Accident Analysis and Prevention 93 (2016) 310–318

Contents lists available at ScienceDirect

Accident Analysis and Prevention

jou rn al hom ep age: w w w . e l s e v i e r . c o m / l o c a t e / a a p

A hybrid simulation approach for integrating safety behavior

into construction planning: An earthmoving case study
Yang Miang Goh a,∗ , Mohamed Jawad Askar Ali b
a
Safety and Resilience Research Unit (SaRRU), Department of Building, School of Design and Environment, National University of Singapore,
4 Architecture Drive, Singapore 117566, Singapore
b
Safety and Resilience Research Unit (SaRRU), Department of Building, School of Design and Environment, National University of Singapore,
4 Architecture Drive, Singapore 117566, Singapore

A R T I C L E I N F O A B S T R A C T

Article history: One of the key challenges in improving construction safety and health is the management of safety
Received 22 April 2015 behavior. From a system point of view, workers work unsafely due to system level issues such as poor
Received in revised form safety culture, excessive production pressure, inadequate allocation of resources and time and lack of
18 September 2015
training. These systemic issues should be eradicated or minimized during planning. However, there is a
Accepted 23 September 2015
lack of detailed planning tools to help managers assess the impact of their upstream decisions on worker
Available online 9 October 2015
safety behavior. Even though simulation had been used in construction planning, the review conducted in
this study showed that construction safety management research had not been exploiting the potential
Keywords:
Hybrid simulation
of simulation techniques. Thus, a hybrid simulation framework is proposed to facilitate integration of
Construction safety safety management considerations into construction activity simulation. The hybrid framework consists
Safety behavior of discrete event simulation (DES) as the core, but heterogeneous, interactive and intelligent (able t o
Activity planning make decisions) agents replace traditional entities and resources. In addition, some of the cognitive
Simulation methodology processes and physiological aspects of agents are captured using system dynamics (SD) approach. The
combination of DES, agent-based simulation (ABS) and SD allows a more “natural” representation of the
complex dynamics in construction activities. The proposed hybrid framework was demonstrated using
a hypothetical case study. In addition, due to the lack of application of factorial experiment approach in
safety management simulation, the case study demonstrated sensitivity analysis and factorial experiment
to guide future research.
© 2015 Elsevier Ltd. All rights reserved.

1. Introduction planning. Computer simulation is an established method for

The Workplace Safety and Health Institute (2015) in Singapore
highlighted that the construction industry contributed to 57% of all
fatal injuries in the first half of 2014. This is despite a decrease in
total number of fatalities in all sectors. A total of 311,623 man-days
were lost during the first half of 2014 due to workplace injuries in
Singapore. Similar alarming trends could be seen throughout the
world, including the United States (Zhang et al., 2015), Hong Kong
(Li et al., 2015), Taiwan (Cheng et al., 2010) and Kuwait (Kartam
and Bouz, 1998). Hence there is heightened interest to improve
construction safety for both humanitarian as well as economic
reasons.
An important aspect of construction safety management is
the quality and depth of safety consideration during construction
∗ Corresponding author.
E-mail address: bdggym@nus.edu.sg (Y.M. Goh).
analysis and planning of construction operations and processes
(Martinez, 2010). It had been applied in a wide range of con-
struction contexts, e.g. planning for material laydown yards
(Alanjari et al., 2014), floating caisson fabrication (Pantouvakis and
Panas, 2013), bored piling (Zayed and Halpin, 2001), earthmoving
(Marzouk and Moselhi, 2004) and bridge construction (Said et al.,
2009). The range of problems that construction simulation models
were meant to resolve is very wide, and some of the typical out-
put variables evaluated include completion time, cost, productivity,
number of resources deployed, and resource utilization. However,
safety behavior considerations such as number of safety violations
and diffusion of safety behavior are usually not considered in con-
struction simulation. This is despite the fact that accidents is a
perennial problem in the construction industry (Zhou et al., 2014)
and safety behavior of workers is an important direct cause of con-
struction accidents (Zhang and Fang, 2013). Safety behavior is also
an important indicator of safety culture, which is fundamental to
safety performance of organizations (Choudhry et al., 2007). Even

