Impact of the Strictness and Cohesiveness of Management Feedback on

Construction Workers’ Safety Behavior

Agent-based Modeling and Simulation

Byungjoo Choi and SangHyun Lee

Department of Civil and Environmental Engineering, University of Michigan, 2350 Hayward Street, Ann Arbor, U.S.A.

Keywords:

Agent-based Model and Simulation, Construction Safety Management, Socio-cognitive Process, Safety

Behavior.

Abstract:

Although workers’ unsafe behaviors are the main causes of accidents in construction projects, there is a no-

ticeable lack of research addressing the mechanisms of workers’ safety behavior. In this paper, an agent-based

model that integrates the cognitive process of safety behavior and workers’ interactions with the environment

–including coworkers, management, and site condition– has been constructed. Then, model experiments were

conducted to investigate the effects of the strictness and cohesiveness of management feedback on safety

behavior. The results indicated that while strictness has signiﬁcant impacts on reducing workers’ unsafe be-

haviors, the impacts become limited in speciﬁc conditions; 1) lenient feedback in the modest-risk condition; 2)

whole range of the strictness in the low-risk condition except for very strict feedback; and 3) very strict feed-

back in high-risk conditions. Also, it was found that construction managers should achieve at least a medium

level of cohesiveness in feedback in order to prevent the negative impacts of the low cohesiveness of manage-

ment feedback. This paper contributes to the body of knowledge on construction safety as well as simulation

literature by developing the socio-cognitive process model of workers’ safety behavior that examines how the

socio-cognitive process interacts with management and the site condition.

1 INTRODUCTION

The construction industry has been notorious for its

poor safety record not only in the U.S., but also in

many other regions including EU, Asia, and Australia

(Lingard et al., 2011; Choi and Lee, 2016). Although

continuous efforts have been made to reduce the num-

ber of accidents at construction sites, the safety in the

construction still lags behind other comparable indus-

tries (Kim et al., 2013). In the U.S., for example,

894 fatal occupational injuries were reported from the

construction industry in 2014, which accounted for

18% of the total fatal occupational injuries across all

industries (BLS, 2016a). Also, non-fatal injury rate

of the U.S. construction industry is 1.4 times greater

than national average of all industries (BLS, 2016b).

Accident investigations have shown that accidents

are caused by the interaction between unsafe condi-

tions (i.e., a physical hazard in work environment) and

unsafe acts (i.e., behavior that deviates from safety

rules and procedures) (Heinrich et al., 1950). Tradi-

tionally, safety management in construction has cen-

tered on eliminating unsafe work conditions on a con-

struction site, and these efforts have resulted in signif-

icant improvements in work conditions on construc-

tion sites over the past decades (Choi et al., 2017).

However, given that more than 80% accidents are at-

tributable to workers’ unsafe behaviors (Salminen and

Tallberg, 1996), increased attention has been paid to

the improvement of workers’ safety behaviors.

In this regard, there have been continuous efforts

to identify factors affecting workers’ safety behav-

ior. Such research efforts have noted that that work-

ers’ safety behaviors are under the inﬂuence of so-

cial controls such as safety climates, safety norms,

and safety culture. Safety climate is deﬁned as ”em-

ployees’ shared perception of organizational safety

policies, procedures, and practices” (Zohar, 1980).

Also, the social norm is deﬁned as ”shared under-

standings of what is acceptable behavior and what is

not acceptable behavior for a group” (Anderson et al.,

2014). A number of previous studies have empirically

shown that workers’ shared perception of safety (i.e.,

safety climate and safety norm) plays a paramount

role in shaping their safety behavior at construction

sites (Mohamed, 2002).

348

Choi, B. and Lee, S.

Impact of the Strictness and Cohesiveness of Management Feedback on Construction Workers’ Safety Behavior - Agent-based Modeling and Simulation.

