers can plan for extra
precautionary measures during these high risk periods
(e.g. lane closure), develop customized injury preven-
tion strategies, or at a minimum, inform workers of
the tasks and interactions known to cause high risk
periods. In addition to using the schedule-based tech-
nique described, the writers also recommend that
practitioners focus attention on high risk work tasks
(e.g. construction zone traffic control, installing traffic
control devices and excavation).
To address the aforementioned study limitations
the writers suggest three complementary research
efforts. First, new safety risk quantification methods
should be explored to produce robust and reliable risk
data independently from specific tasks, trades and
construction objects. For example, Esmaeili and
Hallowell (2011, 2012b) utilized genome concept to
quantify safety risks at attribute level independently
from tasks and objects. Population of this method can
be a major step towards a universal safety risk
assessment in the construction industry. Second,
researchers should consider modelling probability dis-
tributions of accident occurrence for individual tasks.
538 Esmaeili and Hallowell
Using probability distributions instead of the average
estimated points allows an individual to consider
uncertainty in the data and investigate its propagation
thorough the model (Fischhoff et al., 1984). Finally,
the relative impacts of environmental risk factors on
the base safety risk values must be better understood
to create robust models. The external risk factors may
include the intensification of risk due to night-time
work, exposure to weather, and adjacent traffic.
The writers would like to thank Bentley Systems for
the resources and high quality feedback during the
project and all of the Delphi panellists for their
Adelman, L. (1992) Evaluating Decision Support and Export
Systems, John Wiley & Sons, New York.
Adelman, L. and Donnell, M.L. (1986) Evaluating decision
support systems: a general framework and case study, in
Andriole, S.J. (ed.) Microcomputer Decision Support Systems,
QED Information Systems, Wellesley, MA, pp. 285–309.
Adelman, L. and Riedel, S. (1997) Handbook for Evaluating
Knowledge-Based Systems, Kluwer Academic, Boston.
Adelman, L. and Ulvila, J.W. (1991) Evaluating expert sys-
tem technology, in Andriole, S.J. and Halpin, S.M. (eds)
Information Technology for Command and Control, IEEE
Press, New York, pp. 537–47.
Andriole, S.J. (1989) Handbook of Decision Support Systems,
TAB Books Inc, Blue Ridge Summit, PA.
Arditi, D., Ayrancioglu, M. and Shi, J. (2005) Worker
safety issues in night-time highway construction. Engineer-
ing Construction and Architectural Management, 12(5),
Boar, B. (1984) Application Prototyping: A Requirements Defi-
nition Strategy for the 80s, Wiley, New York.
Boje, D.M. and Murnighan, J.K. (1982) Group confidence
pressures in iterative decisions. Management Science, 28
Brauer, R.L. (1994) Risk management and assessment, in
Brauer, R.L. Safety and Health for Engineers, Van Nos-
trand Reinhold, New York, pp. 645–64.
Brockhoff, K. (1975) The performance of forecasting
groups in computer dialogue and face-to-face discussion,
in Linstone, H.A. and Turloff, M. (eds) The Delphi
Method: Techniques and Applications, Addison-Wesley,
Reading, MA, pp. 291–321.
Bryden, J.E. and Andrew, L. (1999) Serious and fatal inju-
ries to workers on highway construction projects. Trans-
portation Research Record, 1657, 42–7.
Bureau of Labor Statistics (2012) Occupational Injuries/Ill-
nesses and Fatal Injuries Profiles, available at http://goo.gl/
CNJb3 (accessed January 2012).
Cabaniss, K. (2001) Counseling and computer technology in
the new millennium: an Internet Delphi study, PhD thesis,
Virginia Polytechnic Institute and State University, Virginia.
Cagno, E., Giulio, A.D. and Trucco, A. (2001) Algorithm
for the implementation of safety improvement programs.
Safety Science, 37, 59–75.
Carter, G.M., Murray, M.P., Walker, R.G. and Walker, W.
E. (1992) Building Organizational Decision Support Sys-
tems, Academic Press, Book News, Inc., Portland, OR.
Center for Construction Research and Training (2008) Con-
struction Chartbook, Center for Construction Research and
Training, Silver Spring, MD.