http://dx.doi.org/10.1016/j.aap.2015.09.015
0001-4575/© 2015 Elsevier Ltd. All rights reserved.
 Y.M. Goh, M.J. Askar Ali / Accident Analysis and Prevention 93 (2016) 310–318
though some of the current construction simulation studies do take
safety into consideration, they are typically limited to basic con-
straints like space limitations (Marzouk and Ali, 2013) and working
hours restrictions (Alvanchi et al., 2012).
Thus, this study presents a hybrid simulation framework to inte-
grate safety behavior considerations into construction simulation
models. The proposed framework utilizes a combination of dis-
crete event simulation (DES), agent-based simulation (ABS) and
system dynamics (SD) to represent the different components of
a construction activity. In general, DES is used to represent work
processes, ABS is used to model individual agents (e.g. workers
and machines), and SD is used to represent complex variables in
agents. Such approach will facilitate a more balanced view of con-
struction activities, where safety considerations will be considered
earlier. A detailed case study is included to demonstrate the pro-
posed framework. In view of the lack of application of simulation
techniques in construction safety management research, the simu-
lation methodology is presented in detail to guide future research
in this area.
2. Simulation approaches
Even though the range of simulation techniques is very wide,
there are three main approaches: SD, ABS and DES (Pidd, 2004;
Carley, 2009). SD is grounded in systems of differential equations
and a SD model is made up of stocks, flows, and auxiliary vari-
ables that are inter-connected (Sterman, 2000). SD is known for its
emphasis on feedback between variables and the delay between
cause and effect. The core of a SD model is the stocks, which vary
at each time step based on the difference between the flow rates in
and out of the stock. A mathematical equation or an if–then rule is
embedded within each variable or flow rate in the model and the
values are analyzed using numerical methods. Variables are usu-
ally continuous and aggregated, where individual entities cannot be
identified. However, the SD approach can be tweaked to track indi-
vidual entities within the model. The modified approach is known
as agent-oriented SD (e.g. Feola et al., 2012).
In comparison to SD, ABS is focused on the design of individual
agents and the adaptive decisions and actions that they perform.
The ABS approach is also known as the “bottom-up” approach
(Miller and Page, 2007), which contrasts with the “top-down”
approach of SD (stipulating high level equations to represent dif-
ferent parts of a system). In ABS, agents can be heterogeneous and
they have the ability to adapt and interact with each other and its
environment in an autonomous fashion. Agents follow certain sets
of rules and system behavior emerge from the interactions among
the agents.
Unlike SD models, DES models advance time from one event to
another, rather than continuously. Each event corresponds to some
significant change in the model and a queue of events is maintained
in the model. Even though DES can be modeled in different ways,
most DES models take a process view of the world, i.e. the core of
the model is a sequence of steps or a flow chart, e.g. in a production
line. Entities and resources such as material, equipment and people
flow in the processes of a DES. By default, entities and resources are
not able to interact with each other and they do not display adaptive
behaviors as in ABS. DES is the most common form of simulation in
construction research (Martinez, 2010).
Hybrid simulation refers to a combination of two or more sim-
ulation approaches in a model. Even though it is possible to model
most real life systems using one of the abovementioned simulation
approaches, increasing level of complexity will often require signif-
icant improvisation of the selected simulation approach (Swinerd
and Mcnaught, 2012). When dealing with multi-faceted systems,
it may be advantageous to integrate two or more simulation
311
approaches so as to enable simple, natural and efficient represen-
tations. For instance, many DES would be more representative of
the real world if entities are agents with the ability to adapt to
changes in the model. It is also advantageous to integrate ABS with
established SD decision making models by having the SD models
embedded into the agents.
3. Simulation and safety management
Safety management is the process of planning, implementing,
checking and improving safety risk controls or interventions. Suc-
cessful construction safety management hinges on detailed and
early planning of construction activities. However, there is a lack of
detailed planning tools to help managers assess the impact of the
construction activities on safety behavior and performance. Due to
its ubiquitous application in other areas of construction planning
and evaluation, simulation appears to be a useful tool to facilitate
safety planning of construction activities. Nevertheless, this study
could not identify a comparable work that uses the full poten-
tial of simulation to identify safety interventions in construction
management. As a sample of current safety management simula-
tion studies, six papers were reviewed in detail: (1) Rudolph and
Repenning (2002), (2) Cooke (2003), (3) Cooke and Rohleder (2006),
(4) Salge and Milling (2006), (5) Sharpanskykh and Stroeve (2011)
and (6) Feola et al. (2012). The first four papers were based on SD,
the fifth is based on ABS and the last is based on agent-oriented SD.
The SD models illustrated safety management theories based on
major accidents. The agent-oriented SD model (Feola et al., 2012)
and ABS (Sharpanskykh and Stroeve, 2011) were more practical
and were focused on evaluation of safety interventions. A detailed
review of the six papers can be found in (Goh and Palak, 2014). It
was observed that SD was the most common approach and ABS is
more suited for modeling safety behavior issues. It was noted that
none of the six papers used factorial experimental design, which
is an important analytical technique in simulation studies (Kelton
and Barton, 2003).
4. Hybrid simulation framework
Instead of using the simulation approaches individually, this
study proposes a hybrid simulation framework to integrate safety
behavioral considerations into construction activity planning.
There are different ways to integrate DES, ABS and SD approaches.
For example, Peña-Mora et al. (2008) used a SD–DES hybrid
approach to model both operational and strategic levels in earth-
moving activities. In their study, SD is used to model the strategic
level, while DES is used to model the operational level. On the other
hand, Swinerd and Mcnaught (2012) conducted a detailed review
of ABS–SD models. Alvanchi et al. (2011) used a combination of SD
and DES in modeling effect of working hours on construction activ-
ity. Some possible hybrid simulation frameworks adapted from
Borshchev (2013) are summarized in Table 1.
In this study, a DES–ABS–SD approach was selected. The pro-
posed conceptual model framework is presented in Fig. 1. As
highlighted earlier, even though it is always possible to stretch any
of the simulation approach to cover all the desired features, the
hybrid approach has the advantage of allowing complex problems
to be represented more “naturally”, leading to improved efficiency
and better communication with stakeholders of the simulation
project.
The framework consists of four quadrants, each highlighting a
critical component of the framework. Since the proposed simula-
tion is focused on operational concerns, it is useful to use DES as the
core of the model. DES is widely accepted in the construction simu-
lation literature as the default approach for modeling construction
312 Y.M. Goh, M.J. Askar Ali / Accident Analysis and Prevention 93 (2016) 310–318

Table 1
Possible hybrid simulation framework for integrating safety behavior into construction activity planning.

