Research on Causes of Unsafe Behaviors Among Special Operations Personnel in Building Construction Based on DEMATEL-ISM-BN

Abstract

To investigate the causal factors and pathways leading to unsafe behaviors among special operation construction workers, this study employed Ability-Motivation-Opportunity (AMO) theory and case study analysis to identify causal factors across the ability, motivation, and opportunity dimensions. These factors were subsequently analyzed using an integrated approach combining the Decision-Making Trial and Evaluation Laboratory method, Interpretive Structural Modeling, and Bayesian Network (DEMATEL-ISM-BN). This analysis revealed the causal interrelationships, hierarchical structure, and primary causal chain for unsafe behaviors. DEMATEL results identified risk-taking propensity, weak responsibility awareness, inadequate supervision mechanisms, insufficient safety training, safety culture deficiency, uncertified operation, and safety knowledge deficiency as key causal factors. ISM results positioned safety culture deficiency, inadequate supervision mechanisms, and insufficient safety investment at the deepest level (Level 5), indicating their status as fundamental underlying causes. BN analysis determined the primary causal chain to be: Safety culture deficiency → Insufficient safety training → Safety knowledge deficiency → Uncertified operation → Risk-taking propensity. This study can provide theoretical support for the management of unsafe behaviors among special operation personnel in building construction.

1. Introduction

Special operations represent high-risk and technologically demanding activities within building construction, where safety management standards directly influence overall project safety [1]. Research indicates that approximately 80% of construction accidents are closely linked to unsafe behaviors among special-operation personnel [2]. Representative cases such as the “May 6 Crane Injury Accident at Shanghai Yangpu Riverside N1-01 Plot” and the “September 25 Scaffold Collapse Accident in Juxian County, Shandong,” highlight issues like regulatory violations and unsafe operations by these workers. Special-operation personnel constitute both critical targets for accident prevention and key agents responsible for achieving intrinsic construction safety [3,4]. Consequently, effectively identifying the causal factors and pathways leading to unsafe behaviors among this group, and scientifically managing these behaviors, holds significant importance for ensuring construction process safety.

In recent years, numerous scholars both domestically and internationally have continuously conducted research on the causes of unsafe behaviors among construction workers; in studies identifying causal factors of unsafe behaviors, researchers have often categorized contributing factors within the “Human-Machine-Environment-Management” framework. For instance, Ni employed grounded theory to analyze three levels: individual, organizational and managerial, and environmental, identifying 13 triggering factors for unsafe behaviors among new-generation construction workers [5]; Zhu explored the influencing factors of unsafe behaviors among decoration and renovation workers from individual, organizational, and environmental perspectives, finding that non-compliance with safety procedures and protocols increases unsafe behavior at the individual level [6]; Raviv used the Analytic Hierarchy Process to quantify accident consequences, revealing the relationship between technical and human factors in tower crane accidents [7]; Li conducted a questionnaire survey to identify the safety atmosphere factors influencing unsafe behavior [8]; Yue investigated how organizational management measures drive collective safety behaviors among construction workers by influencing intra-group factors [9]. Malakoutikhah extracted causal factors of unsafe behaviors from organizational, individual, and socio-economic dimensions based on Grounded Theory [10,11].

In terms of analytical methods for causal factors of unsafe behaviors, existing studies predominantly employ Structural Equation Modeling (SEM) and complex network theory to validate causal relationships among factors. For example, Sun employed structural equation modeling based on the Theory of Planned Behavior to validate the influence of belief factors and situational factors on construction workers’ unsafe behaviors [12]; Gou used structural equation modeling to validate the positive influence of safety perception biases on construction workers’ unsafe behaviors, as well as the mediating effect of risk-taking tendencies between safety perception biases and unsafe behaviors [13]; Zhang collected questionnaire data on the multidimensional manifestations of unsafe behavior among managers and construction workers, using change rates, heat maps, and structural causal models to analyze the data, finding that safety organizational measures have a sustained influence on workers’ safety behaviors [14]; Huang used structural equation modeling to validate the mediating role of attachment relationships between foremen’s safety leadership and construction workers’ safety behaviors [15]. Based on accident cases, Duan constructed a complex network comprising 85 unsafe behavior nodes, thereby identifying key influencing factors of unsafe behavior [16]. Through government case collection, standardized behavioral classification, and complex network modeling analysis, Guo examined the factors influencing unsafe behaviors among construction workers, identifying the characteristics of their risk networks and the core triggering role of individual behaviors [17,18]; Yao employed Structural Equation Modeling to examine the effects of workers’ perceived behavioral control, behavioral attitudes, risk preference, and subjective norms on intentional unsafe behaviors [19].

Obviously, existing research findings have laid a solid foundation for in-depth studies of unsafe behaviors. However, most studies focus on the broad group of construction workers and their established causal factor systems largely adhere to the traditional “Human-Facility-Environment-Management” framework. Unsafe behaviors among special operation personnel are often the product of a complex coupling of deep-seated cognition, psychology, and organizational supervision. The traditional “Human-Machine-Environment-Management” framework cannot meticulously dissect ability elements such as cognitive decision-making and professional skills of special operation personnel in complex situations, nor does it thoroughly analyze the intrinsic motivation types triggering unsafe behaviors. Consequently, existing research lacks sufficient explanatory power regarding the behavioral motivations of this critical group—special operation personnel undertaking high-risk and high-technical-demand tasks in building construction. Furthermore, most existing methods explain factor relationships with limited data support and lack research on the hierarchical relationships and causal pathways of factors causing construction workers’ unsafe behaviors. For instance, Structural Equation Modeling relies on static hypothesis testing, while complex network theory emphasizes topological relationship description; both struggle to achieve dynamic assessment of causal pathways.