Coble, R.J. and Elliott, R.B. (2000) Scheduling for
construction safety, in Coble, R.J., Hinze, J. and Haupt,
T.C. (eds) Construction Safety and Health Management,
Prentice Hall, Englewood Cliffs, NJ, pp. 43–57.
Dalkey, N., Brown, B. and Cochran, S. (1970) Use of self-
ratings to improve group estimates. Technological Forecasting,
Dijksterhuis, A., Bos, M.W., Nordgren, L.F. and von Baa-
ren, R.B. (2006) On making the right choice. the deliber-
ation-without-attention effect. Science, 311, 1005–7.
Eisenhardt, K.M. (1989) Building theories from case study
research. Academy of Management Review, 14(4), 532–50.
Esmaeili, B. and Hallowell, M.R. (2011) Using network
analysis to model fall hazards on construction projects, in
Tjandra, K. (ed.) Proceedings of Safety and Health in Con-
struction, CIB W099, Washington, DC, 24–26 August.
Esmaeili, B. and Hallowell, M. (2012a) Diffusion of safety
innovations in the construction industry. Journal of Con-
struction Engineering and Management, 138(8), 955–63.
Esmaeili, B. and Hallowell, M.R. (2012b) Attribute safety
risk model for measuring safety risk of struck-by acci-
dents, in Cai, H., Kandil, A., Hastak, M. and Dunston,
P.S. (eds) Proceedings of The 2012 Construction Research
Congress (CRC), West Lafayette, 21–23 May.
Federal Highway Administration (2004) Work zone safety
facts and statistics, updated on 24 December 2004, avail-
able at http://goo.gl/X7crE (accessed February 2012).
Fischhoff, B., Watson, S. and Hope, C. (1984) Defining
risk. Policy Sciences, 17, 123–39.
Gambatese, J.A., Hinze, J.W. and Haas, C.T. (1997) Tool
to design for construction worker safety. Journal of Archi-
tectural Engineering, 3(1), 32–41.
Gyi, D.E., Gibb, A.G.F. and Haslam, R.A. (1999) The
quality of accident and health data in the construction
industry: interviews with senior managers. Construction
Management and Economics, 17, 197–204.
Hadikusumo, B.H.W. and Rowlinson, S. (2004) Capturing
safety knowledge using Design-for-Safety-Process Tool.
ASCE Journal of Construction Engineering and Management,
Hallowell, M.R. (2008) A formal model of construction
safety and health risk management, unpublished PhD dis-
sertation, Oregon State University, Corvallis, OR.
Hallowell, M.R. and Gambatese, J.A. (2009) Activity-based
safety risk quantification for concrete formwork construction.
ASCE Journal of Construction Engineering and Management,
Safety risk data 539
Hallowell, M.R. and Gambatese, J.A. (2010) Qualitative
research: application of the Delphi method to CEM
research. ASCE Journal of Construction Engineering and
Management, 136(1), 99–107.
Hallowell, M.R., Esmaeili, B. and Chinowsky, P. (2011)
Safety risk interactions among highway construction work
tasks. Construction Management and Economics, 29(4),
Han, S.H. and Diekmann, J.E. (2001) Approaches for mak-
ing risk-based go/no-go decision for international projects.
ASCE Journal of Construction Engineering and Management,
Helmer, O. (1967) Analysis of the Future: The Delphi Method,
The Rand Corporation, Santa Monica, CA.
Hill, G.W. (1982) Group versus individual performance.
Psychological Bulletin, 91, 517–39.
Hinze, J., Nelson, J. and Evans, R. (2005) Software integra-
tion of safety in construction schedules, in Haupt, T.C.
and Smallwood, J. (eds) Proceedings of the 4th Triennial
International Conference, Rethinking and Revitalizing Con-
struction Safety, Health, Environment and Quality, Port
Elizabeth, South, Africa, 16–17 May, pp. 212–22.
Hinze, J. and Wiegand, F. (1992) Role of designers in con-
struction worker safety. ASCE Journal of Construction
Engineering and Management, 118(4), 677–84.
Hogarth, R.M. (1978) A note on aggregating opinions.