DOI: 10.5220/0006442303480355

In Proceedings of the 7th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH 2017), pages 348-355

ISBN: 978-989-758-265-3

2 RESEARCH OBJECTIVES

Although previous studies have provided ample ev-

idence on the inﬂuence of social factors on work-

ers’ safety behaviors, few research efforts have been

made to uncover the mechanism of social inﬂuence

on workers’ safety behaviors. Since most of the pre-

vious studies have attempted to identify a covariation

between workers’ shared perceptions and their safety

behaviors at a single point (i.e., cross-sectional sur-

vey, variable-based approach), it ’s hard to uncover

underlying mechanism of group behaviors emerging

from individuals’ interactions in an organization with

this methodology (Smith and Conrey, 2007). Fur-

thermore, the previous studies are limited to inves-

tigate how individuals in an organization will re-

act and which group-level phenomena will emerge

when management’s policies /interventions are imple-

mented (Ahn and Lee, 2015).

To address these limitations of the variable-based

approach, this study adopts agent-based modeling and

simulation because it has strength in generating com-

plex social phenomena emerging from the interaction

of individuals in an organization (Macy and Willer,

2002). The agent-based model consists of multiple

agents (i.e., abstraction of workers in this paper) that

interact with each other and/or with the environments

and make their own decision over time based on a

set of theoretically or empirically postulated behav-

ioral rules (Hughes et al., 2012). The modeler can ob-

serve dynamic changes in agents’ behaviors at the in-

dividual as well as group level emerging from the be-

havioral rules by running simulation with the model.

Therefore, the agent-based model provides ”genera-

tive or mechanistic explanation of observed phenom-

ena by postulating or set of mechanics, that generates

the phenomena” (Smith and Conrey, 2007). The gen-

erative explanations from agent-based model reveal

how the observed phenomena from the variable-based

approach emerge in populations of heterogeneous in-

dividuals.

With this background, this study aims to develop

an agent-based model of construction workers’ safety

behavior based on previous theoretical and empirical

ﬁndings regarding workers’ safety behavior. The de-

veloped model is expected to provide a deeper un-

derstanding of the mechanism of social inﬂuence on

workers’ safety behavior. Further, to investigate how

workers react to different safety management inter-

ventions and provide insight into the development

of effective safety management policies/strategies,

”thought experiments” are conducted using the devel-

oped model. Before preceding the next section, the

authors would like to note that the model in this pa-

per is guided by Choi and Lee (2017), but model as-

sumptions have been re-evaluated. Also, while Choi

and Lee (2017) explore the impact of the strictness

of management feedback on workers’ safety behavior

in the modest level of site risk, this paper addition-

ally considers different site conditions and the cohe-

siveness of management feedback in the experiment.

However, this paper is limited to the impacts of the

strictness and cohesiveness of management feedback

on workers’ safety behavior. The results of more com-

prehensive experiments that include additional safety

management interventions (e.g., different frequency

of management feedback and stimulation of workers’

project identiﬁcation) and interactions between the

safety management interventions will be presented in

the future journal paper.

3 LITERATURE REVIEW

Considering that workers’ safety behavior is a re-

sponse to potential hazards in the workplace, work-

ers’ unsafe behavior is in line with risk-taking behav-

ior. The theory of risk homeostasis (Wilde, 1982) pro-

vides an explanation of the cognitive process of risk-

taking behavior. According to the theory, perceived

risk and acceptable risk is two main dimension to de-

termine risk-taking behavior, and individual takes the

risk only when the perceived risk is smaller than an in-

ternal threshold (i.e., acceptable risk). In other words,

an individual perceives the risk (i.e., risk perception)

and assesses the risk based on his/her acceptable risk

(i.e., risk assessment) to determine his/her response to

the risk. The concept of perceived risk and risk accep-

tance have been also applied to explain various types

of risk-taking behavior such as ﬁnance and health risk

behavior. Also, risk perception and risk assessment

have been included in numerous studies on the causes

of workers’ unsafe behaviors (Chi et al., 2013).

Since construction workers are working in a social

context, the cognitive processes and safety behaviors

would be inﬂuenced by interaction with others. In

a construction site, interaction happens not only be-

tween workers but also between workers and manage-

ment. Construction workers create the perception of

coworkers’ risk acceptance based on their observation

of coworkers’ safety behavior (i.e., workgroup norm).

