Identifying Causal Relationships and Response Strategies for Stakeholder Risks in Construction Projects

Stakeholders’ interrelationship in construction projects is a complex relationship, and traditional project management practices have created numerous risks throughout the project lifecycle. Consequently, the integration of modern techniques into risk analysis and assessment has become crucial. These types of projects are surrounded by significant uncertainty, technical, organizational, financial, and environmental, if left unmanaged, leads to delays, increased costs, and declining quality [1], [2]. The construction sector is a major driver of global economic development, despite the inherent risks of its projects. According to the Project Management Institute (PMI), traditional risk management practices, the primary reason for the failure of 14% of construction projects, are shown to be ineffective [3], [4]. While modern risk management is essential to mitigate these risks, traditional risk assessment tools often overlook the complex networks and inherent ambiguities of construction risks [5]. Project stakeholders typically face a variety of risks, such as disagreements between consultants, design constraints, and revision requests. Owners sometimes face challenges related to regulatory compliance and the project's financial viability. Construction companies also suffer from shortages of skilled labor and logistical disruptions. [6]. For example, the owner may incur financial losses, and the consultant’s reputation may be damaged if the contractor is late in completing the project (e.g., due to a shortage of materials). Traditional risk matrices or probabilistic models often oversimplify these relationships, leading to suboptimal mitigation strategies [7]. Recent developments in Multi-Criteria Decision-Making (MCDM) methods promise to address these problems. Decision-Making Trial and Evaluation Laboratory (DEMATEL) is a structural modeling approach for studying cause-and-effect relationships between interacting systems [8]. Combined with fuzzy set theory to quantify linguistic uncertainty, the fuzzy DEMATEL (FD) model is very well suited for construction risk analysis because expert judgments can often be wrong [9], [10]. Although DEMATEL methodology has been applied in both supply chain risk management and occupational safety on construction sites risk factors, its application to assess multi-stakeholder risk interactions in the construction sector [11], [12]. The extensive range of potential risk factors related to stakeholder risk interaction should be examined comprehensively and systematically [13]. Published studies have demonstrated the feasibility of DEMATEL for project risk analysis [14] and the value of fuzzy logic for processing expert knowledge [15]. These findings provided a basis for the proposed model. This research focused on three main stakeholders in project execution (owners, consultants, and contractors) and their associated sub-risks. An approach integrating fuzzy logic and DEMATEL was applied to provide an analytical framework that objectively and effectively addressed the ambiguity in the causal relationships among the three stakeholders, thereby supporting risk reduction and mitigation. Recent work has demonstrated the value of fuzzy MCDM in construction [16], [17]. However, the published studies did not deal with risk interactions among project owners, consultants, and contractors, having taken into consideration inter-relationship. This research bridged that gap by proposing a fuzzy causal analysis model for the three principal stakeholders in construction projects.

Despite adopting the same basic FD approach, the current work differed from previously published papers in three main issues that alter the subject of analysis and its implications. Initially, the three project stakeholders, i.e., owner, consultant, and contractor, were treated as a single 3 $\times$ 3 causal matrix node, replacing the risk type categories as the unit of analysis. The model did not place causes under stakeholders in a hierarchical sequence; instead, it illustrated risk transmission across stakeholders. Second, the expert team was purposefully chosen to be balanced, representing all three stakeholders rather than just the owner’s side, and the 32 sub-factors were classified by the responsible stakeholders rather than by risk type. Third, the analysis demonstrated a stakeholder-specific response strategy framework that connected each causal factor to the responsible party and to tangible governance/digital tools, as well as a directional risk transmission chain (owner $\rightarrow$ contractor $\rightarrow$ consultant) that was not discussed in the previous studies. These variations are summed up in Table 1. The present study was also based on an entirely new expert panel and a novel dataset collected specifically for this work: none of the 9 admissible questionnaires analyzed here were used in the earlier study, and the panel was balanced across the three stakeholder groups instead of being drawn from the owner’s side alone. Beyond the change of risk categories, the substantive difference was therefore the unit of analysis itself: the previous study ranked risk categories inside a single party, whereas the present study treated the three parties as interacting nodes and quantified the direction and strength of risk transmission between them.

By doing so, this research contributes to both academic and practical understanding. Academically, the research encouraged the use of a fuzzy hybrid methodology for MCDM in construction management, providing a general platform applicable to the analysis of multi-stakeholder systems. Practically, the research supported project managers by offering a structured methodology for visualizing risk networks, allocating resources efficiently, and improving stakeholders’ collaboration. The objectives of this research are as follows: (1) identify and classify the primary risk factors associated with owners, consultants, and contractors that should be considered through a literature review and expert opinions to validate the findings; (2) apply fuzzy relationship analysis to calculate the causal relationships between these risk factors, including determining their significance; and (3) calculate the weights of primary criteria based on their relative significance compared to secondary criteria. Ultimately, the research aims to bridge a gap by improving targeted decisions through the development of a risk analysis theory related to construction projects and proposing an integrated quantitative approach that is closer to reality in the construction process, taking into account the dynamic interactive relationship between the main parties involved in this process: owner, consultant, and contractor.

This section presents the research questions, the procedure for identifying owner, consultant, and contractor, risk factors from the literature and screening them with an expert panel, as well as the analytical method. It then describes the FD procedure used to determine the causal relationships among these factors. The technique of FD is prominent and the standard steps are presented with their references. The complete set of intermediate matrices is narrated in Section 3 so that the analysis could be reproduced from this paper. The application-specific details are also described to enhance replicability.