Organizational Behavior and Human Performance, 21,
Jolson, M.A. and Rossow, G. (1971) The Delphi process in
marketing decision making. Journal of Marketing Research,
Kak, A., Zacharias, K., Minkarah, I. and Pant, P. (1995) A
knowledge based software for construction safety, in
Mohsen, J.P. (ed.) Proceedings of the Second Congress on
Computing in Civil Engineering, Atlanta, GA, 5–8 June,
Kaplan, S. and Garrick, B.J. (1981) On the quantitative
definition of risk. Risk Analysis, 1(1), 11–28.
Kartam, N. (1997) Integrating safety and health perfor-
mance into construction CPM. ASCE Journal of Construc-
tion Engineering and Management, 123(2), 121–6.
Karumanasseri, G. and AbouRizk, S. (2002) Decision sup-
port system for scheduling steel fabrication projects.
ASCE Journal of Construction Engineering and Management,
Leu, S.S., Yang, C.H. and Huang, J.C. (2000) Resource
leveling in construction by genetic algorithm-based opti-
mization and its decision support system application.
Automation in Construction, 10(1), 27–41.
Lin, K. and Haas, C. (1996) Multiple heavy lifts optimiza-
tion. ASCE Journal of Construction Engineering and Man-
agement, 122(4), 354–62.
Linstone, H. and Turoff, M. (1975) The Delphi Method:
Techniques and Applications, Addison-Wesley, Reading,
Manoliadis, O., Tsolas, O. and Nakou, A. (2006) Sustain-
able construction and drivers of change in Greece. a Del-
phi study. Construction Management and Economics, 24(2),
Manu, P., Ankrah, N., Proverbs, D. and Suresh, S. (2010)
An approach for determining the extent of contribution
of construction project features to accident causation.
Safety Science, 48(6), 687–92.
Martino, J. (1970) The precision of Delphi estimates. Tech-
nological Forecasting, 1(3), 293–9.
Mitropoulos, P., Abdelhamid, T. and Howell, G. (2005)
Systems model of construction accident causation. ASCE
Journal of Construction Engineering and Management, 131
Molenaar, K.R. and Songer, A.D. (2001) Web-based deci-
sion support systems: case study in project delivery. ASCE
Journal of Computing in Civil Engineering, 15(4), 259–67.
Moser, C.A. and Kalton, G. (1971) Survey Methods in
Social Investigation, 2nd edn, Heinemann Educational,
Murphy, M.K., Black, N.A., Lamping, D.L., McKee, C.
M., Sanderson, C.F.B., Askham, J. and Marteau, T.
(1998) Consensus development methods and their use in
clinical guideline development. Health Technology Assess-
ment, 2(3), 1–88.
National Institute of Occupational Safety and Health, Fatal-
ity Assessment and Control Evaluation (FACE) program
(n.d.) NIOSH FACE reports indexed by industry or
cause of fatality, available at http://goo.gl/CYHuw
(accessed February 2012).
Navon, R. and Kolton, O. (2006) Model for automated
monitoring of fall hazards in building construction. Jour-
nal of Construction Engineering and Management, 132(7),
Navon, R. and Kolton, O. (2007) Algorithms for automated
monitoring and control of fall hazards. ASCE Journal of
Computing in Civil Engineering, 21(1), 21–8.
Occupational Safety and Health Administration Integrated
Management Information System (OSHA IMIS) (n.d.)
Statistics and Data, Standard Industry Classification
(SIC) system search, available at http://goo.gl/iC2dV
(accessed February 2012).
Palaneeswaran, E. and Kumaraswamy, M.M. (2008) An
integrated decision support system for dealing with time
extension claims in construction projects. Automation in
Construction, 17(4), 425–38.
Pandey, S. (2009) Risk quantification for highway recon-
struction activities, unpublished Masters thesis, University
of Colorado, Boulder, CO.
Perera, S. (1983) Resource sharing in linear construction.
ASCE Journal of Construction Engineering and Management,
Rajendran, S. and Gambatese, J.A. (2009) Development
and initial validation of sustainable construction safety
and health rating system. ASCE Journal of Construction
Engineering and Management, 135(10), 1067–75.
Rowe, G. and Wright, G. (1999) The Delphi technique as
a forecasting tool: issues and analysis. International Journal
of Forecasting, 15, 353–75.