Also, workers learn risk acceptance of management

based on the management feedback on unsafe behav-

iors (i.e., management norm). If a worker receives

feedback from the management on a speciﬁc unsafe

behavior, the worker realizes that the unsafe behav-

ior is not acceptable in the current project. Construc-

tion workers establish their internal standard regard-

Impact of the Strictness and Cohesiveness of Management Feedback on Construction Workers’ Safety Behavior - Agent-based Modeling

and Simulation

349

ing safety behavior based on own risk attitude as well

as the two norms. In the same vein, recent studies

have suggested multi-level social inﬂuence model that

consists of organization and workgroup levels and

showed the separate impact of the two levels (Clarke

and Ward, 2006).

The process of social inﬂuence on an individual’s

behavior can be explained by social identity theory.

Social identity is deﬁned as ”a part of an individ-

ual’s self-concept which derives from his knowledge

of his/her membership in a social group (or groups)

together with the value and emotional signiﬁcance at-

tached to that membership” (Tajfel, 1978). The so-

cial identity theory posits that the strength of inﬂu-

ence of group norms on an individual’s behavior is de-

termined by the salience of the group membership in

his/her self-concept. In this regard, Choi et al., (2016)

investigated the effects of workgroup norm, manage-

ment norm, as well as project identiﬁcation on work-

ers’ safety behavior. The results showed that con-

struction workers’ safety behaviors are inﬂuenced by

workgroup norm and management norm, and work-

ers’ project identiﬁcation intensiﬁes the relationship

between management norm and safety behavior and

attenuates relationship between workgroup norm and

safety behavior.

4 MODEL DEVELOPMENT

The agent-based model in this paper integrates the

cognitive process and social inﬂuence process of

workers’ safety behavior. A theoretical model of the

cognitive process and empirical ﬁndings regarding so-

cial inﬂuence from Choi et al., (2016) are incorpo-

rated into the model to establish behavioral rules in

the model.

Figure 1 shows a structure of the model. The

model simulates construction workers (i.e., agent)

who are working on an artiﬁcial project. When the

model is initialized, a site condition and all workers

in the project are created and stored initial values of

the site condition and workers. The site condition has

two attributes; site risk and strictness of management

feedback. The site risk refers to the hazard level of

the project with a range between 0 and 1. It includes

the probability, that workers in the project are exposed

to unsafe work condition, and the severity of the risk

that workers are exposed. The strictness of manage-

ment feedback refers how strictly management reg-

ulates to workers’ unsafe behavior and is deﬁned as

1-risk acceptance. As such, high strictness (i.e., low-

risk acceptance of management) implies that manage-

ment does not tolerate a small risk at the project.

At every time step, every worker is exposed to a

safe or unsafe work condition. The probability of the

unsafe condition is determined by the site risk.The

worker performs a safe behavior if the worker is ex-

posed to the safe condition and does not make a mis-

take. On the other hand, the worker is provided with

the value of actual risk if the worker is under the un-

safe condition. The value of actual risk is also deter-

mined by the site risk. For example, if the site risk is

0.75, there is a 75% chance that a worker is exposed

to the unsafe condition, and the average of actual risk

that a worker will encounter is 0.75. Previously, the

actual risk was assigned based on the uniform distri-

bution between the average actual risk - 0.25 and av-

erage actual risk + 0.25 (Choi and Lee, 2017). This

modeling assumption has been re-evaluated and re-

vised because there is zero probability that workers

are exposed to the actual risk out of the uniform dis-

tribution range. For example, if a worker is under 0.75

site-risk condition, it is impossible for the worker to

encounter the actual risk below 0.5. To address this

issue, the actual risk is assigned based on the beta dis-

tribution which is deﬁned as a continuum between 0

and 1 in this paper.

In the case of the unsafe condition, cognitive pro-

cesses are activated to determine a reaction to the ex-

posed risk. In other words, the worker perceives the

risk (i.e., risk perception) and determines whether the

perceived risk is acceptable or not (i.e., risk assess-

ment). First, the risk perception refers to a worker’s

subjective judgment on the actual risk and thus per-

ceived risk may vary from worker to worker even

under the same actual risk (Shin et al., 2014). The

subjective tendency to overestimate or underestimate

the risk is deﬁned as risk perception coefﬁcient in the

model, and the risk perception coefﬁcient is associ-

ated with individual’s risk attitude. For example, a

worker who has a risk-seeking attitude tends to under-

estimate the actual risk and overestimates their abil-

ity to control the situation. As such, a risk-seeking

worker’s risk perception coefﬁcient will be below 1.0.