To break through these limitations and further enhance the systematic and logical nature of causal analysis, this paper introduces the Ability-Motivation-Opportunity (AMO) theory [20]. Starting from the three dimensions of individual ability, motivation, and opportunity, it systematically organizes the causal factors of unsafe behaviors among special operation personnel. On this basis, the DEMATEL-ISM method is employed to analyze the causal relationships and hierarchical topology among factors, constructing a multi-level hierarchical structure model. The ISM structural model is then mapped into a BN, utilizing the BN’s diagnostic reasoning capability to identify and extract the most critical causal pathways triggering unsafe behaviors. This establishes a comprehensive analytical chain of “factor identification → causal quantification → hierarchical deconstruction → dynamic assessment.” Compared to complex network theory, which primarily describes the topological associations of factors, and SEM, which focuses on static hypothesis verification, the DEMATEL-ISM-BN method’s greatest advantage lies in its ability to explicitly represent the correlational relationships between factors, demonstrate their interaction intensities, and quantify their impact on accidents [21], thereby providing references for managing unsafe behaviors among special operation personnel.

In summary, this study aims to identify the key causes and pathways of unsafe behaviors among special operation personnel in building construction, addressing three core research questions: (1) multi-dimensional identification of causal factors for unsafe behaviors among special operation personnel based on the Ability-Motivation-Opportunity theory; (2) analysis of the causal relationship characteristics among the identified unsafe behavior factors; (3) analysis of the hierarchical characteristics of the causal factors for unsafe behaviors; and (4) analysis of the key causal pathways of unsafe behaviors. The structure of the paper is as follows: (1) Section 2 first defines special operation personnel in building construction and identifies the causal factors of their unsafe behaviors from the three dimensions of ability, motivation, and opportunity. (2) Section 3 introduces the methodological background and specific procedures of the DEMATEL-ISM-BN approach. (3) Section 4 primarily analyzes the relationship characteristics and hierarchical features of the causal factors for unsafe behaviors among special operation personnel based on the DEMATEL-ISM-BN method, and determines the most critical causal pathways for these unsafe behaviors using the fault diagnosis function of the BN. This study aims to provide building construction project managers with insights and suggestions for preventing unsafe behaviors among special operation personnel.

2. Identification of Causal Factors for Unsafe Behaviors Among Special Operation Personnel from the AMO Theory Perspective

2.1. Definition and Operational Characteristics of Special-Operation Personnel in Building Construction

The Regulations on the “Management of Special-Operation Personnel in Construction issued” by the Ministry of Housing and Urban-Rural Development of the People’s Republic of China define construction special-operation personnel as workers engaged in activities that may cause significant hazards to themselves, others, or surrounding equipment/facilities during building and municipal engineering construction [22]. Specific roles include construction electricians, scaffolders, lifting signalers, crane operators, crane installation/dismantling workers, and elevated work platform installers [22].

Special operations involve tasks requiring specialized equipment, techniques, and procedures at construction sites. Their safety and quality directly impact project quality, progress, and overall safety. Analysis of regulations such as the “Work Safety License Regulations” and “Safety Supervision Rules for Construction Lifting Machinery” reveals four defining characteristics: high technological intensity with stringent operational requirements, complex operations demanding advanced safety management, significant cost implications requiring strict budget control, and critical dependence on controlled construction environments. The aforementioned definition clarifies the core risk attributes of special operation personnel within the construction safety system. Based on policy document analysis and incorporating industry practice, this study further elaborates on and summarizes the work characteristics of special operation personnel in building construction into the following four aspects:

(1)

Technical Complexity and Professional Barriers

Special operations are not merely an extension of general labor; they rely on systematic professional theoretical knowledge and precise practical skills. For instance, construction crane operators must not only master machinery operation but also understand professional knowledge such as hoisting mechanics, stability, and signal recognition. This technical complexity creates significant professional barriers, necessitating that personnel undergo rigorous, ongoing training and assessment, thereby placing higher demands on their learning ability, comprehension, and analytical skills.

(2)

High-Risk Nature of Operations and Severity of Accident Consequences

Special operations are directly linked to the most energy-intensive and highest-risk stages of construction, such as work at height, work on live electrical systems, and lifting operations. The work process is inherently high-risk, where minor operational errors or judgment deviations (e.g., misinterpreting signals, exceeding load moment limits, insufficient safe distances) can trigger accidents like falls, electric shocks, collapses, and struck-by incidents. This implies not only a direct threat to individual lives but also the potential for mass casualties and severe negative social impact.

(3)

Strong Coupling within the Human-Machine-Environment System and High Demands for Coordination

Special operation personnel do not work in isolation but operate within a complex, dynamic “human-machine-environment” system. Taking tower crane operation as an example, it requires precise, real-time coordination among multiple roles such as the operator, signaler/rigger, and lashing personnel. Any communication breakdown or coordination failure can lead to accidents. Furthermore, operational safety is highly constrained by the status of mechanical equipment (e.g., whether limit switches are functional) and environmental conditions. This strong coupling requires personnel to possess not only excellent individual skills but also superior team communication abilities, situational awareness, and emergency coordination capabilities.

(4)

Mandatory Safety Management Requirements and Public Safety Attributes

The state has implemented mandatory qualification requirements and whole-process supervision for special operations through regulations such as the Regulations on Work Safety Licensing and the Regulations on the Safety Supervision of Construction Lifting Machinery. This reflects that their work concerns not only the interests of the enterprises themselves but also carries significant public safety attributes.

2.2. Ability-Motivation-Opportunity Theoretical Framework

The AMO model originates from behavioral research in management and psychology, initially proposed by Blumberg et al. in 1982 [20]. This model posits that individual behavior requires three simultaneous conditions: ability (A) representing individual capacity to perform tasks or exhibit behaviors, motivation (M) referring to psychological inclination driving behavioral choices, and opportunity (O) denoting external factors enabling or constraining behavior. The interaction among these elements collectively drives behavioral outcomes [23]. Due to the stability and predictive power of the AMO theoretical model, it is not only applicable to the explanation of behavior, but also provides an effective framework for analyzing the dynamics of behavior [24]. Therefore, by utilizing AMO theory to deconstruct the three-dimensional interactive mechanism of motivation-opportunity-ability, we can systematically summarize the causal factors of unsafe behavior among special operation personnel in housing construction, breaking through the traditional attribution model of influencing factors (“people-objects-environment-management”), and providing a basis for the behavioral management of special operation personnel in housing construction.

