Abstract

Researchers have increasingly used system dynamics (SD) as a modelling tool to understand the behaviour of systems with varying degrees of dynamism and complexity. SD has had a particularly significant impact in improving system representation, modelling, and abstraction of problems within the construction domain. However, there is a lack of comprehensive systematic literature review and content analysis on application of SD in construction engineering and management (CEM). In this study, a systematic literature review and content analysis were used to investigate 213 journal articles published from 1995 through 2021, presenting applications of SD in CEM research. This included analysis of SD research in terms of contributing authors and their affiliations; identifying the major CEM research areas and patterns of SD research within those areas; study of the current focus of SD research, future trends, and potential for future research in these CEM areas; investigating the SD modelling paradigm in terms of hybridization with other modelling techniques; and a review of issues and challenges of SD modelling. This study contributes to the body of knowledge by (1) addressing the lack of a comprehensive systematic review and content analysis in the application of SD in CEM research, (2) providing construction researchers and practitioners with the state-of-the-art in SD research and application within the construction industry, and (3) assessing the potential for SD hybridization with other modelling approaches and proposing areas of future research to improve SD modelling capabilities. This study found that (1) the concept of SD was mostly used in the research areas of decision making and policy analysis, performance, and rework and change, (2) the areas of scheduling and health and safety have acquired more interest in SD relative to previous trends, and (3) researchers have the lowest interest in the research area of bidding and procurement.

1. Introduction

In construction engineering and management (CEM) research, simulation enables practitioners to understand underlying behaviours of construction systems by developing and experimenting with their computer-based representations [1]. Simulation is the process of abstracting systems to achieve a broad understanding of their behaviour. Traditional simulation approaches to solving construction problems fail to capture the intricate interdependencies between construction systems. Examples of such approaches include networking techniques such as critical path method (CPM), program evaluation and review technique (PERT), time–cost trade-off analysis, and resource levelling and allocation. In contrast, simulation enables the building of computer models capable of capturing the overall logic of different activities on a construction site, including relationships between different types of resources (e.g., material, labour, and equipment) and the project environment (e.g., market situation, and weather conditions) [2].

Simulation as a tool to investigate complex behaviours in construction was first studied in the 1960s by Teicholz [3], who introduced a link-node model to studying simple networking concepts and elucidate construction operations. In CEM research, simulation has overarching benefits because experimentation with varying scenarios enables managers to obtain reliable results and optimize processes for efficiency [1]. Some examples of application of simulation in CEM include evaluation and assessment of infrastructure, project scheduling, process modelling to measure productivity or performance, and risk management and control. There are three major paradigms in construction simulation modelling, namely, discrete event simulation (DES), agent-based modelling (ABM), and system dynamics (SD).

DES is a modelling technique used to capture systems, such as construction processes, that occur in discrete units of time [4]. ABM comprises discrete entities called agents, which have their own behaviours, characteristics, and rules of interaction. The SD modelling approach focuses on capturing the dynamic nature of systems that usually exhibit varying properties in relation to time and through multiple feedback processes, interactions, and dependencies [5]. DES allows users to interact with the model and observe the model’s changes as the simulation clock advances. ABM is a bottom–up simulation technique that uses individual agents’ defined behavioural characteristics to produce global behaviour resulting from nonlinear agent interactions [6]. SD is a top–down modelling approach that initially captures the system at a higher (macro) level to identify variables that affect the state of the system. In terms of level of abstraction, DES is modelled with low to medium abstraction, where more details are necessary to represent the system than the other two modelling approaches. ABM can incorporate different levels of detail, ranging from low abstraction with more details to high abstraction with fewer details. SD is usually modelled at higher abstraction and analyses the system at the macro level.

DES, ABM, and SD have all been used to model construction systems in CEM research. DES models are useful for performing processes-based simulation (e.g., resource allocation strategies, utilization, and waiting time) [7]. ABM is most useful for modelling emerging behaviours where the overall behaviour of the system is initially not known (e.g., crew motivation, fatigue, and congestion). However, construction projects exhibit the complexity and dynamism arising from interdependent components, nonlinearity in relationships between components, multiple feedback processes, and the need for both quantitative and qualitative data [8]. SD is an ideal modelling technique for capturing systems that are broad in details, holistic in perspective, continuous in behaviour, and qualitative and quantitative in nature [7]. In this regard, SD is capable of handling the complex and dynamic characteristics typical of most construction systems.