After perceiving the risk, the worker assesses the

risk and determines safety behavior (i.e., safe or un-

safe behavior) by comparing the perceived risk and

risk acceptance. The risk acceptance also varies from

worker to worker because some workers are more

open to take the risk while others are reluctant to ac-

cept the risk. In the model, the risk acceptance is de-

termined using the empirical results from Choi et al.,

(2016). As shown in Equation (1), the risk accep-

tance is affected by risk attitude, workgroup norm,

and management norm, Also, the relationship be-

tween safety norms (i.e., workgroup norm and man-

agement norm) and risk acceptance is moderated by

SIMULTECH 2017 - 7th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

350

Figure 1: Flowchart of Model Structure.

project identiﬁcation.

=(1 − w

)AT

+ w

((1 − p j

)W N

+ p j

) + ε

(1)

where i = worker i; RA = risk acceptance; AT = risk

attitude; WN = workgroup norm; MN = management

norm; pj= project identiﬁcation, w = weight on social

inﬂuence, and t = current time step.

To determine the risk acceptance, the worker

observes coworkers’ safety behavior and perceives

the workgroup norm. The workgroup norm in the

model is deﬁned as an individual worker’s percep-

tion of coworkers’ risk acceptance and is determined

by coworkers’ safety behavior. If the worker wit-

nesses a coworker’s unsafe behavior, the worker will

assume that the coworker performs the unsafe behav-

ior because his/her risk acceptance is higher than cur-

rent risk. On the other hand, in the case of observ-

ing the safe behavior, the worker’s perception of the

coworker’s risk acceptance will be lower than the cur-

rent risk. Also, the management norm is deﬁned as

an individual worker’s perception of management’s

risk acceptance at the current project. Likewise to

the workgroup norm, a worker’s management norm

is determined by his/her experience of management

feedback on the unsafe behavior. If a worker receives

safety feedback from management on his/her unsafe

behavior, he/she will conclude that the current risk is

not acceptable to the management. In other words,

his/her perception of risk acceptance of management

(i.e., management norm) is lower than the perceived

risk. On the other hand, if there is no feedback from

management even if the worker performs an unsafe

behavior, the worker will conclude that risk accep-

tance of the management is greater than the perceived

risk. Finally, if the worker carries out a safe behavior,

there will be no changes in the worker’s management

norm.

After assessing the perceived risk, the worker will

try to perform a safe behavior if the perceived risk

is higher than his/her risk acceptance. On the other

hand, the worker will perform an unsafe behavior if

the perceived risk is acceptable which means the per-

ceived risk is lower than his/her risk acceptance. The

worker’s unsafe behavior can result in a near miss or

accident. On the other hand, it is possible that nothing

happens to the worker because all unsafe behaviors

do not necessarily lead the near miss or accident. The

probability of the near miss or accident occurring is

determined by the actual risk which is assigned based

on the site risk at the beginning of each time step. If

the worker experiences a near miss or accident, he/she

will become more risk-averse because he/she realizes

the risk in the workplace. In the case of happening

nothing, optimistic recovery makes the worker more

risk-seeking (Shin et al., 2014).

Every time step, every worker has a chance to per-

form the safety behavior. The order of performing the

safety behavior is randomly determined to avoid pos-

sible bias due to the order of performing safety behav-

ior. After completing all workers, the model collects

Impact of the Strictness and Cohesiveness of Management Feedback on Construction Workers’ Safety Behavior - Agent-based Modeling

and Simulation

351

the group-level behaviors such as unsafe behavior ra-

tio and moves on to the next time step. The model

repeats the process until the time reaches the prede-

ﬁned maximum time step.

5 VALIDATION

Before conducting the experiment, the model validity

is examined by testing whether the developed model

is qualitatively consistent with empirical ﬁndings of

the literature regarding construction workers’ safety

behavior based on the method suggested by Ahn and

Lee (2015). First, the developed model success-

fully reproduces the effect of risk attitude on workers’

safety behavior in previous studies (Cooper, 2000).

Also, the model reafﬁrms the empirical ﬁndings of the

separate effects of workgroup norm and management

norm on workers’ safety behaviors (Meli et al., 2008).