Rozenfeld, O., Sacks, R. and Rosenfeld, Y. (2009)
‘CHASTE’: construction hazard assessment with spatial
and temporal exposure. Construction Management and Eco-
nomics, 27(7), 625–38.
540 Esmaeili and Hallowell
Rozenfeld, O., Sacks, R., Rosenfeld, Y. and Baum, H.
(2010) Construction job safety analysis. Safety Science,
Russell, J.S., Skibniewski, M.J. and Cozier, D.R. (1990)
QUALIFIER-2: knowledge based system for contractor
prequalification. ASCE Journal of Construction Engineering
and Management, 116(1), 157–71.
Saaty, T.L. (1980) The Analytic Hierarchy Process, McGraw-
Sacks, R., Rozenfeld, O. and Rosenfeld, Y. (2009) Spatial
and temporal exposure to safety hazards in construction.
ASCE Journal of Construction Engineering and Management,
Sage, A.P. (1991) Decision Support Systems Engineering,
Wiley, New York.
Saurin, T.A., Formoso, C.T. and Guimaraes, L.B.M.
(2004) Safety and production: an integrated planning and
control model. Construction Management and Economics,
Sawacha, E., Naoum, S. and Fong, D. (1999) Factors
affecting safety performance on construction sites. Inter-
national Journal of Project Management, 17(5), 309–15.
Shapira, A. and Goldenberg, M. (2005) AHP-based equip-
ment selection model for construction projects. Journal of
Construction Engineering and Management, 131(12),
Snashall, D. (1990) Safety and health in the construction
industry. British Medical Journal, 301, 563–4.
Szymberski, R. (1997) Construction project safety planning.
TAPPI Journal, 80(11), 69–74.
Taleb, N.N. (2004) Fooled by Randomness: The Hidden Role
of Chance in Life and in the Markets, 2nd edn, Random
House, New York.
Taleb, N.N. (2007) The Black Swan: The Impact of the
Highly Improbable, Random House, New York.
Tarrants, W.E. (1980) The Measurement of Safety Perfor-
mance, Garland STPM Press, New York.
Tversky, A. and Kahneman, D. (1974) Judgment under
uncertainty: heuristics and biases. Science, New Series,
Wang, W.C., Liu, J.J. and Chou, S.C. (2006) Simulation-
based safety evaluation model integrated with network
schedule. Automation in Construction, 15, 341–54.
Womack, J.P. and Jones, D.T. (2003) Lean Thinking:
Banish Waste and Create Wealth in Your Corporation, Free
Press, New York.
Yi, K. and Langford, D. (2006) Scheduling-based risk esti-
mation and safety planning for construction projects.
ASCE Journal of Construction Engineering and Management,
Yin, R.K. (2003) Case Study Research: Design and Methods,
3rd edn, Sage, Thousand Oaks, CA.
Safety risk data 541
Copyright of Construction Management & Economics is the property of Routledge and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder’s express written permission. However, users may print, download, or email articles for individual use.
Safety risks interactions among highway construction work tasks.docx
HALLOWELL, M., ESMAEILI, B., & CHINOWSKY, P. (2011). Safety risk interactions among highway construction work tasks. Construction Management & Economics, 29(4), 417-429. doi:10.1080/01446193.2011.552512
Recent research has produced frameworks for integrating safety risk data into project schedules, visual models and otherconstruction planning tools. Unfortunately, only a few studies have attempted to quantify base-level safety risk forconstruction tasks and no study has attempted to quantify the degree to which spatial and temporal interactions among tasks contribute to the potential for injury. A research study was performed to quantify the impact that pair-wise spatial and temporal interactions have on the base-level risk of 25 common highway construction work tasks in the United States. Six hundred risk interactions were quantified by obtaining and aggregating over 23 500 individual ratings from certified experts using the Delphi method. The results indicate that incompatible tasks may increase the base-level risk up to 60%. The most incompatible highway construction tasks are: (1) installing curbs and gutters and installing rigid pavement; and (2)construction zone traffic control and installing rigid pavement. Additionally, watering and dust palliatives and pavement marking is the one compatible task pair and there are 45 neutral task pairs. The resulting database and analysis have the potential to increase the efficacy of existing frameworks for integration of safety risk data with project planning tools. [ABSTRACT FROM AUTHOR]
Safety risk interactions among highway construction work tasks.pdf
Construction Management and Economics
Construction Management and Economics
ISSN 0144-6193 print/ISSN 1466-433X online © 2011 Taylor & Francis http://www.informaworld.com
Safety risk interactions among highway construction work tasks
BEHZAD ESMAEILI and PAUL CHINOWSKY
Department of Civil, Environmental and Architectural Engineering, University of Colorado, 428 UCB, 1111 Engineering Drive, Boulder, 80303 USA
Taylor and Francis
Received 10 August 2010; accepted 21 December 2010