CEM involves “the development and application of techniques that improve organizations’ abilities to plan, structure, forecast, control, and evaluate projects in order to deliver results that meet or exceed performance objectives such as time, cost, productivity, quality, and safety” [9]. Construction projects exhibit complexity stemming from interdependencies between system components, such as human, environmental, technical, and organizational factors that affect construction processes [10]. These interdependencies also involve nonlinear relationships with multiple feedback processes that are able to change through time, which makes the overall problem of CEM system abstraction highly dynamic [8]. SD is appropriate for modelling problems that are “broad in details, holistic in perspective, continuous in behaviour, and also featuring qualitative or quantitative data” [7]. SD is widely used to solve problems with a high degree of complexity and dynamism to help policy- and decision-makers analyse different strategies, formulate policies, and improve the process of decision making [11]. Hence, SD has been a preferred approach in CEM, as it can be used to implement and understand the dynamics of complex processes that are not understood by other means [12].

The concept of SD was first introduced by Jay Forester in the mid-1950s to model complex systems [13]. Since its inception, SD has been widely applied in different fields including agriculture, economics, health care, defense, education, and engineering. Sterman [13] stated that in SD modelling, the initial step is problem articulation (boundary selection), which involves identifying key variables and their behaviours. This is followed by formulating the dynamic hypothesis, which involves identifying model boundaries, subsystems, causal loop diagrams (CLDs), delays, and stock and flow maps. The simulation is then formulated and used to test different model scenarios to eventually design and evaluate policies. Figure 1 depicts the major components of an SD model. In SD modelling, CLDs help to elicit mental models of experts, represent causal relationships, and depict important feedback loops within the system. The polarities (either positive “+” or negative “–”) denote the causal influences among system variables. A positive link implies that variables change in the same direction, while a negative link indicates that variables change in opposite directions [14]. Stocks are represented by accumulation or depletion, which result from differences between inflows and outflows at any point in time. Flows are the rates at which a stock varies over a given amount of time [13]. Delays are described by the lag between inputs and outputs and are used to model elapsed time between cause and effect, which is indicated by a double line perpendicular to the causal link in SD [14].

With the increased use of computer simulation approaches, the role of SD in capturing complex CEM systems is becoming ever more pronounced, both as a standalone modelling approach and as part of hybrid models. In this regard, researchers need a more focused insight into SD’s capabilities for capturing different CEM systems. This can be accomplished through a comprehensive study of the SD literature, such as this state-of-the-art study of SD application within different CEM research areas. Previous studies related to an SD literature review and content analysis in CEM are not exhaustive in terms of the number of papers covered per journal or range of years assessed, having only performed reviews of abstracts and studied citation records. They also classified areas of SD research within a limited set of published SD literature and lacked a focused analysis that proposes potential avenues for SD hybridization. Furthermore, the literature lacks a state-of-the-art study on SD that can guide researchers by analysing the SD research performed since 1995. This paper provides a content analysis and critical review of existing literature related to the application of SD in CEM.

The objectives of this study are to: (1) provide a comprehensive review of SD journal articles and content analysis to profile the selected articles based on researchers’ affiliations, case study projects, and geography; (2) identify CEM research areas and assess past studies and current trends in relation to the role of SD in these research areas; (3) assess the potential for SD hybridization with traditional methods, and other modelling and simulation approaches; and (4) identify modelling issues related to the use of SD in CEM.

Previous SD reviews focused mostly on SD’s application to a specific CEM area, namely, strategic management [15], supply chain management [16], or transportation [17]. Some studied the application of SD on limited aspects of project management [18]. Others studied critical review of SD research to a broader extent outside the scope of CEM [19]. Moreover, previous studies lacked a more focused and purposeful investigation of SD’s application in major CEM research areas. This paper provides a state-of-the-art content analysis and systematic literature review that covers a wide scope of CEM fields and provides researchers with a more focused insight on SD developments and trends in CEM research.

2. Methodology

After reviewing several studies that performed content analysis and literature review related to CEM (i.e., [12, 18, 2022]), this study utilized a multi-phase methodology, as shown in Figure 2 and elaborated in the following subsections.