Hybrid approach Description

ABS interact with SD Humans, machineries and organizations are modeled in ABS and activities and project environment
modeled using SD. The agents’ decisions are dependent on the variables in the SD simulation. The SD
simulation can be based on established models such as the rework cycle Kun et al. (2011).
ABS interact with DES Humans, machineries and organizations are modeled in ABS and activities and project environment
modeled using DES. The agents’ decisions are dependent on the variables in the DES simulation. DES is
very established in construction simulation and can be easily adapted to allow interaction with an ABS
model.
ABS with SD in agents Humans, machineries and organizations are modeled in ABS. SD models are embedded in each agent
to represent the decision making processes in the agents.
DES with agents as entities and/or resources The activities and environment are modeled using DES. At least some of the entities or resources in the
DES process are modeled as agents.

case study is also meant to guide future research on the use of

simulation techniques in construction safety management. The
case study was focused on earthmoving operation because it is
a common construction operation that was frequently simulated
(e.g. Smith et al., 1995; Peña-Mora et al., 2008; Vahdatikhaki and
Hammad, 2014) and most readers should be able to relate to. Fur-
thermore, truck drivers’ safety behavior is especially important to
the safety performance of the earthmoving operation because the
drivers have direct control over the trucks that they operate. In
view of the focus on safety behavior and the need to keep the case
study comprehensible, variables such as load pass time, number
of passes per load, the different timings involved in the dumping
different soil types, grade resistance and the rolling resistance of
the road were not included. The case study describes the use of
sensitivity analysis and factorial experimental design (Law, 2014)
because it was observed that this is a neglected area in safety
management simulation studies. The case study was implemented
using the simulation software AnyLogicTM (Anylogic Company,
Fig. 1. Conceptual framework for the proposed hybrid simulation model.
2014).

processes and activities (Abourizk, 2010; Martinez, 2010). As high-

lighted in the framework (Fig. 1), workflow or sequence of steps for
production and safety will be captured using DES. Some examples of
safety processes include application for safety permits, site inspec-
tions and accident investigation. ABS is used to model human and
machine agents. Unlike the homogeneous entities and resources
in traditional DES, agents in ABS are “naturally” heterogeneous,
adaptive, have decision making capabilities and they interact with
other agents. Thus, ABS is the most suitable for capturing the safety
behavior of workers (human agents). The range of variables related
to safety behaviors can be very wide; some examples are listed in
the top left quadrant of Fig. 1, but as in the case for the other quad-
rants, the list is not meant to be comprehensive. The framework
also suggests the use of SD to capture the internal dynamics of each
human agents. SD has established equations and models to repre-
sent human decision making processes (Sterman, 2000) that can
be utilized in the agents. In the proposed framework, the machine
agents are essentially entities or resources in traditional DES mod-
els. However, when necessary, the entities or resources in DES can
also be programmed to display agent characteristics, e.g. ability to
interact and vary their behaviors. Lastly, the lower right quadrant
highlights the physical and social environment that the processes,
human agents and machine agents operate in. The environment
defines the physical and social relationships among the processes
and agents. It can also contain important environmental factors
such as weather, temperature and humidity.
5. Case study
The aim of this case study is to demonstrate how safety behav-
ior considerations can be incorporated into construction planning
through the proposed DES–ABS–SD framework. In addition, the
5.1. Case scenario
In the hypothetical scenario, dump trucks and excavators are to
be deployed to transport a stockpile of soil to a location to facilitate
land reclamation work. However, the earthmoving contractor had
a series of truck-related accidents and they are concerned about the
safety behavior of the truck drivers. Thus, besides aiming to reduce
completion time and minimize the number of plants deployed, the
manager needs to consider ways to reduce the number of unsafe
behaviors. Specifically, the contractor is concerned with speed-
ing violations and driving under fatigue. The management levers
include adjusting the number of trucks and excavators, increas-
ing available rest time and conducting intensive safety training to
change the safety attitude of drivers. Even though this case study is
hypothetical, basic parameters such as loading duration, duration
of breaks, capacity of trucks and speed limit were based on actual
data from a cut and cover tunneling project.
By default, the company allocated 10 trucks, 2 excavators, and
75 min of rest time per driver. Safety attitude of drivers is measured
on a scale of 0–100 and is represented as a random variable with
a triangular distribution. It is assumed that the manager consulted
his supervisors and estimated the minimum, median and maxi-
mum safety attitude of his drivers as 20, 50 and 80 respectively.
Safety climate surveys (e.g. Huang et al., 2013) and questionnaire
developed based on behavioral models such as theory of planned
behavior (e.g. Goh and Sa’adon, 2015) can be used to help managers
estimate safety attitude more accurately. The default parameters
for the case study are summarized in Table 2. It is assumed that
the manager wants to find out how he can adjust parameters 1, 2,
7 and 8 in Table 2 to improve the job completion time, minimize
resources deployed and reduce unsafe behaviors.
 Y.M. Goh, M.J. Askar Ali / Accident Analysis and Prevention 93 (2016) 310–318 313