Recent research has produced frameworks for integrating safety risk data into project schedules, visual models and other construction planning tools. Unfortunately, only a few studies have attempted to quantify base-level safety risk for construction tasks and no study has attempted to quantify the degree to which spatial and temporal interactions among tasks contribute to the potential for injury. A research study was performed to quantify the impact that pair-wise spatial and temporal interactions have on the base-level risk of 25 common highway construction work tasks in the United States. Six hundred risk interactions were quantified by obtaining and aggregating over 23 500 individual ratings from certified experts using the Delphi method. The results indicate that incompatible tasks may increase the base-level risk up to 60%. The most incompatible highway construction tasks are: (1) installing curbs and gutters and installing rigid pavement; and (2) construction zone traffic control and installing rigid pavement. Additionally, watering and dust palliatives and pavement marking is the one compatible task pair and there are 45 neutral task pairs. The resulting database and analysis have the potential to increase the efficacy of existing frameworks for integration of safety risk data with project planning tools.
Project management, risk analysis, safety.
Over the last 40 years the construction industry has accounted for an injury and fatality rate that is nearly five times greater than the all-industry average (Bureau of Labor Statistics, 2010). Although injury rates have declined dramatically in this time, in each of the past 15 years the construction industry has accounted for over 1200 deaths and 460 000 disabling injuries in the United States (National Safety Council, 2009). In addition to physical pain and emotional suffering expe- rienced by the victims and their families, these inci- dents have substantial societal costs totalling an estimated $15.64 billion annually (National Safety Council, 2009). Furthermore, it has also been shown that injuries alone account for 7.9% to 15% of the costs of new construction (Everett and Frank, 1996). These costs cripple entrant firms and have a strong, negative impact on the gross domestic product (GDP).
Following the Occupational Safety and Health Act of 1970, numerous attempts have been made to improve understanding of construction safety. For example, Bernold and Guler (1993) identified common activities and physical motions that contribute to back injuries; Hinze
(1998) suggested a new classification method for identifying root causes of injuries; Chi
(2005) identified key contributing factors to fall inci- dents; Hinze
(2005a) studied the root causes of struck-by accidents; Sobeih
(2009) identified causes of musculoskeletal disorders; Lombardi
(2009) evaluated factors affecting workers’ perception of risk; and Mitropoulos and Guillama (2010) suggested a protocol to evaluate the potential for injury when constructing residential framing. Though the contributions of these previous studies are considerable, they are limited in application because they evaluate injuries, activities and preventive measures as individual issues and isolated subjects (Sacks
Author for correspondence. E-mail: email@example.com
Construction projects are characterized by complexity and uncertainty which stems from an ever- changing environment. The dynamic nature of construction projects requires safety measures to be adapted to new situations. Consequently, many experts believe that injury prevention activities should be conducted early in the project life cycle (Hinze, 1997). One emerging proactive safety management strategy is to integrate safety information into project schedules (Kartam, 1997; Chantawit
, 2005; Hinze
, 2005a). Recently, Yi and Langford (2006) and Sacks
(2009) developed techniques for ‘safety loading’ safety risk data into critical path method (CPM) schedules. According to Yi and Lang- ford (2006), the quantity of safety risk varies during the project schedule and limited resources should be allocated to projects in proportion to their safety risk at any given time. To analyse temporal safety risk, both studies concluded that safety risk data should be numerically integrated into the project schedule. Prior to these efforts, resource allocation for safety manage- ment was inefficient because resources (e.g. safety personnel) were assigned to projects for longer peri- ods than they were actually required for (Sacks
In order to effectively integrate safety risk data with project schedules, managers must identify and quan- tify safety risk for all scheduled tasks. Though the framework for schedule integration established by Yi and Langford (2006) only requires base-level risk data for the performance of individual tasks, in isolation, under typical circumstances, several authors have postulated that the actual risk of construction opera- tions also depends on the interactions that occur among tasks throughout space and time (Lee and Halpin, 2003; Sacks
, 2009; Rozenfeld
, 2010). These studies argue that interactions among incompatible tasks may contribute to a greater risk than the sum of the base-level task risks alone. Unfortunately, no study has quantified these potential interactions.