2.3. Determination of Causal Factors for Unsafe Behaviors

This study systematically analyzes the causes of unsafe behaviors among special operations personnel based on the AMO theoretical framework. The opportunity dimension, as an external factor influencing special operations, is examined across three levels: physical and environmental opportunities [25], organizational and managerial opportunities [26,27], and social and cultural opportunities [28,29]. The ability dimension serves as the foundation for special operations, encompassing safety awareness [30], job competency [31], and teamwork skills [32]. Based on this foundation, a combination of literature review, case analysis, and expert consultation was employed to systematically organize and summarize the contributing factors to unsafe behaviors among special operations personnel in building construction.

First, 126 accident investigation reports on special operation incidents in building construction in China between 2022 and 2024 were screened and compiled from publicly available data sources, including the accident notification platform of the Ministry of Housing and Urban-Rural Development and provincial emergency management departments. By reviewing the accident processes and analyzing the direct and indirect causes, causal factors of unsafe behaviors were extracted. To concretely illustrate the extraction process from cases to causal factors, the ‘September 25 Scaffolding Collapse Accident in Juxian County, Shandong Province’ is analyzed as an example. The investigation revealed that, within the capability dimension, the scaffolding erectors worked without valid licenses, failed to master the requirements of the specialized construction plan, possessed insufficient safety knowledge reserves, and demonstrated poor hazard identification skills—failing to detect critical risks such as an unstable foundation. Within the motivation dimension, to accelerate the construction schedule, on-site managers tacitly condoned workers’ risk-taking tendencies and forcibly proceeded to the next phase without conducting the required sectional inspection and acceptance. Within the opportunity dimension, the on-site supervision mechanism was weak: safety officers did not verify the qualifications of the special operation personnel; the company exhibited insufficient safety investment, using non-conforming scaffolding components; and safety education and training were inadequate, failing to organize specialized safety and technical briefings. This case clearly demonstrates that the occurrence of unsafe behaviors often results from the synergistic failure of multiple factors across all three AMO dimensions. It particularly highlights the systemic interconnections among factors such as unlicensed operation, supervision gaps, and inadequate safety investment in special operations, thereby providing a foundation for the subsequent systematic identification and screening of causal factors.

In this research, the keyword search method was used to identify the key factors influencing Unsafe Behaviour of Special personnel in Building Construction by reviewing the relevant literature on authoritative data platforms such as the CNKI and Web of Science were searched using keywords like “Building Construction,” “Special Operations,” and “Safety Capability” to supplement the data with relevant literature. For instance, Li used Meta-analysis, found strong correlations between safety climate/safety communication and construction workers’ safety behaviors, and moderate correlations between safety environment/safety supervision and safety behaviors [33]. Cheng based on the SHEL model and factor analysis, concluded that operational skills are a primary influencing factor of safety capability, and that risk decision-making ability within safety capability positively correlates with safety behavior [34,35]. Hashiguchi found that both younger and older workers’ behaviors are influenced by intrinsic and extrinsic motivation [36]. Muhammad found that work pressure and safety climate significantly impact workers across different age groups [37]. Man discovered that safety supervision and safety training can significantly positively influence construction workers’ safety behaviors [38].

Based on causal factors identified from case studies and influencing factors derived from literature, we conducted preliminary screening of unsafe behavior determinants. Subsequently, we sought expert input from professors specializing in safety management at our institution and senior construction safety management practitioners regarding factor refinement. Ultimately, 18 causal factors contributing to unsafe behaviors were systematically categorized within the Ability-Motivation-Opportunity framework. Results are presented in

Table 1. Meanings of causal factors for unsafe behaviors.

Dimension	Causal Factors		Meaning
Ability	S1	Unlicensed Operation	Operating without a special operations qualification certificate (e.g., unlicensed personnel operating construction elevators, causing a fall accident due to unfamiliarity with the emergency stop procedure).
	S2	Weak Hazard Identification Capability	Unable to identify or assess on-site hazards (e.g., scaffolders fail to notice cracks in the scaffolding foundation due to subsidence and continue to raise the scaffolding).
	S3	Inadequate Safety Operation Skills	Technical inexperience or failure to master standard operating procedures (e.g., a newly hired signalman who has not mastered the blind spot control techniques for “no entry within the swing radius of the load” and failed to clear the area when directing the hoisting of steel bars, resulting in the load swinging and colliding with workers below).
	S4	Poor Risk Assessment and Decision-Making Capability	Misjudgment in response to sudden situations (e.g., a tower crane operator insisting on lifting heavy objects in windy weather, resulting in loss of control and a fall).
	S5	Ineffective Safety Ineffective Safety Communication and Cooperation	Poor teamwork leads to miscommunication (e.g., crane operators and signalmen misunderstanding each other’s hand signals, causing the load to hit workers below).
	S6	Poor Psychological Resilience	Emotional fluctuations or tension can affect operational safety (e.g., an elevator operator distracted by a family dispute fails to pay attention to the floor where the elevator stops, causing someone to fall).
	S7	Insufficient Safety Knowledge	Lack of necessary safety regulations and professional technical knowledge (e.g., welders working in confined spaces without testing harmful gas concentrations, resulting in poisoning).
	S8	Deficient Equipment Maintenance Capability	Unfamiliarity with equipment maintenance or fault handling procedures (e.g., excavator operators failing to replace worn hydraulic hoses in a timely manner, resulting in burst pipes and loss of control during construction).
Motivation	S9	Flawed Safety Values	Insufficient awareness of the importance of safety, prioritizing efficiency and progress over safety (e.g., tower crane operators overloading the crane to speed up construction progress).
	S10	Weak Responsibility Awareness	Lack of responsibility for the safety consequences of one’s own actions (e.g., scaffolding erectors failing to check the tightness of fasteners, leading to a risk of collapse).
	S11	Risk-Taking Propensity	Habitually engaging in high-risk behavior and taking chances (such as workers at heights not wearing safety harnesses or climbing steel structures with bare hands).
Opportunity	S12	Weak Supervision Mechanism	Safety supervision is lacking or merely perfunctory (e.g., safety officers do not regularly check the operating certificates of special operations personnel and tacitly allow uncertified personnel to operate welding machines).
	S13	Inadequate Safety Equipment Allocation	Lack of necessary protective equipment or equipment failure (e.g., electricians not equipped with insulated gloves, or lift limiters not repaired when faulty).
	S14	Inadequate Safety Education and Training	Training content is vague, infrequent, or lacks specificity (e.g., newly hired crane signal operators only receive theoretical training and do not participate in on-site command practical exercises).
	S15	Poor Safety Climate	Construction sites generally disregard safety rules, creating a negative atmosphere (e.g., workers mock welders who wear protective masks as required for being “afraid of death,” and there is collective rejection of safety measures).
	S16	Unreasonable Task Load	Overworking leads to fatigue or psychological stress (e.g., crane operators working under excessive pressure misjudge the distance of the load due to fatigue, causing a collision).
	S17	Inadequate Safety Investment	Companies cut safety resources to save costs (e.g., construction companies purchase substandard safety ropes or refuse to install fall protection systems on aerial work platforms).
	S18	Safety Culture Deficiency	Companies lack long-term safety values and institutional support (e.g., management never holds safety meetings, and only fines are imposed after accidents without any corrective measures).