Table 2 allowed to rest for a maximum of 75 min per day. After unloading
Default parameters for hypothetical case study.
or resting, drivers will return to the queue for loading. Fig. 3 shows
No. Parameters Values the discrete event simulation implementation of Fig. 2 in AnyLogic
1 Number of trucksa 10 7. The simulation ends when all the soil had been re-located and all
2 Number of excavatorsa 2 the trucks had returned to their base location.
3 Number of loading ports 2
4 Volume of soil to be moved per day 2040 m3
5 Truck capacity 18 m3 5.3. Truck and human agents – ABS–SD
6 Haul distance 15 km
7 Rest time per daya 75 min The trucks and the drivers are modeled as agents with the abil-
8 Median safety attitudea 50
ity to alter their behaviors. Each driver starts off with a randomly
9 Speed limit 60 km/h
10 Threshold attention level 50
assigned safety attitude level. Safety attitude affects the selection
a
of speed during driving and it can be altered when drivers interact
Adjustable parameter.
during rest breaks. Fig. 4 shows the state chart for modeling the
behavior of truck drivers. Within the “working” state, the driver is
driving and s/he can select the desired driving speed, but it will
never go below 45 km/h or exceed 75 km/h. If the speed exceeds
60 km/h, a speeding violation is recorded in the simulation. The
model keeps track of each driver’s fatigue level (measured by atten-
tion resource level) and the driver can rest if s/he has not exceeded
the maximum rest time allocated. The driver can only rest at the
dumping point and will have to drive while under fatigue if s/he
runs out of rest time.
The speed selection of the truck driver is modeled based on the
simplified trip disutility concept model suggested by (Tarko, 2009),
which states that the drivers trade-off a portion of their safety for
their time gain. The safety attitude of the driver acts as a speed
Fig. 2. Basic workflow for the earthmoving activity. deterrent and the perceived value of time gain or the production
pressure acts as speed enforcement. This relationship is in line with
the theory by Fuller (2005), who indicated that drivers adjust their
5.2. Processes – DES
behavior to maintain the current workload below their capacity.
The speed of the truck in the model depends on the SD Eqs. (1)–(3),
Fig. 2 shows the workflow for the earthmoving activity. The
trucks queue to enter loading ports where excavator(s) is (are) used Truck Speed = max(l, min(u, l + (Sf + (100 − Sa)) ∗ (Er − R))) (1)
to load soil onto the truck. Once loaded, the trucks haul to the dump-
ing point to unload the soil. Truck drivers can rest at the designated where u is the upper speed limit of the truck (75 km/h), l the lower
resting area in the dumping point whenever they want, but they are speed limit of the truck (45 km/h), Sf is the default scaling factor

Fig. 3. Discrete event simulation for the earthmoving process.

314 Y.M. Goh, M.J. Askar Ali / Accident Analysis and Prevention 93 (2016) 310–318

Resource Recovery Ratet = Min [Maximum Resource Recovery

∗(1 − Workload statust ), (100% − Attention Resource levelt )

/∆t + Resource Consumption Ratet ]

(5)

Attention Resource levelt = Attention Resource levelt−∆t

+(Resource Recovery Ratet−∆t

− ∆t )
Resource Consumption Ratet ∗ ∆t
(6)

Once the Attention Resource levelt falls below the threshold

attention level, the truck driver is deem to be in the tired state,
i.e. under fatigue. The driver under fatigue will continue driving
until s/he reaches the resting location (dump point) and only if he
has not exceeded the maximum rest time allocated.