To address the research gaps, three research questions (RQs) were formulated:

RQ1: What risk variables from the owner, consultant, and contractor operate as upstream generators of risk for the other parties and which variables act largely as downstream receivers of risk communicated from them?

RQ2: Which risk factors carry the greatest combined significance and impact on overall project performance, and how should this ranking guide the allocation of mitigation resources across stakeholders?

RQ3: How are the identification and efficacy of suggested response methods impacted by the interconnections and causal links between various stakeholder risks?

The population of the study included twelve experts who were working on construction projects. According to measures of risk factors, the classical procedure and FD approach were employed to obtain an overall view by means of which these study points were established to categorize risk factors as main or subsidiary factors for analyzing internal causes among main and other specific criteria (weight valuation). The suggested procedure for analyzing and evaluating the primary sub-factors in the model of evaluating risk factors using the FD analysis method is depicted in Figure 1.

Twelve experts with more than fifteen-year of professional experience in risk management or construction project management were recruited through purposive sampling. The experts had worked in the roles of owner, consultant, or contractor in the private or public sector with high educational degree. They completed a pairwise-comparison questionnaire with full admissibility checking. The returned questionnaires were checked for eligibility, completeness, and internal consistency. Entries validity means a clear and single term from five-point scale, while the completeness purpose targeted one linguistic term from direct-relation matrix in each pairwise cell. Internal consistency required the choice of ‘No influence’ to be placed in the diagonal cells for the non reversed row and column roles matrix in order to confirm parties not influencing their choice in the matrix.

Three returned questionnaires were excluded because they did not meet the eligibility criteria; the analysis therefore used only nine eligible responses.

Reducing delays, cost overruns, and subpar quality in construction projects require effective risk management. Conventional approaches to risk assessment frequently depend on qualitative studies, which are imprecise in addressing unexplicit or interrelated risk variables. Fuzzy logic and recent developments in MCDM methodologies, such as DEMATEL method, have enhanced the measurement of intricate risk interconnections. With an emphasis on the functions of owners, consultants, and contractors as well as the use of FD, this review of the literature compiled recent research on risk factor assessment in building projects. The fragmented nature of stakeholder risk identification has been highlighted by prior research. The contractors are more susceptible to supply chain and operational disruptions, whereas owners are primarily exposed to contractual and financial risks [6]. In their capacity as intermediaries, consultants usually have to cope with faulty designs and inadequate communication [18]. However, these studies usually ignore linkages in favor of focusing on risks separately. DEMATEL is a helpful tool for mapping the causal relationships between components.

For instance, DEMATEL was preferred to evaluate risks in infrastructural projects though it was acknowledged that the linguistic challenges associated with expert judgment was unable to be completely addressed [19]. Fuzzy logic integration tackles these constraints by transforming subjective expert judgments into quantifiable truths [20]. Fuzzy sets enlarge the reliability of DEMATEL in scenarios with inadequate knowledge, especially in large-scale projects [12]. Recent study has demonstrated the value of FD in sustainability risk assessment and validated its ability to categorize components based on their impact and dependence [21]. Nevertheless, few studies have completely addressed the triple-stakeholder perspective (owner, consultant, and contractor) using this hybrid technique. The FD model outperformed traditional risk assessment methods in identifying the fundamental causes of construction industry delays [11]. The adaptability of this method was clarified by representing the dynamic interdependence of hazards in megaprojects using Bayesian networks in combination with the FD model [22], [23]. Despite recent improvements, there are still gaps in the systematic assessment of how risks propagate across stakeholder borders throughout project implementation. Table 2 represents aspects of FD methods adopted in previous studies.

Consultants are exposed to technical and contractual risks, including inefficient design, ambiguous definition of scope, and documentation errors, leading to rework and disputes [1], [27]. Contractors continuously face operational risks, such as shortage of labor, equipment malfunctions, and safety hazards, which impair productivity and increase accident-related liabilities [28], [29]. These risks collectively contribute to delays, budget overruns, and decreased quality, undermining stakeholders’ confidence and project viability. Robust contracts, proactive screening, and collaborative planning are essential for effective risk mitigation that ensure stakeholder accountability and objective alignment [30]. To do so, it is necessary to specify which criteria in the relevant sections of the contract to be applied to the contractor and to recognize that these criteria do not necessarily lead to choices that are rational for both parties, since what is most rational from the contractor’s viewpoint might be irrational from the owner’s viewpoint [6].

Secondary risk factors were identified by the researcher and past experiences in the field of engineering project management and implementation through brainstorming (BS). They were identified from various previous studies, from research recently published in peer-reviewed scientific journals, and from reports of the Ministry of Planning, in addition to the main engineers syndicate. After that, expert interviews (EI) were continuously conducted with project management experts, contractors, and consultants to identify other secondary risk factors. Finally, the secondary risk factors were classified into primary factors through interviews and by taking the opinions of experts. Accordingly, the purpose of this section is to review owner decision procedures based on his criteria, list these main decision criteria with their associated sub-criteria, and establish a compilation of related risks. In Table 3, the factors are explained along with the detection methodology or reference that some secondary/sub-criteria are based on BS techniques and EI in construction project management.