.

3. Research Methods

3.1. DEMATEL-ISM-BN Methodological Analysis

In safety research for building construction, the causal system of unsafe behaviors among special operation personnel is structurally complex and lacks sufficient data. Directly constructing a network model can easily lead to unreliable conclusions due to unclear structure and inaccurate parameterization. Therefore, based on systematically identifying influencing factors of unsafe behaviors using the AMO theory, this paper proposes a hybrid model integrating DEMATEL, ISM, and BN.

The DEMATEL method leverages expert knowledge to describe inter-factor relationships through directed graphs and causal diagrams, measuring impact degree, affected degree, cause degree, and center degree to determine factor importance within the system [39]. ISM employs reachability matrices to reflect intrinsic relationships among causal factors in complex systems, constructing multi-tiered hierarchical models that visually display hierarchical linkages via topological graphs [40,41]. Integrating DEMATEL’s comprehensive influence matrix with the identity matrix yields an overall influence matrix, computationally converted into ISM’s required reachability matrix. Compared to standalone ISM, DEMATEL-ISM not only reveals relationships between causal factors but also quantifies interaction intensity and impact magnitude on unsafe behaviors [42]. Bayesian Networks, probabilistic networks grounded in Bayes’ theorem, perform probabilistic inference to predict target variables through statistical analysis [43,44]. The multi-level hierarchical structure model constructed based on ISM is mapped into a BN model using the Bayesian network software GeNIe 3, enabling the extraction of key causal pathways of unsafe behaviors through posterior probability calculations performed by the diagnostic reasoning function of the BN model. The framework of the DEMATEL-ISM-BN model is shown in Figure 1.

3.2. Procedural Steps of the DEMATEL-ISM-BN Method

This study integrates the DEMATEL-ISM-BN model to analyze causal factors for special-operation personnel unsafe behaviors through the following steps:

Step (1): Select the set of influencing factors $S=\left\{S_1,S_2,\dots S_n\right\}$ ($n$ is the number of influencing factors), and construct the initial direct influence matrix $D$. Invite multiple experts to determine the interrelationships among factors, resulting in the direct influence matrix $D={\left[d_{ij}\right]}_{n\times n}$. $d_{ij}$ represents the degree of influence of factor $i$ on factor $j$, with influence levels of strong, moderate, weak, and none assigned values of 4–0, respectively. When $i=j$, the value is assigned as 0. Standardize the direct influence matrix $D$ to obtain the normalized influence matrix $B$ (with elements $b_{ij}$), calculate the comprehensive influence matrix $T$ (with elements $t_{ij}$), and identify the key influencing factors.

$$B={\left[b_{ij}\right]}_{n\times n}={\displaystyle \frac{1}{\underset{1\le i\le n}{\max }{\displaystyle \sum _{j=1}^n{d}_{ij}}}}D$$

(1)

$$T={\left[T_{ij}\right]}_{n\times n}=B+B^2+\cdots +B^n=B{\left(I-B\right)}^{-1}$$

(2)

In the formula, $I$ is the unit matrix, representing the influence of factors on themselves.

Step (2): Calculate the impact degree $f_i$, affected degree $e_i$, center degree $z_i$, and cause degree $y_i$.

$$f_i={\displaystyle \sum _{j=1}^n{t}_{ij}}$$

(3)

$$e_i={\displaystyle \sum _{j=1}^n{t}_{ji}}$$

(4)

$$z_i=f_i+e_i$$

(5)

$$y_i=f_i-e_i$$

(6)

Step (3): Calculate the overall influence matrix $H$ (with elements $h_{ij}$) by determining the threshold $\lambda$ and determining the reachability matrix $K$ (with elements $k_{ij}$) of the ISM model. Then calculate the reachable set $Q\left(S_i\right)$, the predecessor set $A\left(S_i\right)$, and the common set $C$. Repeat the calculation of and based on different values of $Q\left(S_i\right)$ and $A\left(S_i\right)$, then, update the sets in sequence to construct a multi-level hierarchical structure model.

$$H={\left[h_{ij}\right]}_{n\times n}=I+T$$

(7)

$$\lambda =\overline{t_{ij}}+\sigma$$

(8)

$$k_{ij}=\left\{\begin{array}{cc}1& h_{ij}>\lambda \\ 0& h_{ij}\le \lambda \end{array}\right.$$

(9)

$$Q\left(S_i\right)=\left\{S_j\in S,k_{ij}=1\right\}$$

(10)

$$A\left(S_i\right)=\left\{S_j\in S,k_{ji}=1\right\}$$

(11)

$$C=\left\{S_i|S_i\in S,A\left(S_i\right)\cap Q\left(S_i\right)=Q\left(S_i\right)\right\}$$

(12)

In the formula, $\overline{t_{ij}}$ represents the mean of all values in the comprehensive influence matrix $T$, and $\sigma$ is the standard deviation of the comprehensive influence matrix $T$.

4. Case Study

4.1. Causal Interdependence Analysis Based on DEMATEL

This study selects a building construction project in Xi’an as the research object. An expert panel comprising 10 specialists (including 4 project managers, 3 construction technicians, and 3 academic researchers) assessed the pairwise influence degrees between factors based on their professional judgment. All survey responses were consolidated, and the results were recorded in matrix form according to Step (1). The collected data were then normalized to establish the direct influence matrix, as presented in

Table 2. Direct Influence Matrix.