5.4. Social environment

When more than one driver is resting at the resting area, the
drivers interact. It is assumed that whenever two drivers inter-
act, they can potentially influence each other’s safety attitude. This
process is based on the Theory of Planned Behavior (Ajzen, 1991),
which indicates that intentional behavior, e.g. safety violation, is
dependent on attitude, subjective norm and/or perceived behav-
ioral control. In this case, the effect of subjective norm is assumed
to be dominant. In the model, when two truck drivers meet, the
driver with the higher absolute deviation from the median safety
attitude is able to influence the other truck driver’s safety attitude.
The affected truck driver will adopt a new level of safety attitude
calculated based on the average of the safety attitude of the two
Fig. 4. Truck driver state chart. drivers. As an illustration, two drivers (A and B) met at the rest
point. Driver A’s safety attitude is 40 point higher than the median
(50), and Driver B is 10 point lower than the median. In this case,
(80) , Sa is the safety attitude of the driver (triangular (20, 50, 80)), Driver A is the “influencer” and Driver B’s new safety attitude is
R is the current rate of work and Er the expected rate of work. derived based on the average of the two drivers’ safety attitude
Eq. (1) captures the effect of production pressure on speed selec- ((90 + 40)/2 = 65).
tion and how higher (lower) safety attitude can decrease (increase) The final simulation model is an integration of the DES–SD–ABD
the effect of production pressure. The rate of work, R, is calculated model in which the operational flow of the earthmoving operation
dynamically and at any given point of time, R is determined based is defined by the DES system. The haul times, the resting decision
on and the co-worker influence of the truck drivers are determined by
the underpinning ABS–SD models.
total volume of soil moved
R= (2)
time worked 5.5. Sensitivity analysis
Expected rate of work is given by
A simple one-factor-at-a-time (or one-way) sensitivity analysis
N ∗ Tc (Chua et al., 1997; Goh and Chua, 2013) was conducted to demon-
E (3)
= strate how to evaluate the need for more detailed data collection
t ∗ Tn
r
on selected parameters. In this case study, the sensitivity of the
where N is the expected total number of hauls, Tc is the capacity of response variables to changes in the arbitrary constants such as
the truck, t is the total time assigned for the earthmoving operation median safety attitude, scaling factor (Eq. (1)), threshold atten-
and Tn is the total number of trucks assigned. tion level (see Table 2) and maximum recovery rate (Eq. (4)) were
Fatigue is an important factor influencing safety performance. evaluated. The values of the parameters were increased by 10% of
This study uses the SD model of mental fatigue by Alvanchi et al. their base value individually. It is noted that the magnitude of the
(2012), which is based on the Theory of Limited Resources (Eqs. increase was arbitrarily chosen to assess the sensitiveness of the
(4)–(6)) to model the mental fatigue of the truck driver during work. response variables to the factors or independent variables. This is
acceptable as the sensitivity analysis is a preliminary analysis to
Resource Consumption ratet = Min[(Resource Consumption Index
guide further data collection. For each variation, the simulation
model was executed 40 times and the percentage change in the
−Maximum Resource Recovery Rate) average value of the response variables were captured in Table 3.
Attention Resource levelt (4)
∗Workload Statust, Based on Table 3, the maximum recovery rate had the highest
∆t
impact on the number of speeding violations and average time in
+Resource Recovery Ratet ]
tired status. Similarly, a 10% increase in the threshold attention
 Y.M. Goh, M.J. Askar Ali / Accident Analysis and Prevention 93 (2016) 310–318 315

Table 3
Sensitivity analysis of the assumed values.

Response variable Base value % Change in response due to 10% increase in:

Median safety attitude Scaling factor Threshold attention level Maximum recovery rate

Number of speeding violations 4.3 −10.5% +38.5% +28% −97.6%

Average time in tired status (min) 83.3 −0.2% −2.1% + 14.2% −35.2%
Job completion time (min) 635.0 0 0 0 0

Table 4 interventions and the interactions between factors (or controllable

Factor levels for the four-factor experiment.
Factor No. Factor Negative level (−) Positive level (+)
1 Number of trucks 8 10
2 Number of excavators 1 2
3 Resting duration (min) 60 75
4 Median safety attitude 40 60
level leads to more than 10% change in number of speeding viola-
tions and average time in tired status. It is observed that only the
number of speeding violations is sensitive to a 10% increase in the
scaling factor. Job completion time is not sensitive to changes in
the variables. However, maximum recovery rate, threshold atten-
tion level and scaling factor should be carefully evaluated because
of their impact on number of speeding violations and average time
in tired status. Some methods to increase the credibility of these
variables are calibration based on available response data, compar-
ing with expert opinion and comparing with established models
(Law, 2014).
The sensitivity analysis is a useful rough gauge of the impact
of changes in parameters on the response variables. The results
of the sensitivity analysis can provide guidance on whether more
data should be collected to better represent an input parameter.
However, it does not consider interactions between the different
input parameters.
5.6. Factorial experimental design
While the sensitivity analysis is useful for guiding further
data collection, a more rigorous method for evaluating impact
of changes in input variables is factorial experimental design
(Montgomery and Runger, 1999). A factorial experiment is espe-
cially important when the possible combination of interventions to
improve system performance is not clearly stipulated. The factorial
experiment is meant to guide managers in identifying impactful
variables). Despite its importance, none of the six safety simula-
tion papers reviewed (Rudolph and Repenning, 2002; Cooke, 2003;
Cooke and Rohleder, 2006; Salge and Milling, 2006; Sharpanskykh
and Stroeve, 2011; Feola et al., 2012) conducted factorial exper-
iments. A 2k factorial design is performed herein to encourage
its usage when utilizing the proposed framework. Table 4 shows
the four key factors that the manager can modify and the two
credible levels (negative and positive) identified by the manager.
Since there are four factors and each can be varied based on the
two levels, there are 16 possible combinations of the factors (see
Table 5). For each of the 16 system configurations, the simulation
was replicated 40 times and the results were analyzed in the sta-
tistical software SPSS. The average response values are captured in
Table 5.
The main effect of the factor j is the average change in the
response due to moving the factor j from its “ − ” level to the “+”
level while holding all other factors fixed. This average is taken
over all the possible combination of factors. For example, the effect
of factor 1, e1v, on the response variable “No. of speeding violations”
is calculated as
(V2 − V1) + (V4 − V3) + (V6 − V5) + · · · + (V16 − V15)
e1 v = (7)
8
where V1 is the ‘Number of speeding violations’ corresponding to
design point 1, V2 is the ‘Number of speeding violations’ corre-
sponding to design point 2 and so on.
Determining the interactions between the factors is one of the
key advantages of conducting a factorial experiment over a sensi-
tivity analysis (varying one variable at a time). Two factors interact
when the effect of a factor is dependent on the value of another fac-
tor (Law, 2014). If interaction is ignored, effects can be erroneously
assumed leading to ineffective interventions. As an example, the
effect of the interaction between factor 1 and 2 on the response

Table 5
Experiment design matrix.