Causal linkages and interdependencies among elements of complex systems could be identified using FD, a matrix-based structural modeling method [8]. FD extends DEMATEL by transforming qualitative linguistic terms using fuzzy numbers to obtain subjective expert judgment to handle linguistic impressions ambiguity [10]. Many types of risk like delays, defects, and cost overrun risks have nonlinear intercorrelation and depend on dynamic variables including financial limitations, interruptions of supply chain, and design problems that make FD capability valuable in dealing with these issues [5]. DEMATEL supports important risk prioritization (e.g., contractual uncertainty, staff shortage) and enhancement of focused mitigation measures under a condition of separating driving factors from dependent ones [46]. The soundness of judgment under application uncertainty like sustainable project success factors assessment could be integrated by FD model [21]. FD integration with complementary modelling techniques, like system dynamics, could be supporting resource allocation and harmonizing strategic goals with project management decisions [25]. It is compatible to handle the complexity of contemporary construction projects with both flexible, complementary qualitative methods and rigorous quantitative methods [47].

There were two primary parts in the questionnaire. As described below, the goal of this data analysis phase facilitates easy understanding of the questionnaire results.

Before the implementation of this method, building project risk criteria and sub-criteria were extracted from the viewpoint of nine experts in the employer, public, and private sectors. Twelve professionals took part in the survey, but the survey forms of three experts were excluded or cancelled due to incomplete responses, inaccuracies, or lack of clarity in their answers as described in Section 2.2. Therefore, nine experts were selected for the survey, and six were men while three were women. Table 4 presents the general information of the expert sample that gives a considerable description of it. The subjects were recruited by purposive sampling. The experts included in this study had different education levels, more than 15 years of professional experience, and similar professional responsibilities. The aggregated direct-relation matrix generated based on expert judgments is shown in Table 5.

An aggregated individual assessment process was performed by calculating each expert fuzzy linguistic arithmetic mean of each pair of criteria (as shown in step 1 below). All nine experts have an equal weight with a balanced combination of three members for each role (owner, consultant, and contractor) as shown in Table 4.

To achieve the failure diagnosis model, the researcher compared several major criteria with a series of key steps as illustrated below.

Step 1. Generate fuzzy direct-relation matrix: based on the $n$-criterion relationship model, the first step is to obtain an $n \times n$ matrix. The scalar product between row and column in this matrix could be formulated as a fuzzy number. If the judgments of multiple experts are employed, everybody has to fill out the matrix. The direct-relation matrix $z$ is generated by taking the arithmetic average of all experts’ opinions [10]. The language phrases accessible to the experts and the triangular fuzzy number allocated to each of them are shown in Table 6.

where, $z$ represents the initial fuzzy direct-relation matrix of size $n \times n$.

Step 2. Normalize the fuzzy direct-relation matrix: the normalized fuzzy direct-relation matrix $\widetilde{X}=[\widetilde{x}_{ij}]$ was obtained using Eqs. (2)–(3), and the results are shown in Table 7.

where,

where:

$\widetilde{z}_{ij}$ denotes the fuzzy influence of factor $i$ on factor $j$, expressed as a triangular fuzzy number (TFN) $(l_{ij},m_{ij},u_{ij})$;

$n$ is the total number of risk factors ($i,j\in\{1,2,\ldots,n\}$);

$x_{ij}$ is the normalized fuzzy direct relation between factor $i$ and factor $j$;

$r$ serves as the normalization constant, calculated as the maximum sums of the upper bounds;

$u_{ij}$ is to ensure all fuzzy values remain within the $[ 0,1]$ interval; and

$l_{ij}$, $m_{ij}$, and $u_{ij}$ represent the respective lower, middle, and upper bounds of the triangular fuzzy numbers in the direct-relation matrix.

Step 3. Calculate the fuzzy total-relation matrix: the fuzzy total-relation matrix could be calculated by the following formula:

where, $\widetilde{T}$ signifies the fuzzy total-relation matrix; $I$ is the identity matrix of dimension $n\times n$; and $k$ represents the power index denoting the sequence of direct and indirect influences ($k\rightarrow\infty$).

$\widetilde{x}^{1}$, $\widetilde{x}^{2}$, …, $\widetilde{x}^{k}$ are the consecutive powers of the normalized fuzzy direct-relation matrix $\widetilde{x}$ (see Table 7), which reflects the indirect influence channels of length 1, 2, …, $k$ accordingly. Their fuzzy sum ($\oplus$) converges to the fuzzy total-relation matrix $\widetilde{T}$ as $k \rightarrow \infty$. The identity matrix mentioned above was utilized in Eqs. (5)–(7) below, not in Eq. (4) itself.

The operator $\oplus$ denotes fuzzy addition for triangular fuzzy numbers. If each element of the fuzzy total-relation matrix is expressed as , it can be calculated as follows:

The double-prime notation used in $l_{ij}^{\prime\prime}$, $m_{ij}^{\prime\prime}$, and $u_{ij}^{\prime\prime}$ from Eqs. (5)–(7) distinguishes the boundary values of the total-relation matrix $\widetilde{T}$ from those of the initial or normalized matrices. The crisp $n\times n$ matrices $x_l$, $x_m$, and $x_u$ in Eqs. (5)–(7) were obtained by aggregating the lower ($l$), middle ($m$), and upper ($u$) bounds of each element of the normalized fuzzy direct-relation matrix $\widetilde{X}$ (Table 7), respectively. Each value was then applied to the matrix inverse $(I-x)^{-1}$. The elements of the fuzzy total-relation matrix were denoted in three ways throughout this derivation: $\widetilde{T}$ as the matrix name in Eq. (4), $l^{\prime\prime}/m^{\prime\prime}/u^{\prime\prime}$ with a double prime here are from Eqs. (5)–(7), and $l^{t}/m^{t}/u^{t}$ with a superscript $t$ are from Eqs. (8)–(11) below. All three were related to the same fuzzy total-relation matrix $\widetilde{T}$ (Table 8).