Finally, the moderating effect of workers’ social iden-

tiﬁcation with the current project on the relationship

between social norms (i.e., workgroup norm and man-

agement norm) and safety behavior (Choi et al., 2016)

is also reproduced by the developed model.

6 SIMULATION EXPERIMENT

To investigate how workers’ socio-cognitive process

interacts with management and site condition, im-

pacts of the strictness of management feedback on

workers’ safety behaviors are investigated in different

site risk conditions. The impact of the strictness of

management feedback is tested by applying different

values in the model and comparing the results (i.e.,

parameter sweep). For this purpose, the mean of un-

safe behavior ratio of all workers for each of the 150

simulated days is measured. The range of the strict-

ness of management feedback is determined from 0.1

to 0.9 because extremely lenient or strict management

feedback do not reﬂect safety management practices.

Also, the parameter sweeps are repeated in three dif-

ferent site risk conditions (i.e., low (0.25), modest

(0.5), and high-risk (0.75) site conditions) to explore

the effects of interactions between the socio-cognitive

process, management feedback, and site condition.

In addition to the strictness of management feed-

back, effects of the cohesiveness of management

feedback are investigated using another experiment.

The impact of the cohesiveness of the management

feedback is tested by applying different ranges of

the strictness of management feedback with the same

mean in the model and compare the unsafe behav-

ior ratio. A wider range of the strictness means less

cohesive management feedback. The mean of the

strictness of management feedback is determined 0.7

because construction managers typically have a rela-

tively strict standard regarding safety behavior (Choi

et al., 2017). The simulation is repeated in three dif-

ferent ranges of the strictness of management feed-

back (i.e., high cohesiveness; Uniform distribution

[0.7, 0.7], medium cohesiveness; Uniform distribu-

tion [0.55, 0.85], and low cohesiveness; Uniform dis-

tribution [0.4, 1.0]) in the modest-risk condition (0.5).

Common parameter settings for each simulation

is described in Table 1. A 200-worker organization,

which consists of 20 crews and each of crew has

10 workers, is simulated. Within each crew, every

worker is able to observe coworkers’ safety behav-

iors, while observation across crews is limited in the

model (Ahn et al., 2013). At the beginning of the sim-

ulation, workers are initialized with the value of the

risk perception coefﬁcient and risk attitude which are

randomly assigned based on the uniform distribution

with the consideration of heterogeneity of the work-

ers. Finally, the value of the weight of social inﬂuence

in Equation (1) is determined based on the result of

Choi et al., (2016).

Table 1: Common parameter settings for simulations.

Parameter Setting

Number of workers 200 (20 x 10)

Simulation days 150

Within-crew

connection probability

1.0

Outside-crew

connection probability

0.03

Risk perception

coefﬁcient

Uniform distribution

[0.6, 1.2]

Risk

attitude

Uniform distribution

[0.1, 0.9]

7 RESULTS & DISCUSSION

For the ﬁrst experiment. the experiment runs 270 sim-

ulations for each value of the strictness of manage-

ment feedback (from 0.1 to 0.9) and site risk (i.e.,

0.25, 0.5, and 0.75) to produce a sufﬁciently large

sample size. Statistical signiﬁcance of the effect of

the strictness of management feedback is examined

using MannWhitney U test because the data could

not guarantee the assumptions required for parametric

statistical test (i.e., normality assumption). To exam-

ine the effect of the strictness, differences in the un-

safe behavior ratio between the low (0.2) and medium

level (0.5) of the strictness and between the medium

SIMULTECH 2017 - 7th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

352

(0.5) and high level (0.8) of the strictness are tested

in the three different site risk conditions. Since data

in medium strictness are used in both of the compar-

isons, the results of MannWhitney U are corrected by

Bonferroni correct to address multiple comparison is-

sue.

Figure 2 shows the effect of the strictness on

workers’ safety behavior in the modest-risk condition.