Design point Factora No. of speeding Average time in Job completion

violations tired status (min) time (min)
1 2 3 4

1 − − − − 116.4 203.8 711.3

2 + − − − 100.0 181.6 676.9
3 − + − − 95.1 180.5 677.9
4 + + − − 2.2 154.6 632.0
5 − − + − 146.0 147.5 716.5
6 + − + − 128.6 122.4 682.5
7 − + + − 134.1 123.5 687.6
8 + + + − 4.4 93.5 638.2
9 − − − + 124.8 188.5 711.9
10 + − − + 100.9 177.5 677.5
11 − + − + 98.0 175.3 678.8
12 + + − + 2.6 149.8 632.4
13 − − + + 141.0 156.4 712.9
14 + − + + 131.6 117.5 681.6
15 − + + + 131.9 125.9 688.0
16 + + + + 5.4 84.8 639.0
a
1 – number of trucks; 2 – number of excavators; 3 – rest duration; 4 – median safety attitude.
316 Y.M. Goh, M.J. Askar Ali / Accident Analysis and Prevention 93 (2016) 310–318

Fig. 5. Main effects and factor interactions.

variable “No. of speeding violations” is calculated as
1 (V4 − V3) + (V8 − V7) + (V12 − V11) + · · · + (V16 − V15)
e12v =
2 4
(V2 − V1) + (V6 − V5) + (V10 − V9) + · · · + (V14 − V13)
− 4
(8)
Fig. 5 shows a summary of the results of the simulation exper-
iments. As can be observed, even though many of the interactions
are statistically significant, the actual magnitude of the interactions
diminish as the order of the interaction increases. Thus, the main
effects dominate all the responses.
5.7. Implications
Fig. 6 summarizes the results of the analysis. The manager is
able to adjust the numbers of trucks and excavators, rest time and
median safety attitude to balance activity completion time and
safety performance. Safety performance is represented by the num-
ber of speeding violation and average time driven under fatigue,
which are influenced by production pressure. Production pressure
 Y.M. Goh, M.J. Askar Ali / Accident Analysis and Prevention 93 (2016) 310–318
Fig. 6. Influence diagram summarizing influence of factors on responses.
arise when the expected production rate per driver and/or when
the gap between target completion time and actual completion
time is increased. Actual completion time is decreased when the
numbers of trucks and excavators are decreased. Although increas-
ing rest time will reduce the average time driven under fatigue, it
causes actual completion time to be increased. On the other hand,
median safety attitude helps to reduce the effect of production
pressure, but the effect is minor as compared to the effect of produc-
tion pressure. Therefore, within the confines of the values stated in
Table 4, increasing the numbers of trucks and excavators from 8 to
10 and 1 to 2 respectively, will help to reduce activity duration, and
reduce production pressure on drivers. This indicates the impor-
tance of allocating sufficient plants for the activity. Increasing rest
time and having intensive safety training targeted on improving
safety attitude have effect on safety performance, but these effects
can be easily negated by production pressure arising due to lack of
resources.
6. Limitations and future research
The study proposed a hybrid simulation framework for model-
ing construction activities so that safety behavior can be considered
during planning. The case study showed how the hybrid sim-
ulation framework can be implemented. In view of the lack of
application of factorial experimental approach in current safety
management simulation studies, a detailed analysis was demon-
strated. The case study was hypothetical and simple in many
aspects, but this limitation does not affect the core purpose of this
paper, which is to present the hybrid simulation framework for
integrating safety behavior considerations into construction simu-
lation models. Future research will implement the framework using
actual data.
One of the key challenges in implementing the framework is
the modeling of the agents. This paper presented some suggested
approaches, but future studies should review a wider range of
fundamental psychological research on risk perception and safety
behavior (e.g. Slovic, 2000, 2010) and select the appropriate model
to represent safety behavior. In addition, more detailed calibration
of assumed variables and comparison with experts and estab-
lished models should be conducted to validate the simulation
model. A “standard agent” can be created in the future to stan-
dardize the different behavioral, physiological and psychological
aspects important to construction safety and productivity. With the
“standard agent”, the complexity and effort for developing simula-
tion models will be reduced significantly.
317
In terms of analysis, the effects and interactions derived from the
factorial experiments should be interpreted within the bounds of
the factor levels. Future studies should utilize optimization tech-
niques, such as genetic algorithms, to optimize the controllable
factors automatically.
Even though this study adopted AnyLogicTM to develop the case
study, the conceptual framework and simulation approach can be
implemented in any modeling tool or a general purpose program-