The normalized matrix was multiplied by the resultant matrix after the inverse was first computed and subtracted from matrix $I$, which yielded the fuzzy total-relation matrix, as shown in Table 8.

Step 4. Defuzzify into crisp values: The total-relation matrix’s crisp value has been obtained using the converting fuzzy data into crisp scores (CFCS) approach as proposed by Opricovic and Tzeng [48]. The steps of CFCS method are as follows:

In Eqs. (8)–(11), $l_{ij}^{n}$, $m_{ij}^{n}$, and $u_{ij}^{n}$ are the normalized lower, middle, and upper bounds of the total-relation element $(i,j)$; the superscript $n$ indicates normalization relative to the range $\Delta_{\min}^{\max}$ as defined in Eq. (11), which is the spread between the highest upper bound and the lowest lower bound taken across all elements of the fuzzy total-relation matrix $\widetilde{T}$ (Table 8). $l_{ij}^{t}$ and $u_{ij}^{t}$ (superscript ‘$t$’) are the equivalent unnormalized total-relation limits, which are the same values as $l^{\prime\prime}/u^{\prime\prime}$ from Eqs. (5)–(7).

Calculating the upper and lower bounds of normalized values:

In Eqs. (12)–(13), $l_{ij}^{s}$ and $u_{ij}^{s}$ are the standardized (scaled) lower and upper limits of element $(i,j)$, calculated from the normalized values $l_{ij}^{n}$, $m_{ij}^{n}$, and $u_{ij}^{n}$ specified above. The superscript $s$ represents this standardization step, which is done before final defuzzification in Eq. (14).

The output of the CFCS algorithm is crisp values. After calculating total normalized crisp values, the figures are illustrated in Table 9:

In Eq. (14), $c_{ij}$ is the final defuzzified crisp value of the total-relation element $(i,j)$ calculated from $l_{ij}^{s}$ and $u_{ij}^{s}$. A separate symbol was adopted so that the crisp value was not confused with the normalized fuzzy direct-relation value $\tilde{x}_{\mathrm{ii}}$ in Eq. (2) and Table 7. The final crisp entry of the total-relation matrix was then obtained as $T_{ij}=\min l^{t}+c_{ij}. \Delta$ and these entries are reported in Table 9.

In Eqs. (11)–(14), three aspects of CFCS were used for the generation of numerical results in Table 9. First, the range $\Delta$ of Eq. (11) was computed for each column: for each column $j$ of the fuzzy total-relation matrix, $\min l^{t}$ and $\max u^{t}$ were calculated over the entries of that column, and Eqs. (8)–(10) normalized the three bounds of every element of column $j$ in relation to the range of that column. Second, $u^{s}=u^{n}/(1+u^{n}-m^{n})$ was calculated as the standardized upper limit of Eq. (13) in the original formulation of Opricovic and Tzeng [48]; $l^{s}=m^{n}/(1+m^{n}-l^{n})$ was the lower bound calculated using Eq. (12). Third, the value on the normalized $[ 0,1]$ scale was given by Eq. (14), and the crisp item presented in Table 9 was restored by restoring the column’s original scale: $x^{*}=\min l^{t}+x_{ij}\times\Delta$. These three conventions are shown stepwise for one element in the working example below Table 9.

The following practical example exhibits the way to handle the linguistic assessment from a single expert to the final and precise value, from the owner to the consultant.

Working example: following opinions of nine experts from linguistic evaluation to the final crisp value (Owner $\rightarrow$ Consultant). To make the computational chain of Steps 1–4 completely visible, the Owner $\rightarrow$ Consultant entry is tracked through each of the transformations it experiences below.

Step 1. Linguistic aggregation and evaluation

The nine experts appraised the effect the owner had on the consultant using the five-point scale as shown in Table 6. The panel reached the consensus on “very high influence (VH)”. This was mapped to the triangular fuzzy number $(l,m,u)=(0.750,1.000,1.000)$ using the scale. The aggregated judgment was the element-wise arithmetic mean of the nine ratings, $\widetilde{z}$ (Owner $\rightarrow$ Consultant) = $(1/9)\sum_k(l_k,m_k,u_k)=(0.750,1.000,1.000)$, which is the value shown in the Owner row/Consultant column of Table 5. The three bounds were independently averaged one by one if the panel was not consensual.

Step 2. Normalization (Eqs. (2)–(3))

When the constant $r$ is normalized, it is the upper bounds, maximum summation of column and row, as shown in Table 5. The owner provides 0.000 + 1.000 + 1.000 = 2.000, that is all columns and rows maximum. Dividing the three boundaries by $r$ yields $\widetilde{x}$ (Owner $\rightarrow$ Consultant) = $(0.750/2.000,1.000/2.000,1.000/2.000)=(0.375,0.500,0.500)$, which is the equivalent item in Table 7.