	S1	S2	S3	S4	S5	S6	S7	S8	S9	S10	S11	S12	S13	S14	S15	S16	S17	S18
S1	0	4	4	4	0	0	0	4	0	0	4	0	0	0	0	0	0	0
S2	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S3	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S4	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S5	0	0	0	4	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S6	0	0	0	4	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S7	4	3	3	3	4	0	0	3	0	0	3	0	0	0	0	0	0	0
S8	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S9	0	3	3	3	0	0	0	3	0	4	3	0	0	0	0	0	0	0
S10	0	4	4	4	0	0	0	4	0	0	4	0	0	0	0	0	0	0
S11	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S12	4	1	1	1	2	0	3	1	0	0	3	0	4	4	3	0	0	0
S13	0	0	0	2	3	0	0	0	0	0	0	0	0	0	4	0	0	0
S14	3	2	2	2	3	0	4	2	4	3	2	0	0	0	0	0	0	0
S15	0	0	0	3	4	0	0	0	0	0	0	0	0	0	0	0	0	0
S16	0	0	0	3	0	4	0	0	0	0	0	0	0	0	0	0	0	0
S17	0	0	0	1	2	0	3	1	0	0	0	0	4	0	3	0	0	0
S18	2	1	1	1	2	0	3	1	3	2	1	0	0	4	0	0	0	0

.

After obtaining the direct influence matrix, the normalized comprehensive influence matrix was derived using Equations (1) and (2). Subsequently, the impact degree, center degree, and cause degree of each factor were calculated through Equations (3)–(6). The results are presented in

Table 3. DEMATEL Analysis Results.

Factors	Impact Degree	Affected Degree	Center Degree	Cause Degree	Weight	Ranking	Factor Type
S1	0.74074	0.59224	1.33298	0.1485	0.06979	5	Cause Factor
S2	0.0000	0.92933	0.92933	−0.92933	0.04866	12	Effect Factor
S3	0.0000	0.92933	0.92933	−0.92933	0.04866	12	Effect Factor
S4	0.0000	1.789	1.789	−1.789	0.09366	1	Effect Factor
S5	0.14815	0.94579	1.09394	−0.79764	0.05727	9	Effect Factor
S6	0.14815	0.14815	0.2963	0.0000	0.01551	17	Cause Factor
S7	0.98354	0.52538	1.50892	0.45816	0.079	4	Cause Factor
S8	0.0000	0.96636	0.96636	−0.96636	0.05059	11	Effect Factor
S9	0.81344	0.30316	1.1166	0.51028	0.05846	8	Cause Factor
S10	0.74074	0.41117	1.15191	0.32957	0.06031	7	Cause Factor
S11	0.0000	1.0034	1.0034	−1.0034	0.05253	10	Effect Factor
S12	1.53365	0.0000	1.53365	1.53365	0.08029	3	Cause Factor
S13	0.39145	0.2963	0.68775	0.09515	0.03601	16	Cause Factor
S14	1.44729	0.2963	1.74359	1.15099	0.09129	2	Cause Factor
S15	0.28121	0.41427	0.69548	−0.13306	0.03641	15	Effect Factor
S16	0.28121	0.0000	0.28121	0.28121	0.01472	18	Cause Factor
S17	0.72801	0.0000	0.72801	0.72801	0.03812	14	Cause Factor
S18	1.31257	0.0000	1.31257	1.31257	0.06872	6	Cause Factor

. Factors with a cause degree > 0 are classified as cause factors, while those with cause degree ≤ 0 are categorized as effect factors. Causal factors exert significant influence on other factors in the system, whereas effect factors are more susceptible to being influenced by causal factors. Higher center degree indicates greater importance of a factor, necessitating prioritized management attention. Based on this, a causal factor relationship diagram was plotted with centrality on the horizontal axis and cause degree on the vertical axis, as shown in Figure 2.

(1)

Center Degree Analysis

The center degree reflects the overall connectivity of each contributing factor within the system. Higher values indicate that the factor occupies a more central position in the unsafe acts causation system, exhibiting stronger influence on other factors and being more influenced by them. Considering

Table 4. Reachability Matrix.

	S1	S2	S3	S4	S5	S6	S7	S8	S9	S10	S11	S12	S13	S14	S15	S16	S17	S18
S1	1	1	1	1	0	0	0	1	0	0	1	0	0	0	0	0	0	0
S2	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S3	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S4	0	0	0	1	0	0	0	0	0	0	0	0	0	0	0	0	0	0
S5	0	0	0	1	1	0	0	0	0	0	0	0	0	0	0	0	0	0
S6	0	0	0	1	0	1	0	0	0	0	0	0	0	0	0	0	0	0
S7	1	1	1	1	1	0	1	1	0	0	1	0	0	0	0	0	0	0
S8	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0	0	0	0
S9	0	1	1	1	0	0	0	1	1	1	1	0	0	0	0	0	0	0
S10	0	1	1	1	0	0	0	1	0	1	1	0	0	0	0	0	0	0
S11	0	0	0	0	0	0	0	0	0	0	1	0	0	0	0	0	0	0
S12	1	1	1	1	1	0	1	1	1	1	1	1	1	1	1	0	0	0
S13	0	0	0	1	1	0	0	0	0	0	0	0	1	0	1	0	0	0
S14	1	1	1	1	1	0	1	1	1	1	1	0	0	1	0	0	0	0
S15	0	0	0	1	1	0	0	0	0	0	0	0	0	0	1	0	0	0
S16	0	0	0	1	0	1	0	0	0	0	0	0	0	0	0	1	0	0
S17	1	1	1	1	1	0	1	1	0	0	1	0	1	0	1	0	1	0
S18	1	1	1	1	1	0	1	1	1	1	1	0	0	1	0	0	0	1

and Figure 1 comprehensively, factors such as poor risk assessment and decision-making capability (S4, 1.789), inadequate safety education training (S14, 1.744), weak supervision mechanism (S12, 1.534), insufficient safety knowledge (S7, 1.509), and safety culture deficiency (S18, 1.313) demonstrate high centrality degrees. This indicates these factors serve as critical hubs with extensive systemic influence and require urgent attention.