In Figure 2, the horizontal axis represents changes in

values of the strictness, and the vertical axis refers

to the unsafe behavior ratio. As shown in Figure

2, the strictness of management feedback contributes

to reducing workers’ unsafe behavior in the modest-

risk condition. First, there are signiﬁcant differences

in the median of the unsafe behavior ratio between

high strictness (Mdn = 0.355) and medium strictness

(Mdn = 0.380), U = 865,140.0, p = 2.38 × e

−63

Also, signiﬁcant differences between medium strict-

ness (Mdn = 0.380) and low strictness (Mdn = 0.395),

U = 1,030,089.0, p = 2.82 × e

−26

are found while the

differences are are less than the previous comparison.

The signiﬁcant effects imply that the strictness

of management feedback is directly associated with

workers’ perception of management norm and re-

duces workers’ unsafe behaviors. Workers are more

likely to receive safety feedback from management

and decrease their risk acceptance if management has

stricter risk acceptance because the management will

not ignore the small risk and give safety feedback to-

ward unsafe behaviors. In addition, patterns of the ef-

fects of the strictness can be found in Figure 2. In very

lenient management feedback situation (i.e., from 0.1

to 0.3), changes in the strictness have very limited im-

pact on the unsafe behavior ratio. For example, dif-

ferences in the median between 0.1 strictness (Mdn =

0.395) and 0.2 strictness (Mdn = 0.395) are not sig-

niﬁcant, U = 1,312,200.0, p = 0.499. However, the

impact becomes signiﬁcant as the strictness increases.

The effects of strictness of management feedback

on safety behavior in the low-risk condition are rep-

resented in Figure 3. As shown in Figure 3, the strict-

ness of management feedback has a relatively weaker

impact on safety behavior than in the modest-risk con-

dition. The effects of the strictness on safety behav-

ior are signiﬁcant only in very strict condition (from

0.7 to 0.9). Differences in the median of unsafe be-

havior ratio between low strict (Mdn = 0.280) and

medium strict (Mdn = 0.280) management feedback

are not signiﬁcant, U = 1,309,279.5, p = 0.912. On the

other hand, there are signiﬁcant differences between

medium strict (Mdn = 0.280) and high strict (Mdn =

0.275), U = 1,143,821.0, p = 2.43 × e

−10

Figure 4 represents the effects of strictness of

management feedback on workers’ safety behavior in

Figure 2: Inﬂuence of the Strictness of Management Feed-

back in Modest-Risk Condition.

Figure 3: Inﬂuence of the Strictness of Management Feed-

back in Low-Risk Condition.

high-risk condition. As shown in Figure 4, the strict-

ness of management feedback reduces workers’ un-

safe behavior in the high-risk condition. The median

of the unsafe behavior ratio in low (Mdn = 0.400) and

medium (Mdn = 0.350) strictness varies signiﬁcantly

and shows meaningful differences, U = 658,044.0,

p = 2.05 × e

−133

. Also, there are signiﬁcant differ-

ences between medium (Mdn = 0.350) and high (Mdn

= 0.325) strictness in the high-risk condition, U =

1,032,552.0, p = 8.12 × e

−26

. Also, patterns of the

effects can be found in Figure 4. While the strictness

has signiﬁcant effects on unsafe behaviors in the low

and medium level of strictness, the effects become

limited very strict condition (i.e., from 0.8 to 0.9). For

example, differences in the median between 0.8 strict-

ness (Mdn = 0.325) and 0.9 strictness (Mdn = 0.325)

are not signiﬁcant, U = 1,300,326.5, p = 0.328.

One possible explanation of the limited impact of

the very strict management feedback is that the mod-

est strict management feedback (i.e., 0.7 or 0.8) is al-

Impact of the Strictness and Cohesiveness of Management Feedback on Construction Workers’ Safety Behavior - Agent-based Modeling

and Simulation

353

ready strict enough to cover the risk in the high-risk

condition because workers in this condition are more

likely to be at a high level of risk. Also, the line con-

nects the mean of unsafe behavior ratio in each strict-

ness value shows inverse ”S” pattern. While slope of

the line stays stable in lenient and strict condition, the

slope becomes relatively steep in modest strict con-

dition (from 0.3 to 0.7). It implies that construction

managers should ﬁrst have the medium level of man-

agement strictness to reduce workers’ unsafe behav-

ior in the high-risk condition. Then, the construction

managers should ﬁnd other ways to improve workers’

safety behavior such as increasing the frequency of

management feedback or promoting workers’ project

identiﬁcation.