Step 3. Fuzzy total-relation value (Eqs. (4)–(7))

In Table 7, form the three crisp matrices $x_l$, $x_m$, and $x_u$, each of which is adjusted by $x(I-x)^{-1}$. For the Owner $\rightarrow$ Consultant position, this gives $\widetilde{t}$ (Owner $\rightarrow$ Consultant) = $(0.571,1.362,3.021)$ as in the entry of Table 8. The middle limit has increased from 0.500 to 1.362 since the total-relation matrix also contained the indirect impact of the owner on the consultant, which is conveyed via the contractor.

Step 4. Defuzzification (Eqs. (8)–(14))

In the Consultant column of Table 8, the lowest lower limit is $\min l^{t}$ = 0.121 (Consultant $\rightarrow$ Consultant) and the biggest upper bound is $\max u^{t}$ = 3.021 (Owner $\rightarrow$ Consultant). Therefore, using Eq. (11) $\Delta$ = 3.021 $-$ 0.121 = 2.900, the normalized limits from Eqs. (8)–(10) are $l^{n}$ = (0.571 $-$ 0.121)/2.900 = 0.155, $m^{n}$ = (1.362 $-$ 0.121)/2.900 = 0.428 and $u^{n}$ = (3.021 $-$ 0.121)/2.900 = 1.000. The standardized boundaries are given by Eqs. (12)–(13) as $l^{s}$ = 0.428/(1 + 0.428 $-$ 0.155) = 0.336 and $u^{s}$ = 1.000/(1 + 1.000 $-$ 0.428) = 0.636. Eq. (14) then gives $c_{ij}$ = [0.336 $\times$ (1 $-$ 0.336) + 0.636 $\times$ 0.636]/(1 $-$ 0.336 + 0.636) = 0.628/1.300 = 0.48. Restoring the column’s original scale gives $T_{ij}$ = $\min l^{t}+c_{ij}\Delta$ = 0.121 + 0.483 $\times$ 2.900 = 1.521, which is precisely the crisp value for Owner $\rightarrow$ Consultant reported in Table 9.

Use of the value in the rest of the analysis. The value 1.521 is above the threshold of 1.173 as defined in Step 5, so the Owner $\rightarrow$ Consultant link is in the network relationship map. The same value is in the Owner row sum $D$ = 1.110 + 1.521 + 1.451 = 4.083 and the Consultant column sum $R$ = 1.521 + 0.869 + 1.320 = 3.710. This gives the Cause/Effect classification. The remaining cells of the matrix were each subjected to precisely the same five steps.

Step 5. Set threshold value When creating the internal relations matrix, a threshold parameter must be defined to discriminate between different relationships. As a result, a network relationship map (NRM) was made, and incomplete relations were disregarded. NRM included relations, whose matrix $T$-values were greater than a particular threshold. The average value of the $T$ matrix could be adopted to determine the threshold for relations. Given a threshold intensity, without considering the previously described causal relation, the elements in the $T$ matrix whose values were less than the threshold intensity were set to zero. The study’s cutoff value was 1.173. The aforementioned causal relationship was not taken into consideration since all items in matrix $T$ below 1.173 were set to zero.

Threshold sensitivity: Since $D$ and $R$ are the row/column sums of the complete and un-thresholded total-relation matrix (Table 9), the Cause/Effect classification of each stakeholder is mathematically independent of the threshold value: Owner ($D - R$ = +0.739), Contractor ($D - R$ = +0.108), and Consultant ($D - R$ = –0.84) all have the same sign and therefore the same identification at every threshold. What the threshold does decide is which of the six cross-party linkages are shown in the network relationship map. Table 10 sweeps this criterion across a range that includes the mean-based cutoff employed in our investigation (1.173).

The sweep validates the robustness of the presented structure. The two owner-driven links survive every cutoff from 0.90 to 1.45. The consultant has no outbound linkages at any threshold above 1.014; therefore, its position as a pure receiver of influence is maintained throughout the entire feasible range. The map is stable and interpretable in the range approximately 0.90–1.45: below 0.90, all pairs are connected, and the diagram loses its discriminating power, while at 1.55, the cutoff exceeds the largest entry of the matrix (1.521), and no linkage survives at all, fixing the practical upper limit of the criterion. Thus, the mean-based limit of 1.173 employed in this study is well within this steady range. It is vital to specify exactly what does and does not change over the sweep. The number of cross-party links kept in the network relationship map does fluctuate with the threshold, from six at $\theta$ = 0.90 to five at $\theta$ = 1.00, four at the study’s cutoff of $\theta$ = 1.173, three at $\theta$ = 1.30, two at $\theta$ = 1.45, and none at $\theta$ = 1.55. Conversely, the cause/effect classification for the three stakeholders is the same for all thresholds tested, with Owner and Contractor as causes and Consultant as an effect. This is because $D$ and $R$ are derived from the full and un-thresholded crisp matrix in Table 9 and therefore are not affected by the cutoff. The interpretations given in the next sections are based on that categorization and on the four links maintained at $\theta$ = 1.173, and none of them is changed elsewhere throughout the stable range investigated.

Across this spectrum, there are two solid results. First, the Owner $\rightarrow$ Consultant and Owner $\rightarrow$ Contractor linkages persist at all thresholds up to 1.45, representing the owner’s position as the major upstream driver, yet not an artifact of the particular cutoff selected. Second, the Consultant never has a surviving outgoing link once the threshold exceeds 1.014; its two outgoing values (Consultant $\rightarrow$ Owner = 0.98, Consultant $\rightarrow$ Contractor = 1.014) are the two smallest of the six off-diagonal entries so the consultant’s status as a pure recipient is stable and well beyond the study’s chosen cutoff. It only breaks down at implausibly permissive thresholds below 1.0. At the main-category level, there are no factors that the transition between cause and effect at any threshold evaluated. The crisp total-relation matrix after applying the threshold $\theta$ = 1.173 is presented in Table 11.