The objective of the present study was to quantify the impact that the interactions of common highway construction tasks have on base-level safety risk levels. Risk interactions are defined as the pair-wise impacts that tasks have on each other due to task compatibility or incompatibility. Interactions were measured as the
percentage increase or decrease in safety risk resulting from the concurrent performance of the tasks in the same physical workspace
. The research focused on the highway construction sector because this is one of the most dangerous in the construction industry (Bai, 2002; Bureau of Labor Statistics, 2010) and highway construction tasks are limited in number and well defined (Pandey, 2009).
Spatial and temporal interactions
Traditionally, safety has not received the attention that it deserves in comparison with other objectives in jobsite planning (Anumba and Bishop, 1997). Recently, however, researchers have begun to study the impact of site layout schemes on safety performance. For example, Shapira and Lyachin (2009) showed that crowded jobsites, resources constraints and overlap of activities may increase safety risks. In an effort to inte- grate safety into site layout planning, Elbeltagi
(2004) presented a method of modelling safety zones around temporary facilities. They used genetic algo- rithms to optimize the distances between facilities in order to minimize their negative interactions. Similarly, El-Rayes and Khalafallah (2005) suggested a model to consider the influence of crane operations, hazardous materials and travel routes on safety. Navon and Kolton (2006) took a different approach and showed how interactions among site layouts and planned tasks can produce fall hazards. This body of literature confirms the importance of studying risk interactions but has two main limitations: (1) the models are conceptual and are not based upon an underlying data- base; and (2) the interactions among tasks were ignored in the quantitative analyses.
While spatial safety management typically occurs during site layout planning, temporal safety manage- ment typically involves safety–schedule integration. Kartam (1997) made the first attempt to integrate safety data into schedules; however, as Hinze
(2005b) recognized, there was not an actual relation between schedule and safety resources in Kartam’s model. Consequently, Hinze
(2005b) developed software called SalusLink which allows safety person- nel to load safety components into the schedule of a project. Taking schedule integration a step further, Yi and Langford (2006) suggested a framework to inte- grate safety risk into schedules using a similar method to resource loading (i.e. assigning a safety risk quan- tity to each scheduled activity). This framework can be used to identify periods with a relatively high level of safety risk and allows managers to use resource levelling techniques to level the safety risk in a sched- ule. Similar to the spatial modelling of safety, these schedule-based techniques are not based on robust underlying data nor do they consider the interactions among tasks.
(2009) recently proposed CHASTE, a model that simultaneously considers spatial and tempo- ral interactions of work tasks. By using information available in 4D geographic models and user-provided data for ‘loss-of-control events’, the method can be used
Safety risk management
to produce a 4D view of the regions of the worksite with high levels of safety risk (Sacks
, 2009). The most significant limitation of this framework is that, in order to quantify the risk for ‘loss-of-control events’, the hazards related to each task must be identified and quantified by the user, which can be time intensive and laborious. As discussed by Jannadi and Almishari (2003), quantifying these risk values is not practical for most firms. To address the limitations in the current body of literature and to enhance the efficacy of the aforementioned safety integration models and frame- works, a database of task interactions was created.
Safety risk quantification
The prevailing methods of injury risk assessment typi- cally involve qualitative risk ratings on either linguistic or numerical scales (e.g. Hallowell and Gambatese, 2009). Typically, injury risks are evaluated using a combination of frequency ratings, severity ratings and exposure durations. When sufficient historical data are available, safety risk can be calculated by finding the product of likelihood of occurrence and magnitude of impact (Baradan and Usmen, 2006; Navon and Kolton, 2006). To date, no research has evaluated the impact of risk interactions in risk assessment. Rather, base-level task risks are evaluated individually and are rarely aggregated.