(2)

Cause Degree Analysis

The cause degree determines whether a factor is cause (positive values) or resultant (negative values) in the causal chain. According to Table 4 and Figure 1, factors including weak supervision mechanism (S12, +0.884), inadequate safety education and training (S14, +0.559), safety culture deficiency (S18, +0.587), and insufficient safety knowledge (S7, +0.251) are cause factors with strong driving effects. Among these, institutional and cultural deficiencies (S12 and S18) directly compromise training effectiveness and operators’ safety knowledge accumulation, representing root causes of competency gaps and cognitive biases. In contrast, risk-taking propensity (S11, −1.003), poor risk assessment and decision-making capability (S4, −1.789), weak hazard identification capability (S2, −0.929), and inadequate safety operational skills (S3, −0.909) manifest as resultant factors, acting as immediate triggers of unsafe acts.

It should be noted that certain factors, such as “Unlicensed Operation (S1)” and “Flawed Safety Values (S9),” are influenced by deeper-level factors (e.g., safety culture, supervisory mechanisms) while also directly affecting more superficial factors (e.g., risk-taking propensity, inadequate safety operation skills). Although these factors may exhibit dual roles as both causes and effects in local relationships, their classification within the system as either causal or resulting factors is determined by the positive or negative attributes of their “cause degree” according to the overall system analysis results of the DEMATEL method. This classification approach aids in identifying key variables with global driving effects at the systemic level, thereby avoiding the obscuration of governance priorities due to local complexities.

4.2. Hierarchical Causation Analysis Based on ISM

Based on the characteristics of the DEMATEL comprehensive influence matrix, this study adopts the sum of the mean and standard deviation from statistics as the basis for determining the threshold value λ [45]. The calculation yields: λ = 0.08570. Using Equations (7)–(9), the total influence matrix is established. This total influence matrix is then binarized to generate the reachability matrix K required for ISM analysis. The rule applied is as follows: if a matrix element $h_{ij}\ge \lambda$, then $K_{ij}=1$, indicating that factor $i$ influences factor $j$; otherwise, $K_{ij}=0$, indicating no influence. The resulting reachability matrix is shown in Table 4.

Finally, Equations (10)–(12) are used to determine the reachable set, the antecedent set, and their intersection. The specific steps are as follows: First, for each factor $S_i$, its reachable set $Q_{(S_i)}$ (i.e., the set of all factors reachable from $S_i$) and its antecedent set $A_{(S_i)}$ (i.e., the set of all factors that can reach $S_i$) are identified based on the reachability matrix K. Second, factors satisfying the condition $Q_{(S_i)}\cap A_{(S_i)}=Q_{(S_i)}$ are identified. These factors reside at the top level (the first hierarchy) of the system because they cannot influence factors at a higher level but can be influenced by other factors. Finally, these top-level factors are removed from the system, and the steps above are repeated to sequentially identify factors at the second level, third level, and so on, until all factors have been assigned to a hierarchy. The final hierarchical structure model constructed is shown in Figure 3.

Figure 3 reveals that the causal factor structure for unsafe behaviors among special operations personnel is relatively complex, presenting a five-tiered hierarchy. This includes 3 deep-level factors, 10 intermediate-level factors, and 5 surface-level factors, with specific findings as follows:

(1)

Deep-Level Factors

Safety Culture Deficiency (S18), Weak Supervision Mechanism (S12), and Inadequate Safety Investment (S17) reside at Level 5, constituting the most fundamental causes of unsafe behaviors. Within the hierarchical structure, these factors only emit directed arrows and do not receive any, indicating their profound influence on all other causal factors. These elements primarily belong to the organizational opportunity dimension, representing root-cause issues at the organizational and management level.

(2)

Intermediate-Level Factors

Inadequate Safety Education and Training (S14) and Inadequate Safety Equipment Allocation (S13) are positioned at Level 4. Flawed Safety Values (S9), Unreasonable Task Load (S16), Insufficient Safety Knowledge (S7), and Poor Safety Climate (S15) are located at Level 3. Weak Responsibility Awareness (S10), Poor Psychological Resilience (S6), Ineffective Safety Communication and Cooperation (S5), and Unlicensed Operation (S1) are situated at Level 2. Notably, Unlicensed Operation (S1) exerts a direct influence on multiple ability-related surface-level factors (e.g., S3, S4, S8), serving as a significant trigger for individual competence deficiencies. These intermediate factors are influenced by deep-level factors while simultaneously affecting surface-level factors, exhibiting complex inter-level relationships.

(3)

Surface-Level Factors

Risk-Taking Propensity (S11), Poor Risk Assessment and Decision-Making Capability (S4), Inadequate Safety Operation Skills (S3), Weak Hazard Identification Capability (S2), and Deficient Equipment Maintenance Capability (S8) are located at Level 1. As surface-level factors, they constitute the direct influencing factors of unsafe behaviors. Most belong to the individual capability dimension, with Risk-Taking Propensity (S11) falling under the motivation dimension and acting as a psychological trigger for unsafe behavior occurrence.

4.3. Bayesian Network-Based Causal Path Analysis

First, based on the constructed multi-level hierarchical interpretative structural model of causal factors for unsafe behaviors in special operations personnel, the structural model was converted using Bayesian network software GeNIe 3 to establish a BN structural model for unsafe behaviors. Experts defined two states for each factor: Occurrence (YES) and Non-occurrence (NO), with initial state probabilities uniformly assigned as 1/2. Using expert knowledge and historical experience, the initial probabilities of parent nodes and conditional probabilities between nodes were determined to establish Conditional Probability Tables (CPTs) reflecting dependency relationships. Due to the complexity of individual CPTs and space limitations, these tables are not elaborated here. After importing the acquired prior probabilities of parent nodes and inter-node conditional probabilities, probability updating calculations yielded the Bayesian network topology shown in Figure 4. Then, utilizing the fault diagnosis function of the constructed BN structural model for unsafe behaviors, the occurrence probability of the node “Unsafe Behavior of Special Operations Personnel” was set to 100% to infer the posterior probability of each factor, as shown in Figure 5.