Step 6. Final output: Calculate the row sums ($D$) and column sums ($R$) of the crisp total-relation matrix $T$ using the following formulas.

In Eq. (15)–(16), the sums of row and column influences are defined:

$D$ (Degree of Dispatching) represents the sum of the $i$th row of the crisp total-relation matrix $T$, indicating the total influence factor $i$ exerts on other factors.

$R$ (Degree of Receiving) represents the sum of the $j$th column of the crisp total-relation matrix $T$, indicating the total influence factor $i$ receives from other factors.

$T_{ij}$ refers to the crisp value of the total relation between factor $i$ and factor $j$ from the matrix $T$ obtained in Step 4.

The overall importance of factor $i$ in the system is $D + R$ and net costs associated with factor $i$’s presence in the system are given by $D - R$. These values can then be computed based on $D$ and $R$. The final result is shown in Table 12 below.

A model of the salient relationships is shown in Figure 2. This can also be represented as a diagram where the values of prominence (a factor’s overall importance, $D + R$) are laid off as abscissa and those of relation (net cause/effect value of a factor, $D - R$) as the ordinate. Each factor’s position and relative interaction with a coordinate point ($D + R$, $D - R$) in the Minkowski coordinates are determined by positioning the factors in those coordinates.

Step 7. Interpret the results: As shown in Table 12 and Figure 2, the horizontal coordinate ($D + R$) reflects the significance of each factor in the overall system and can be used to evaluate the factors. That is, ($D + R$) represents factor $i$’s contribution to the total system as well as the contributions of other system factors to this factor. According to priority, the owner is in first order, and the contractor & consultant are the next rank.

The vertical/linear vector ($D - R$) signifies how strong an impact a factor has on the system. Positive $D - R$ is usually called the cause in general, and the negative $D - R$ effect. For this study, both owners and contractors are seen as independent variables; Consultants are treated as dependent variables.

From an administrative perspective, the data for Table 12 and Figure 2 were interpreted as follows:

• Owner’s Risk as a Main Causal Factor ($D$ = 4.083, $D - R$ = 0.739)

This is due to the owner’s authority to finance, modify the scope of work, and grant contractual approvals, placing them at the top of the project’s decision-making hierarchy. According to the study’s second hypothesis, owner’s risk (such as financial delays) is known as ‘project-stream root’, and its consequences are transmitted through the flow of information and decisions to the consultant (via modifications to design requirements) and then to the contractor (via modifications to execution). This explains why the owner’s $D$ value is higher compared to other parties.

• Consultant’s Risk as a Mainly Influenced Factor ($D - R$ = -0.84)

The Consultant does not have the authority to make independent decisions regarding the project scope or financing; rather, the consultant acts as an intermediary within the contractual structure, transmitting design outputs to the contractor after receiving the owner’s requirements. The consultant is more influenced by the decisions of the other two parties ($R$ = 3.71) than it influences them, due to its intermediary position in the flow of information.

The exact cross-party relationships are explicit in Table 9 (the crisp total-relation matrix). After thresholding, four directed linkages are Owner $\rightarrow$ Consultant (1.521), Owner $\rightarrow$ Contractor (1.451), Contractor $\rightarrow$ Consultant (1.320) and Contractor $\rightarrow$ Owner (1.253). Consultant $\rightarrow$ Owner and Consultant $\rightarrow$ Contractor turn into zero. The asymmetry is important to the research conclusion: although the consultant receives risk from the Owner and Contractor, the consultant is not solely ‘an effect’ in isolation, while no meaningful risk is communicated back to either stakeholder. Each stakeholder risk cannot be assessed by single stakeholder and in a-three party model with independently running for each party. A risk assessment ranks cause and effects only among the sub-factors associated with each stakeholder. The model cannot generate the relation intensity an Owner $\rightarrow$ Consultant or Contractor $\rightarrow$ Consultant that cannot be defined unless the three stakeholders are placed as interacted nodes in direct-relation matrix. The structure impact straightforward practically that consultant’s internal processes are symptom treated and targeted as mitigation resources. The exposure of consultant risk is driven by upstream choices. Paper contribution is transmission of diagonal pattern without consideration to stakeholder’s number. It falsifiable causal and ranked map, that justifies Owner $\rightarrow$ Contractor interface prioritization as the primary laver for system-wide risk reduction, more than consultant’s own processes. It quantifies the previously widespread but unsubstantiated assumption that problems “flow down” from owner to contractor.

The results of the FD model by sub-risk criteria can also be derived by repeated Step 1–Step 7 to analyze and evaluate such sub-risk criteria. The ultimate results of risk priority model in construction projects are shown in Tables 13, 14, and 15. For each sub-risk factor, the row sum $D_i$ and the column sum $R_i$ of the crisp total-relation matrix were calculated, and the prominence $D_i + R_i$ and the net causal impact $D_i - R_i$ were derived from them: the first describes the strength of the element’s involvement in the system, the second indicates whether the factor is a cause or an effect. As shown in Figure 3, Figure 4, and Figure 5, the final results shown in the causal relationship diagrams are as follows:

A. Owners Risk Factor (Sub-Risk)

Table 13 shows the final outcome of the owner’s risk factors in construction projects. It demonstrates the relative importance of the secondary factors by finding the final ranking of the factors, as well as determining the cause and effect of each factor.