2.1. Stage 1: Journal Selection

In the first stage, peer-reviewed journals with important impact and prominence in the field of CEM and which hosted published research works in the area of SD between 1995 and 2021 were selected. During the selection process, previous studies on journal rankings in CEM (e.g., [23]) were referenced. Journals that had a Cite Score of ≥2.0 in the 2020 Scopus ranking and journals that had a journal impact factor of ≥2.0 in the 2020 Institute for Scientific Information’s (ISI) Web of Science journal citation report were considered. Only peer-reviewed journals in the area related to SD were selected. Books, book chapters, book reviews, conference papers, editorials, forums and discussions, letters to the editor, indexes, introductions, forewords, seminar reports, briefing sheets, and comments were excluded. The following 21 journals were thus selected: Accident Analysis and Prevention (AAP), Automation in Construction (AC), Building and Environment (B&E), Canadian Journal of Civil Engineering (CJCE), Computer-Aided Journal of Civil and Infrastructure Engineering (CAJCIE), Construction Innovation (CI), Construction Management and Economics (CME), Engineering Construction and Architectural Management (ECAM), European Journal of Operational Research (EJOR), International Journal of Civil Engineering (IJCE), International Journal of Construction Management (IJCM), International Journal of Project Management (IJPM), Journal of Civil Engineering and Management (JCiEM), Journal of Computing in Civil Engineering (JCCE), Journal of Construction Engineering and Management (JCEM), Journal of Infrastructure Systems (JIS), Journal of Management in Engineering (JME), Journal of the Operational Research Society (JORS), Korean Society of Civil Engineers-Journal of Civil Engineering (KSCE-JCE), Resources, Conservation and Recycling (RCR), and System Dynamics Review (SDR).

2.2. Stage 2: Article Selection and Indexing

In the second stage, relevant articles from the selected journals were selected and indexed. Article searches were performed using the major available databases, namely, the Web of Science, Scopus, Google Scholar, and the American Society of Civil Engineers (ASCE) library. Further article search was performed in the Wiley Online Library, Taylor & Francis, Vilnius Tech, Emerald, and Science Direct databases. For a more inclusive but focused search, the keyword system dynamics was searched in entire articles across each journal. This was carried out for two reasons: first, introducing other keywords that have been alternatively used in SD articles would produce search results that are not within the context of SD. For example, the keyword dynamic modelling resulted in several articles not related to SD (e.g., ABM, fuzzy cognitive mapping, and robotics). Second, executing a topic/abstract/keyword (T/A/K) search would limit those articles that proposed an SD model but do not use the specific words in any one T/A/K search result.

The online search was performed to include articles from the advent of SD as a tool by Forester in 1956 [24]. However, the search did not produce enough relevant articles related to CEM prior to 1995, because relevant articles that may have been published within that period were not archived in the database. For example, Scopus coverage of CEM-related articles started in 1995. Therefore, article selection was restricted to include articles published in the English language, in the year range between 1995 and 2021 (inclusive), and in-press articles not yet published in 2021. After performing the initial search in the major databases (e.g., Web of Science, Scopus, and Google Scholar), further searches were performed in the database of each journal to find any missing articles and ensure completeness. As a result, 1,488 articles were downloaded and then indexed in Microsoft Excel. These results were further examined by reviewing the abstract, methodology, and summary sections of the texts to filter out articles that did not meet the predetermined inclusion criteria, which were (1) the article should specifically address the issue of utilizing SD for modelling, and (2) the article should discuss a topic in the area of CEM. The abstract, introduction, and methodology sections of each paper were then examined against these criteria. A total of 213 articles met the inclusion criteria and were selected for further analysis. The complete list of selected articles used for the content analysis and critical review is provided in Table S1 (see Supplementary Materials).

2.3. Stage 3: Content Analysis and Critical Review

After journal selection and article identification, further analysis of the selected articles was performed by studying articles that carried out a similar analysis in other related areas [11, 2527]. This content analysis included profiling the articles based on: (1) journal, year of publication, and number of authors per article, (2) university affiliation and geography of the authors, (3) project types that were considered in the articles, (4) research areas addressed in the articles, and (5) software used to model the SD problem in an article.

3. Results and Discussion

3.1. Descriptive and Content Analysis
3.1.1. Profile of Selected Articles Based on Journal Types and Year of Publication

The 213 articles selected for further analysis were profiled based on the contributing journals and year of publication. The percentage contribution of each journal to the total number of articles is shown in Figure 3. More than 50% of the articles were published in seven journals: JCEM (16%), JME (10%), ECAM (8%), IJPM (7%), CME (6%), and IJCM (6%). Figure 4 shows the yearly contribution of each journal, tallied per a five-year period. Out of all articles, 60% (128 articles) were published after 2010. Close to 40% (82 articles) of the total articles were published after 2015, of which 53% of these publications were registered in four journals: JME (18%), ECAM (16%), JCEM (10%), and JIS (9%).