Figure 4 indicates that the occurrence probability of unsafe behaviors among special operations personnel at this construction site is 29%, signifying a moderate overall risk level. through the posterior probability results in Figure 5, under the premise of unsafe behavior occurrence, the primary causal chain is identified as: safety culture deficiency (S18) → inadequate safety education training (S14) → insufficient safety knowledge (S7) → unlicensed operation (S1) → risk-taking propensity (S11). First, the posterior probability of Deficient Safety Culture as a root factor increases to 72.5%, confirming its role as a deep-seated fundamental driver within the ISM model. Inadequate Safety Education and Training, acting as a critical transmission node, sees its probability rise to 78.9%, highlighting its pivotal role in translating cultural deficiencies into specific competency gaps. At the terminus of the chain, the probability of Risk-taking Tendency reaches 85.4%, indicating that this motivational factor is the most direct behavioral manifestation under accident conditions. This causal chain reflects the profound controlling effect of institutional deficiencies on individual behavior and represents the priority pathway for improving system safety performance. The factors on this causal chain are significant contributors to the unsafe behaviors of special operation personnel and must be prioritized for targeted management in safety interventions.

Secondly, the BN diagnostic results show a high degree of consistency with the factors identified with high centrality and high cause degree in the DEMATEL analysis. In the diagnostic reasoning, the posterior probabilities of factors ranking high in centrality (e.g., S14, S12, S18, S7) all increased significantly. This indicates that the key factors identified as central in the static analysis also exhibit the highest risk exposure probabilities in dynamic accident scenarios. For instance, the posterior probability of Weak Supervisory Mechanism, a core causal factor in DEMATEL, increased substantially, further confirming its strong influence in driving the system towards an unsafe state.

Therefore, the formulation of safety management strategies must possess clear priorities and pathway specificity. Intervention measures should primarily focus on the origin of the most significant causal chain—namely, the fundamental factors within the opportunity dimension, such as reshaping the safety culture, strengthening the supervisory mechanism, and ensuring adequate safety investment. Subsequently, emphasis should be placed on cultivating the safety capability of special operation personnel by enhancing the quality of safety education and training and strictly enforcing licensed operation management to curb the emergence of unsafe motivations at the source. The BN model quantitatively demonstrates that interventions targeting upstream factors in the chain (e.g., S18) yield the greatest systemic risk reduction benefits through the causal network. Consequently, this study not only identifies the key pathways but also provides a solid scientific basis for developing efficient and precise prevention and control strategies.

5. Results Analysis and Discussion

5.1. Analysis of Results Based on the Ability, Motivation, Opportunity Three-Dimensional Framework

(1)

Ability Dimension

The ability dimension encompasses eight factors: unlicensed operation (S1), weak hazard identification capability (S2), inadequate safety operation skills (S3), poor risk assessment decision-making capability (S4), ineffective safety communication and cooperation (S5), poor psychological resilience (S6), insufficient safety knowledge (S7), and deficient equipment maintenance capability (S8). Existing studies generally acknowledge individual capability as the direct foundation of behavior. For instance, Rahman emphasized the importance of construction workers’ safety capability for industrial safety sustainability [46]. The present study not only corroborates this perspective but also further reveals the specific roles and formation pathways of capability factors within the causal network through an integrated model. DEMATEL indicates S4, S7, and S1 have high centrality, signifying strong structural relevance. Meanwhile, Poor Risk Assessment and Decision-Making Ability (S4) is identified as a typical “effect factor.” Its trainable nature aligns with the conclusions of Nykänen [47], collectively highlighting targeted directions for safety education interventions.

Furthermore, the BN model reasoning results reveal that causal factors within the capability dimension predominantly occupy relay or terminal amplification positions in the formation process of unsafe behaviors, playing crucial transmission and amplification roles. Although Poor Risk Assessment and Decision-Making Ability (S4) ranks highest in centrality, its negative cause degree indicates that it primarily functions as an outcome variable resulting from the comprehensive effects of upstream factors. This observation aligns with the ISM model analysis, where S4 is positioned at the surface level of the model structure, directly reflecting the influence of workers’ risk assessment and decision-making abilities on their behaviors. Overall, capability deficiencies not only serve as significant direct triggers of unsafe behaviors but also exert reciprocal influence on motivational factors; for instance, knowledge gaps and lack of practical skills can easily induce risk-taking tendencies.

Therefore, it is recommended to strengthen the foundational competency development of special operation personnel through systematic measures: improving the licensed operation system, enhancing the quality and coverage of professional knowledge training, and intensifying on-site practical assessments. These actions will comprehensively elevate their safety capability levels and effectively control the occurrence probability of unsafe behaviors.

(2)

Motivation Dimension

The motivation dimension encompasses three factors: flawed safety values (S9), weak responsibility awareness (S10), and risk-taking propensity (S11). DEMATEL analysis indicates that S9 and S10 have high cause degree, representing typical intrinsic motivational factors; while S3, despite having high centrality, has negative causality, suggesting it is more likely a manifestation of prior management and cognitive accumulation. BN inference results further show that in the maximum causal chain, risk-taking propensity (S11) serves as the terminal node, reflecting the psychological manifestation of systemic risk factors transmitted through successive layers. Individuals are prone to develop risk-taking tendencies when lacking effective training, insufficient knowledge, and failing to meet compliance requirements for job qualification. This suggests that motivation is jointly driven by capability and opportunity factors, constituting a crucial component of the behavioral outcome layer. This finding logically supports the view of Ghoddousi regarding the promoting effect of motivation on workers’ safety behaviors, while our model further reveals the specific causal pathway from capability to motivation [48]. Additionally, within the ISM framework, weak responsibility awareness (S10) and flawed safety values (S9) are positioned in the intermediate layer, indicating they are influenced by cultural and institutional factors in the opportunity layer while also exerting regulatory effects on behavioral outcomes in the capability layer. Therefore, in managing unsafe behaviors, efforts should not be limited to capability and institutional development. Instead, psychological cognition should be addressed through institutional guidance and role modeling to strengthen workers’ sense of responsibility and safety values, thereby establishing a behavioral baseline at the cognitive level.