Figure 3 shows the cause-and-effect diagram based on the primary criteria. All the causal properties of the primary criteria are illustrated, where the significant relationships between the secondary criteria are depicted.

Interpretation of the Results for Owner Risk. As illustrated and tabulated above, each factor may be assessed by the owner risk factor ranks first by measurement values in Table 12 and Figure 2, with $D + R$ = 7.427 in proportion to the other criteria. At the same time, the $D - R$ indicates that occupier risk is a source of risk in a construction project. For the second criterion, results were interpreted by combining Table 13 and Figure 3 as described here.

• The horizontal vector ($D + R$) shows the extent of significance of each criterion in the whole system. In other words, ($D + R$) is the effect of factor $i$ on the whole system and that of other system factors on it. Their importance in degrees (from the most important one) is RC1 (Financial Instability), followed by RC8, RC4, RC5, RC7, RC2, RC10, RC6, RC9 and finally RC3 is the weakest factor.

• The vertical vector ($D - R$) is the influence strength of factors described in the system. In the overall sense, a $D - R$ value $>$ 0 is indicative of a causal variable, and a $D - R$ value $<$ 0 is an effect. In the study, RC5 (Owner's Lack), RC6 (Poor Communication of Experience), RC7 (Site Readiness Issues), and causal variables, and in model as effect are considered RC1-RC2 -RC3 -RC4 –and RC8.

B. Consultant Risk Factor (Sub-Risk)

Table 14 represents the final result of the analysis of consultant’s secondary factors during the implementation of construction projects. It also shows the relative importance of each sub-factor, in addition to determining the causal relationship between all criteria. Figure 4 shows a cause-and-effect diagram based on main criteria. It shows all the causal features of the main criterion.

Interpretation of consultant risk results. Each factor could be assessed using the following criteria, as shown in Table 14 and Figure 4: With a $D + R$ value of 6.573, which indicates the relative relevance of the criterion in relation to the other criteria, consultant risk factors came in third place, according to the results displayed in Table 12 and Figure 2. However, the $D - R$ results show that the construction project has an impact on owner risk variables. Regarding the secondary criteria, the findings were interpreted as follows in light of Table 14 and Figure 4:

• The horizontal vector ($D + R$) corresponds to the importance of each factor in the overall system. Put another way, ($D + R$) represents the effect of factor $i$ on the whole system and all other sets of factors in it. In terms of importance value, highest to lowest, RC13 was (2), followed by RC15/RC17/RC14/RC11/RC18/RC19 (6), then in decreasing order of degrees, RC12 and RC16, and RC20.

• The vertical $D - R$ vector indicates the effect of a factor on the system. In other words, if $D - R$ is positive, then it is a causal variable, and when $D - R$ is negative, it should be called an effect. In the present study, RC11 (Design Inefficiencies/Flaws), Communication Insufficiency, RC14, RC15, and RC16 were treated as causal variables, and RC12, RC13, RC17, RC18, RC19, and RC20 were regarded as effect variables.

C. Contractor Risk Factor (Sub-risk)

Table 15 shows the final result of the analysis of the contractor secondary factors during the implementation of construction projects. It also shows the relative importance of each sub-factor, in addition to determining the causal relationship between all criteria. Figure 5 shows a cause-and-effect diagram based on the main criteria.

Interpretation of contractor risk factors. Based on the figure and table above, the contractor risk criterion ranking reveals the second-highest priority (as per Table 12 and Figure 2) based on $D + R$ value of 7.112, showing its importance relative to the other criteria. In contrast, the $D - R$ results show that contractor risk factors are direct risks to the construction project. Regarding the secondary criteria, considering the results presented in Table 15 and those indicated in Figure 5.

• The horizontal vector ($D + R$) shows how important each component is to the system as a whole. To put it another way, ($D + R$) shows how factor $i$ affects the system as a whole, as well as how other system factors affect it. RC23 (Contractual Disputes) was ranked 1 in terms of relevance, followed by RC25, RC21, RC28, RC26, RC27, RC24, RC31, RC22, RC29, RC30, and RC32.

• The vertical vector ($D - R$) shows how much of an impact the factor has on the system. Generally speaking, an effect is represented by a negative $D - R$ value and a causative variable by a positive $D - R$ value. In this study, RC22, RC26, RC27, RC29, RC30, RC31 and RC32 were considered as causal variables, while RC21, RC23, RC24, RC25, and RC28 were considered as effect variables.

Owner ($D - R$ = +0.739, most important reason). Has unilateral control over funding, scope definition, and contract approvals, choices made once and are then binding on the other two parties. This is why Owner risk enters the system as the least restricted and most simply causative variable.

Contractor ($D - R$ = +0.108, weak causation). Bound by the design output of the Consultant, but maintains genuine operational autonomy over means, techniques, sequencing, and subcontracting that the contract does not fully stipulate – thus the execution-side choices still create new risk rather than just propagate it.

Consultant ($D - R$ = -0.84, effect). The most restrictive contractual constraint of the three: technical output is bound to the owner’s brief and relevant codes/standards, leaving relatively little independent decision-making freedom—structurally a technical conduit between Owner requirements and Contractor execution, which is why it removes risk from both other parties without creating a comparable amount of its own.