ming language. Using a commercial software has many advantages
in terms of having a more user-friendly development environment
and technical support, but users can be restrained by the user inter-
face and it also prevents collaborations among developers. The
price of AnyLogicTM may also prevent researchers and user from
adopting the proposed approach more willingly. Future research
will explore how the hybrid simulation framework can be imple-
mented in general purpose programming languages like Java or
Python.
7. Conclusions
The construction industry needs to develop new approaches to
improve its safety and health performance. One way to improve
safety and health performance is to encourage early planning of
construction processes and consider the possible impact on safety
behavior. This study proposed a hybrid simulation framework to
integrate safety behavior considerations in construction activity
simulation. The hybrid simulation framework is based on a combi-
nation of discrete event simulation (DES), agent-based simulation
(ABS) and system dynamics (SD). Since DES is the most common
simulation approach in the construction industry, it is natural to
use DES as the core of the proposed framework. However, in the
proposed framework, the entities and resources in the DES are
agents that are heterogeneous, able to make decisions and can
interact with other agents. Due to the ability of SD to account
for feedback and delays and its established approaches for mod-
eling complex decision making, SD was used to model agents’
decision making and physiological processes. The DES–ABS–SD
framework was demonstrated through a hypothetical earthmoving
case study. A sensitivity analysis and a factorial experiment were
conducted to encourage correct utilization of these analytical meth-
ods in construction safety management simulation. The case study
showed how managers can utilize the hybrid simulation model to
select suitable interventions to balance production and safety goals.
In contrast to current construction planning simulation models,
the proposed hybrid simulation approach allows integrates safety
318 Y.M. Goh, M.J. Askar Ali / Accident Analysis and Prevention 93 (2016) 310–318
behavior issues into the model. This will help to engender a greater
focus on construction safety issues upstream and can potentially
lead to better safety performance during construction.
Acknowledgement
This research is funded by the Singapore Ministry of Education
(grant number: R-296-000-143-133).
References
Abourizk, S., 2010. Role of simulation in construction engineering and
management. J. Constr. Eng. Manag. ASCE 136 (10), 1140–1153.
Ajzen, I., 1991. The theory of planned behavior. Organ. Behav. Hum. Decis. Process.
50 (2), 179–211.
Alanjari, P., Razavialavi, S., Abourizk, S., 2014. A simulation-based approach for
material yard laydown planning. Autom. Constr. 40, 1–8.
Alvanchi, A., Lee, S., Abourizk, S., 2011. Modeling framework and architecture of
hybrid system dynamics and discrete event simulation for construction.
Comput-Aided Civ. Infrastruct. Eng. 26 (2), 77–91.
Alvanchi, A., Lee, S., Abourizk, S., 2012. Dynamics of working hours in construction.
J. Constr. Eng Manage. ASCE 138 (1), 66–77.
Anylogic Company, 2014. Anylogic – multimethod simulation software.
Borshchev, A., 2013. The big book of simulation modeling: multimethod modeling
with anylogic 6. AnyLogic North America, New York.
Carley, K.M., 2009. Computational modeling for reasoning about the social
behavior of humans. Comput. Math. Org. Theory 15 (1), 47–59.
Cheng, C.-W., Lin, C.-C., Leu, S.-S., 2010. Use of association rules to explore
cause–effect relationships in occupational accidents in the Taiwan
construction industry. Saf. Sci. 48 (4), 436–444.
Choudhry, R.A., Fang, D.P., Mohamed, S., 2007. The nature of safety culture: a
survey of the state-of-the-art. Saf. Sci. 45 (10), 993–1012.
Chua, D.K.H., Kog, Y.C., Loh, P.K., Jaselskis, E.J., 1997. Model for construction budget
performance – neural network approach. J. Constr. Eng. Manage. 123 (3),
214–222.
Cooke, D.L., 2003. A system dynamics analysis of the westray mine disaster. Syst.
Dyn. Rev. 19 (2), 139–166.
Cooke, D.L., Rohleder, T.R., 2006. Learning from incidents: from normal accidents
to high reliability. Syst. Dyn. Rev. 22 (3), 213–239.
Feola, G., Gallati, J.A., Binder, C.R., 2012. Exploring behavioural change through an
agent-oriented system dynamics model: the use of personal protective
equipment among pesticide applicators in Colombia. Syst. Dyn. Rev. 28 (1),
69–93.
Fuller, R., 2005. Towards a general theory of driver behaviour. Acc. Anal. Prev. 37
(3), 461–472.
Goh, Y.M., Chua, D., 2013. Neural network analysis of construction safety
management systems: a case study in Singapore. Constr. Manage. Econ. 31 (5),
460–470.
Goh, Y.M., Palak, G., 2014. Evaluating the potential and challenges of utilising