3.1.2. Profile of Authors by Geography and University Affiliations

The analysis results show that of the 213 journal articles, only 7% were authored by a single author. A major proportion of the articles were authored by two authors (34%) or three authors (33%), 17% were authored by four authors, and 8% were authored by five or more authors. A scoring method (equation (1)) widely implemented in previous studies [28, 29] was also adopted to rank the most prolific contributors and identify these authors based on their university affiliations and geographical location:where n is the number of authors of an article, and i is the specific author’s order in the group of authors who participated in a given article. For example, if an article is authored by three authors, the first, second, and third author will have a score of 0.47, 0.32, and 0.21, respectively. The total score of an author is determined by adding the scores that author has from all articles. For ranking based on university scores and geography, appropriate calculations are made in a similar fashion by considering an author’s affiliation and location, respectively. Table 1 shows the ranking of the ten most prolific contributing authors along with their score, number of articles each had authored, and their respective affiliations. The ranking of contributing institutions is shown in Table 2U. The top contributing authors were predominantly from the United States of America and Asian countries. In terms of the authors’ geographical locations, the top five countries by score were: United States (54.00), United Kingdom (24.42), China (16.43), Australia (16.10), and Iran (15.36). The United States ranked first from the list of top contributing countries and scored significantly higher. This can be attributed to the significant foundation already laid in SD research, pioneered by U.S. researchers (e.g., [8]). However, when contributions by continent were considered, Asia ranked first with a total score of 72.52. This can be attributed to the increased demand in construction and the subsequent engagement in scientific approaches to construction in Asian countries (e.g., China, Iran, and Thailand). North and Central America both ranked second with a total score of 66.90. Europe scored a distant third with a total score of 32.96.

3.1.3. Profile of Projects in the Selected Articles

In this study, the types of projects presented in the selected articles were profiled in accordance with the type of construction work involved. This approach is adopted from previous studies that performed similar project type classification [11]. Table 3 presents the profile of projects in the selected articles. Of the 213 articles, 182 could be categorized under one of the following project types: Infrastructure projects, General type, Building projects, Power and energy projects, and Heavy industrial projects. For this study, articles that primarily addressed performance as a topic and/or discussed multiple metrics that are aggregated to indicate performance were selected to be analysed under the Performance category. A project was characterized as General type when no specific project type was given (e.g., design processes, analysis of project delivery systems unrelated to specific projects, and development of qualitative SD models). The remaining 31 articles either did not mention project types or discussed project types with limited number of mentions (e.g., civil air defense systems). As shown in Table 3, most of the construction project types modelled using SD in the selected articles were Infrastructure projects (34.74%), followed by General type (26.76%), and Building projects (18.31%).

The analysis indicates that SD research is heavily linked with infrastructure projects. This stems from the significance of infrastructure projects in a county’s growth, as these projects play a key role in spearheading the economic development of several economic sectors [29, 30]. Infrastructure projects also cover a wide range of construction works that are usually complex and encompass a diverse nature of project requirements [31].

3.1.4. Profile of Application Areas in the Selected Articles

In this study, previous works by Lyneis and Ford [18], Abotaleb and El-adaway [12], and Liu et al. [22] were used as a reference to examine most common application areas studied by researchers. Moreover, major construction management knowledge areas identified by the Project Management Institute [32] were also used as input. The most frequently occurring keywords and phrases were analysed to assist in identifying the focus of pertinent past, current, and future research. Consequently, the major construction application areas identified in this study are: Decision making and policy analysis; Performance; Rework and change; Scheduling; Risk and contingency; Resource management; Productivity; Cost planning, estimation, and control; Bidding and procurement; Health and safety; and Claim and contract administration. Based on these CEM research areas, 188 of the articles were categorized under one of these 11 categories. The list of selected articles categorized based on application areas are provided in Table S2 (see Supplementary Materials).

Figure 5 shows the profile of the application areas in the selected articles. The trend of these application areas since 1995 was summarized in five-year intervals for 1995–2019 and two-year intervals for 2020–2021, as shown in Figure 6.

It is important to note that intersections exist between the aforementioned application areas (e.g., effect of schedule delay on project cost), and some researchers have addressed more than one construction application area in a given article. In such cases, the research area given the most focus by the researchers was selected. Moreover, some researchers have produced articles in which multiple research areas are aggregated, such as for aggregation of cost performance, schedule performance, and client satisfaction, to model project performance. This has been given due consideration while outlining major areas in CEM research for this study.