(3)

Opportunity Dimension

The opportunity dimension encompasses seven factors: weak supervision mechanism (S12), deficient safety equipment allocation (S13), inadequate safety education training (S14), poor safety climate (S15), unreasonable task load (S16), inadequate safety investment (S17), and safety culture deficiency (S18). Integrated analysis confirms this dimension constitutes the core causal source. DEMATEL shows S12, S14, and S18 possess extreme centrality and positive cause degrees, representing the system’s most influential drivers; ISM positions them at the deepest level as root causes; BN’s primary causal chain originates from S18, propagating through S14 to impact individual capabilities. This evidence demonstrates organizational management, training systems, and cultural foundations are prerequisites for behavioral safety—only by embedding safety into institutional frameworks and cultural norms can effective behavioral constraints be achieved. Inadequate opportunity management induces unsafe motivation and capability defects. Critical countermeasures include strengthening supervision mechanisms, reshaping safety culture, and increasing resource allocation to enhance systemic safety efficacy. This strongly resonates with the conclusion of “safety organizational measures having a sustained impact” found by Zhang, while our model further reveals the inherent mechanism behind this sustained influence [14].

In summary, unsafe behaviors stem from the interplay of motivation, opportunity, and capability dimensions: capability deficits act as direct triggers; unsafe motivation provides intrinsic drive; and deficient opportunity constitutes institutional root causes. Effective prevention thus requires system governance integrating institutional optimization, psychological guidance, and capability enhancement.

5.2. Management Implications and Practical Recommendations

Based on the integrated analysis results of the DEMATEL-ISM-BN model, this study has identified the characteristic relationships, hierarchical structure, and key pathways among causal factors. Accordingly, this section proposes targeted insights and recommendations from both the perspectives of on-site safety management and government supervision.

On-site safety management should establish a collaborative governance system grounded in “Opportunity,” supported by “Ability,” and guided by “Motivation.” First, management measures should focus on the longest causal chain identified by the BN model (Deficient Safety Culture → Inadequate Safety Education and Training → Insufficient Safety Knowledge Reserve → Unlicensed Operation → Risk-taking Tendency), implementing source governance and process interruption. Specifically, in the Opportunity dimension, enterprises should prioritize the development of a safety culture and foster a positive safety climate by integrating systematic safety requirements into management processes and incentive mechanisms. At the same time, it is essential to strengthen the supervisory mechanism, for instance, by introducing IoT and intelligent recognition technologies to enable full-process monitoring, and establishing a safety investment guarantee mechanism to optimize the physical work environment. In the Capability dimension, the efforts in the Opportunity dimension provide institutional and environmental support for capability enhancement, with their effectiveness directly reflected in the optimization of safety education and training. By establishing an integrated “training-assessment-authorization” mechanism, instances of unlicensed operation and insufficient safety knowledge reserves can be fundamentally eliminated. Finally, in the Motivation dimension, a stable “Opportunity” foundation and solid “Capability” collectively influence workers’ psychological cognition, effectively mitigating risk-taking tendencies induced by capability deficiencies or institutional flaws. Furthermore, establishing incentive policies for safe behaviors can positively guide their safety values and sense of responsibility.

Government and industry policymakers need to commit to building a macro-level institutional environment that balances guidance and constraints. First, the institutional and standard system should be improved by revising and refining the Regulations on the Management of Special Operation Personnel in Construction. It should be explicitly stipulated that the effectiveness of corporate safety culture development, training quality assessments, and the dynamic supervision of certified personnel are incorporated into mandatory evaluation criteria. Additionally, the establishment of a “Safety Credit System for Special Operation Personnel” should be promoted, linking their behavioral records to industry access, awards, and recognitions. Second, a Guidance Manual for Safe Behaviors of Special Operation Personnel should be compiled to provide the industry with standardized and visualized best practices.

6. Conclusions

Based on the AMO (Ability-Motivation-Opportunity) theory, this study systematically identified 18 causal factors for unsafe behaviors among special operations personnel in building construction through literature review, case analysis, and expert consultation. By integrating the DEMATEL-ISM-BN methodology, this study determined the criticality, hierarchical relationships, causal pathways, and the most significant causal chain of these factors. Key conclusions are as follows:

(1)

Using the DEMATEL method to analyze the causal characteristics of each factor, the study found that within the motivation dimension, weak responsibility awareness and flawed safety values exhibit high centrality and causality, serving as prerequisites for developing risk-taking propensity. Within the opportunity dimension, weak supervision mechanisms, inadequate safety education and training, and safety culture deficiency occupy core positions in the causal network, constituting systemic drivers of unsafe behavior. Within the ability dimension, poor risk assessment and decision-making abilities, insufficient safety knowledge reserves, and unlicensed operation are key manifestations of ability deficiencies.

(2)

A multi-level hierarchical structure model was developed using the ISM method, which revealed a five-tier causal relationship behind unsafe behaviors. Causative factors in the opportunity dimension predominantly reside at the lowest level, while factors in the motivation and ability dimensions occupy the middle and upper levels. Among these, safety culture deficiency and inadequate safety investment within unsafe opportunities represent the fundamental causes of unsafe behaviors by personnel, warranting particular attention. Weak supervision mechanisms, and inadequate safety investment within unsafe opportunities are the root causes of unsafe behaviors among special operations personnel and warrant particular attention.

(3)

Through the BN model, the primary chain of unsafe behaviors among special operations personnel was identified as: safety culture deficiency (S18) → Inadequate safety education and training (S14) → Insufficient safety knowledge (S7) → Unlicensed operation (S1) → Risk-taking propensity (S11). Identifying these chain factors enables construction managers to implement targeted, sequential behavioral controls for special operations personnel and develop scientifically grounded plans, thereby reducing the blind implementation of measures.

Considering the limitation that the DEMATEL-ISM-BN integrated model constructed in this study cannot quantitatively simulate the implementation effects of different intervention measures across dynamic time dimensions, future research could employ system dynamics to dynamically simulate and quantitatively compare the effectiveness of various intervention strategies (such as enhanced training, increased supervision frequency, and raised safety investment ratios) in reducing the incidence of unsafe behaviors over short-, medium-, and long-term horizons.