computational simulation to design safety management and culture
interventions. In: CIB W099 International Conference on Achieving Sustainable
Construction Health and SafetyIngvar Kamprad Design Centre (IKDC), Lund,
Sweden.
Goh, Y.M., Sa’adon, N.F., 2015. Cognitive factors influencing safety behavior at
height: a multimethod exploratory study. J. Constr. Eng. Manage. Publ.
Online.
Huang, Y.H., Zohar, D., Robertson, M.M., Garabet, A., Lee, J., Murphy, L.A., 2013.
Development and validation of safety climate scales for lone workers using
truck drivers as exemplar. Transp. Res. Part F: Traffic Psychol. Behav. 17,
5–19.
Kartam, N.A., Bouz, R.G., 1998. Fatalities and injuries in the Kuwaiti construction
industry. Acc. Anal. Prev. 30 (6), 805–814.
Kelton, W.D., Barton, R.R., 2003. Experimental design for simulation: experimental
design for simulation. In: Proceedings of the 35th Conference on Winter
Simulation: Driving Innovation. Winter Simulation Conference, New Orleans,
Louisiana, pp. 59–65.
Kun, H.U., et al., 2011. Factors influencing the risk of falls in the construction
industry: a review of the evidence. Construct. Manage. Econ. 29 (4), 397–416.
Law, A.M., 2014. Simulation Modeling and Analysis, 5th ed. McGraw-Hill, New
York.
Li, H., Lu, M., Hsu, S.-C., Gray, M., Huang, T., 2015. Proactive behavior-based safety
management for construction safety improvement. Saf. Sci. 75 (0), 107–117.
Martinez, J.C., 2010. Methodology for conducting discrete-event simulation studies
in construction engineering and management. J. Constr. Eng. Manage. ASCE
136 (1), 3–16.
Marzouk, M., Ali, H., 2013. Modeling safety considerations and space limitations in
piling operations using agent based simulation. Expert Syst. Appl. 40 (12),
4848–4857.
Marzouk, M., Moselhi, O., 2004. Multiobjective optimization of earthmoving
operations. J. Constr. Eng. Manage. 130 (1), 105–113.
Miller, J.H., Page, S.E., 2007. Complex Adaptive Systems: An Introduction to
Computational Models of Social Life. Princeton University Press.
Montgomery, D.C., Runger, G.C., 1999. Applied Statistics and Probability for
Engineers, 2nd ed. John Wiley & Sons, New York.
Pantouvakis, J.P., Panas, A., 2013. Computer simulation and analysis framework for
floating caisson construction operations. Autom. Constr. 36, 196–207.
Peña-Mora, F., Han, S., Lee, S., Park, M., 2008. Strategic-operational construction
management: hybrid system dynamics and discrete event approach. J. Constr.
Eng. Manage. 134 (9), 701–710.
Pidd, M., 2004. Computer Simulation in Management Science. John Wiley & Sons,
Chichester.
Rudolph, J.W., Repenning, N.P., 2002. Disaster dynamics: understanding the role of
quantity in organizational collapse. Adm. Sci. Q. 47 (1), 1–30.
Said, H., Marzouk, M., El-Said, M., 2009. Application of computer simulation to
bridge deck construction: case study. Autom. Constr. 18 (4), 377–385.
Salge, M., Milling, P.M., 2006. Who is to blame, the operator or the designer? Two
stages of human failure in the Chernobyl accident. Syst. Dyn. Rev. 22 (2), 89–
112.
Sharpanskykh, A., Stroeve, S.H., 2011. An agent-based approach for structured
modeling, analysis and improvement of safety culture. Comput. Math. Org.
Theory 17 (1), 77–117.
Slovic, P., 2000. The Perception of Risk. Earthscan Publications, Sterling, VA/London.
Slovic, P., 2010. The Feeling of Risk: New Perspectives on Risk Perception.
Earthscan, Washington, DC/London.
Smith, S.D., Osborne, J.R., Forde, M.C., 1995. Analysis of earth-moving systems
using discrete-event simulation. J. Constr. Eng. Manage. 121 (4), 388–396.
Sterman, J.D., 2000. Business Dynamics: Systems Thinking and Modeling for a
Complex World. Irwin/Mac-Graw Hill, Boston.
Swinerd, C., Mcnaught, K.R., 2012. Design classes for hybrid simulations involving
agent-based and system dynamics models. Simul. Model. Pract. Theory 25 (0),
118–133.
Tarko, A.P., 2009. Modeling drivers’ speed selection as a trade-off behavior. Acc.
Anal. Prev. 41 (3), 608–616.
Vahdatikhaki, F., Hammad, A., 2014. Framework for near real-time simulation of
earthmoving projects using location tracking technologies. Autom. Constr. 42,
50–67.
Workplace Safety and Health Institute, 2015. Overall Workplace Injuries and
Fatalities Increase in First Half of 2014. Singapore.
Zayed, T.M., Halpin, D.W.,2001. Construction i: simulation of bored pile
construction. In: Proceedings of the 33rd Conference on Winter simulation.
IEEE Computer Society, Arlington, Virginia, pp. 1495–1503.
Zhang, M., Fang, D., 2013. A cognitive analysis of why Chinese scaffolders do not
use safety harnesses in construction. Constr. Manage. Econ. 31 (3), 207–222.
Zhang, S., Sulankivi, K., Kiviniemi, M., Romo, I., Eastman, C.M., Teizer, J., 2015.
Bim-based fall hazard identification and prevention in construction safety
planning. Saf. Sci. 72 (0), 31–45.
Zhou, Z., Goh, Y.M., Li, Q., 2014. Overview and analysis of safety management
studies in the construction industry. Saf. Sci. 72 (February 2015), 337–350.