Figure 4: Inﬂuence of the Strictness of Management Feed-

back in High-Risk Condition.

The effects of the cohesiveness of management

feedback on safety behavior are represented in Figure

5. As shown in Figure 5, while the cohesiveness con-

tributes to reducing workers’ unsafe behavior when

the cohesiveness becomes medium level, the cohe-

siveness does not affect safety behavior after achiev-

ing a medium level of the cohesiveness. There are sig-

niﬁcant differences between low cohesiveness (Mdn

= 0.355) and medium cohesiveness of management

feedback (Mdn = 0.335), U = 3,624.0, p = 0.001.

This is because if the strictness of the management

feedback has a wider range (i.e., low cohesiveness),

management ignores the large risk when the strict-

ness of management becomes very low (i.e., negative

effect of low cohesiveness of management feedback).

In such case, workers may perceive very low manage-

ment norm, which increases workers’ risk acceptance

and unsafe behaviors. Although it might also be pos-

sible that management sometimes is sensitive to the

small risk in the low cohesiveness condition (i.e., pos-

itive effect), the positive effects are limited to recover

the negative effects. On the other hand, differences in

the median of unsafe behavior ratio between medium

cohesiveness (Mdn = 0.335) and high cohesiveness of

management feedback (Mdn = 0.335) are not signif-

icant, U = 4,987.5, p = 0.976. It implies that man-

agement should achieve at least medium level of the

cohesiveness of management feedback in order to pre-

vent the negative impacts of the low cohesiveness of

management feedback on workers’ safety behaviors.

Figure 5: Inﬂuence of the Cohesiveness of Management

Feedback.

8 CONCLUSIONS

In this study, an agent-based model has been de-

veloped to simulate workers’ socio-cognitive process

of safety behavior and its interaction with manage-

ment interventions (i.e., strictness and cohesiveness

of management feedback) and different site condi-

tions (i.e., site risk). The theoretical model of a cog-

nitive process (e.g., the theory of risk homeostasis)

and empirical ﬁndings regarding social inﬂuence on

workers’ safety behavior (e.g., (Choi et al., 2016))

are incorporated to simulate workers’ socio-cognitive

process of safety behavior. By running simulations

on the model with different strictnesses and cohesive-

ness of management feedback in different site risk, it

has been demonstrated that (1) the strictness of man-

agement feedback has signiﬁcant impact on reducing

workers’ unsafe behavior in the modest-risk condi-

tion, but the impacts are not limited in the lenient

management feedback, (2) the strictness of manage-

ment feedback only affects workers’ safety behav-

ior with very strict management feedback in the low-

risk condition, (3) the effect of strictness of manage-

ment feedback on workers’ safety behavior is signif-

icant, but the impact of very strict management feed-

SIMULTECH 2017 - 7th International Conference on Simulation and Modeling Methodologies, Technologies and Applications

354

back becomes limited in the high-risk condition, and

(4) construction managers should at least achieve the

medium level of cohesiveness to avoid negative re-

sults. This paper contributes to the body of knowledge

on construction safety as well as simulation litera-

ture by developing the socio-cognitive process model

of workers’ safety behavior that examines how the

socio-cognitive process interacts with management

and the site condition.

REFERENCES

Ahn, S. and Lee, S. (2015). Methodology for creating em-

pirically supported agent-based simulation with sur-

vey data for studying group behavior of construction

workers. Journal of Construction Engineering and

Management, 141(1):04014065.

Ahn, S., Lee, S., and Steel, R. (2013). Effects of workers

social learning: Focusing on absence behavior. Jour-

nal of Construction Engineering and Management,

139(8):1015–1025.

Anderson, K., Lee, S., and Menassa, C. (2014). Impact

of social network type and structure on modeling nor-

mative energy use behavior interventions. Journal of

Computing in Civil Engineering, 28(1):30–39.

BLS (2016a). Chart 3. number and rate of fatal occupa-

tional injuries by industry sector, 2014. In National

Census of Fatal Occupational Injuries in 2014. U.S.

Department of Labor.

BLS (2016b). Table 2. numbers of nonfatal occupational

injuries and illnesses by case type and ownership, se-

lected industries, 2015. In Employer-Reported Work-

place Injuries and Illnesses 2015. U.S. Department of

Labor.