The current study primarily presents a risk classification, and owners, consultants, and contractors are presented as interconnected subsystems within the project management system. The three points below clearly explain the risk transfer process, control procedures, and feedback mechanism.

• Owner, Consultant, and Contractor as Interconnected Subsystems: Instead of presenting the macro relationship matrix Table 9 and the natural resource management diagram as separate classifications for each party, the study presents them as representations of the relationships among the three parties. The difference between the outcome and the relationship ($D - R$) indicates the net influence flow between the subsystems, where $D$ values reflect the ‘output’ of a subsystem to the rest of the system, while $R$ values represent its ‘input’ from it.

• Risk Transfer: Table 16 (Response Strategies) provides clear examples of risk transfer between subsystems, including the transfer of financial risk to experts, the transfer of cost risk to the contractor through fixed-price contracts, and the transfer of regulatory obligations through requirements tracking systems.

• Control Procedures and Feedback Mechanisms: In addition, Table 16 lists the control and feedback mechanisms identified in the study findings. These mechanisms include periodic meetings and progress reports, digital information management platforms (such as BIM 360 and Aconex) for sharing real-time updates, and periodic design reviews at different stages (30%, 60%, and 90%) with independent audits. From a systems engineering perspective, these control and feedback mechanisms can help identify and correct deviations within one stakeholder subsystem, such as a consultant’s design delay, before they affect another subsystem, such as the contractor.

This explicitly accounts for the risk-transfer mechanism by mapping the four surviving cross-party linkages from Table 11 as directed edges across the three subsystems, together with the control/feedback checkpoints specified in Table 16. Figure 6 illustrates the interacting subsystems of the owner, consultant, and contractor.

This section includes response identification strategies collected and classified through various research sources, EI, and BS. Table 16 illustrates the response strategies to causal factors to mitigate their risks and impact on the affected factors. Note that the table includes the causal factor, the main response strategies, and an explanation of the response methodologies.

This study indicated the necessity of adopting a proactive approach to identifying and analyzing the root causes and risks of key stakeholders involved in project implementation. This study employed a framework integrating fuzzy logic and DEMATEL techniques to understand the complex causal relationships among stakeholders involved in project execution. Primary and secondary risk factors were examined, thus leading to the following conclusions.

The following direct findings were obtained from the FD analysis.

First, according to the model, the Consultant ($D - R$ = -0.84) is the main receiver of transmitted risk, while the Owner ($D - R$ = +0.739) and the Contractor ($D - R$ = +0.108) are causative generators of risk (Table 12).

Second, four cross-party transmission linkages over the significance threshold are identified by the crisp total-relation matrix (Table 11): Owner $\rightarrow$ Consultant, Owner $\rightarrow$ Contractor, Contractor $\rightarrow$ Consultant, and Contractor $\rightarrow$ Owner. The consultant does not maintain any outgoing link above the threshold.

Third, within the owner matrix, the sub-factors with the highest prominence were financial instability (RC1, $D + R$ = 5.993) and stakeholder conflicts (RC8, $D + R$ = 5.962), while poor communication (RC5, $D + R$ = 5.756) ranked fourth. Because the owner, consultant, and contractor sub-matrices were computed separately, $D + R$ values are comparable only within each stakeholder matrix and are not compared across matrices here.

Fourth, the owner-side variables classified as causal (positive $D - R$)—poor communication (RC5), site readiness issues (RC7), site condition uncertainties (RC9), and external risks (RC10) are those most closely linked to performance concerns on the consultant and contractor sides (Section 3.2).

Beyond what the model specifically evaluates, these findings have two more general implications for practice.

First, mitigation resources are likely to be used most effectively if they are focused on the Owner’s decision-making processes rather than being distributed equally among the three stakeholders because the Owner is both the strongest causal source and a common origin point for risk that reaches the other two parties.

Second, although the FD results identify communication and site-condition-related factors as net causal factors (positive $D - R$) within the owner matrix, rather than as the factors with the highest prominence. It has been discovered that in practice these effects are often exacerbated by inadequate planning and coordination; this connection is presented as an interpretive observation for practitioners, rather than a consequence of the causal analysis itself. This observation is consistent with the general project management literature rather than as a direct output of the model.

Some preventive measures could be suggested to prevent possible occupational hazards considering the result. First, emphasis should be placed on the cause group criterion as it drove the impact group criteria. It is much more challenging to prioritize the criteria of the cause group than those of the impact.

a. Enhance communication management between the parties: To guarantee transparency and cut down on decision-making delays, create digital platforms for information sharing between the Owner, Consultant, and Contractor. Organize regular coordination meetings with all parties’ representatives to go over possible dangers beforehand.

b. Enhance the consultant’s function as a successful mediator by involving them early in the planning stage to prevent specification differences. Provide consultants with explicit performance requirements, such as timely response and dispute resolution.

This study has intrinsic limitations despite the organized insights it offered. First, notwithstanding their specialization, the panel of nine experts that produced the results represents a particular rather than a general viewpoint. Second, because the analysis is conducted in a particular nation, the identified causal weights might change in other regulatory or economic contexts. Lastly, the conclusions are based on expert opinion and are not supported by empirical evidence from finished building projects. To confirm the anticipated causal relationships in practical contexts, future research should concentrate on using this paradigm in longitudinal case studies.