Chi, S., Han, S., and Kim, D. (2013). Relationship between

unsafe working conditions and workers behavior and

impact of working conditions on injury severity in u.s.

construction industry. Journal of Construction Engi-

neering and Management, 139(7):826–838.

Choi, B., Ahn, S., and Lee, S. (2016). Role of social norms

and social identiﬁcations in safety behavior of con-

struction workers. i: Theoretical model of safety be-

havior under social inﬂuence. Journal of Construction

Engineering and Management, page 04016124.

Choi, B., Ahn, S., and Lee, S. (2017). Construction workers

group norms and personal standards regarding safety

behavior: Social identity theory perspective. Journal

of Management in Engineering, page 04017001.

Choi, B. and Lee, S. (2016). Role of social norms and so-

cial identiﬁcations in safety behavior of construction

workers. ii: Group analyses for the effects of cultural

backgrounds and organizational structures on social

inﬂuence process. Journal of Construction Engineer-

ing and Management, page 04016125.

Choi, B. and Lee, S. (2017). Modeling the effect of a socio-

psychological process on construction workers safety

behavior. In Computing in Civil Engineering (2017).

American Society of Civil Engineers.

Clarke, S. and Ward, K. (2006). The role of leader inﬂu-

ence tactics and safety climate in engaging employ-

ees’ safety participation. Risk Analysis, 26(5):1175–

1185.

Cooper, M. D. (2000). Towards a model of safety culture.

Safety Science, 36(2):111–136.

Heinrich, H. W., Petersen, D., and Roos, N. (1950). Indus-

trial accident prevention. McGraw-Hill New York.

Hughes, H. P. N., Clegg, C. W., Robinson, M. A., and Crow-

der, R. M. (2012). Agent-based modelling and simula-

tion: The potential contribution to organizational psy-

chology. Journal of Occupational and Organizational

Psychology, 85(3):487–502.

Kim, H., Lee, H., Park, M., and Choi, B. (2013). Auto-

mated information retrieval for hazard identiﬁcation

in construction sites. In Computing in Civil Engineer-

ing (2013), pages 897–904. American Society of Civil

Engineers.

Lingard, H., Cooke, T., and Blismas, N. (2011). Coworkers’

response to occupational health and safety: An over-

looked dimension of group-level safety climate in the

construction industry? Engineering, construction and

architectural management, 18(2):159–175.

Macy, M. W. and Willer, R. (2002). From factors to actors:

Computational sociology and agent-based modeling.

Annual review of sociology, pages 143–166.

Meli, J. L., Mearns, K., Silva, S. A., and Lima, M. L.

(2008). Safety climate responses and the perceived

risk of accidents in the construction industry. Safety

Science, 46(6):949–958.

Mohamed, S. (2002). Safety climate in construction site

environments. Journal of Construction Engineering

and Management, 128(5):375–384.

Salminen, S. and Tallberg, T. (1996). Human errors in fa-

tal and serious occupational accidents in ﬁnland. Er-

gonomics, 39(7):980–988.

Shin, M., Lee, H.-S., Park, M., Moon, M., and Han, S.

(2014). A system dynamics approach for modeling

construction workers safety attitudes and behaviors.

Accident Analysis and Prevention, 68(0):95–105.

Smith, E. R. and Conrey, F. R. (2007). Agent-based model-

ing: A new approach for theory building in social psy-

chology. Personality and social psychology review,

11(1):87–104.

Tajfel, H. (1978). Differentiation between social groups:

studies in the social psychology of intergroup re-

lations. European monographs in social psychol-

ogy.14. Published in cooperation with European As-

sociation of Experimental Social Psychology by Aca-

demic Press, London ; New York.

Wilde, G. J. S. (1982). The theory of risk homeostasis:

Implications for safety and health. Risk Analysis,

2(4):209–225.

Zohar, D. (1980). Safety climate in industrial organizations:

theoretical and applied implications. Journal of ap-

plied psychology, 65(1):96–102.

Impact of the Strictness and Cohesiveness of Management Feedback on Construction Workers’ Safety Behavior - Agent-based Modeling

and Simulation

355