3.1.5. Profile of Software Used in the Selected Articles

Analysis of the reviewed literature indicates that seven software packages have been used to implement SD: AnyLogic®, Dynamo™, DynaRisk, iThink®, Powersim, Systems Thinking, Experimental Learning Laboratory with Animation (STELLA), and Vensim. About 50% (107) of the articles discussed either a general model or did not mention the type of software used. Profiling was therefore performed for those remaining articles that either demonstrated or discussed the use of specific software in their models. As shown in Figure 7, Vensim was used in 55% of these articles, followed by AnyLogic® (16%), STELLA (12%), iThink® (11%), Powersim (4%), and Dynamo™ (3%). DynaRisk holds the lowest ranking with 1% frequency of usage. The trend of software use in the selected articles since 1995 was summarized in five-year intervals for 1995–2019 and two-year intervals for 2020-2021, as shown in Figure 8. Selection of software for use in SD modelling depends on several factors, such as availability and capability. For instance, Dynamo™ is no longer distributed commercially, so fewer and fewer papers are implementing it. Vensim is a relatively earlier software with discrete event functionality and simulation capabilities for the Markov chain and Monte Carlo methods. AnyLogic® is a newer software that is able to support a combination of SD, DES, and ABM, is able to perform hybrid modelling, and offers graphical user interface (GUI) for users to execute several types of standalone or hybrid simulation. Both STELLA and iThink® offer a GUI to simplify user experience and are mainly SD and DES modelling software with limited ABM capabilities.

3.2. Systematic Review

This section presents a systematic review of SD research in the main CEM application areas, first summarizing the corresponding literature in these areas. The application areas were selected by: utilizing previous knowledge from the works of Lyneis and Ford [18], Abotaleb and El-adaway [12], and Liu et al. [22]; referring to the major construction management knowledge areas identified by the Project Management Institute (PMI) [32]; and analysing the most frequently occurring keywords and phrases present in the selected articles. Next, the gaps in the current research are analysed to propose potential areas for future research.

3.2.1. Decision Making and Policy Analysis

Beyond capturing the construction system to study causal relationships between system elements and feedback mechanisms, SD can be effectively applied to analysing scenarios to devise policies and support decision making. In this regard, researchers and stakeholders have leaned towards the application side of SD and implementation of this modelling approach to devise solutions to different problems. In the literature, researchers have utilized SD to facilitate a systems-level approach to higher-level decision-making problems. The majority of early research focused on industry-level studies and infrastructure projects, including dynamic simulation of different maintenance policies for highway projects (e.g., [3335]). Studies related to decision making and policy analysis have since focused on sustainability. Yao et al. [36] proposed a SD model for evaluating the sustainability of highway infrastructure projects and exploring policy scenarios to improve poor sustainability performance areas. Xu and Coors [37] proposed an integrated approach for assessing sustainability of urban residential development, in which SD was used to quantitatively investigate and help decision-makers identify the developmental tendency of sustainability indicators. Zhang et al. [38] proposed an SD model for assessing sustainability of construction projects.

Within decision making and policy analysis research, the issue of sustainable infrastructure management has been a recurring theme. Related articles focused on studying the dynamics of maintenance and rehabilitation of highway projects for policy analysis and decision making [3941], environmental and economic impacts of infrastructure-highway projects [42], and financing of infrastructure projects [43, 44]. Furthermore, sustainability studies also used SD to explore the dynamics of causal relationships and strategies and to realize sustainability improvement programs [4547].

Future research may potentially capture more of the complexities and dynamic relationships between factors while analysing their impact on strategies. Further research on incorporating feedback delay into SD models would allow researchers to account for the delay resulting from strategy selection and strategy implementation in decision making. Moreover, producing better SD models that consider the effect of different policies on subsystems, detailed at different levels of aggregation within the model to support project and organizational-level decisions, should also be investigated. Potential to mitigate problems related to policy optimization and scenario analysis exists, which can enable decision makers to produce better solutions. Furthermore, more studies need to be performed to study the capabilities of hybrid models to capture human and social behaviours and analyse the social impacts of policies in decisions modelling.

3.2.2. Performance

Defining performance is an extensive research topic. Performance can be assessed using multiple metrics for gauging construction processes, practices, and outcomes and analysing their measurements based on previous or defined acceptable standards. Hence, performance can be defined differently based on the objective of a study [48]. Earlier studies on the applicability of SD for modelling performance focused on strategic management to enhance performance in construction organizations. Relevant research was conducted at the project and organizational levels, with more research on the latter.

At the project level, Peña-Mora et al. [49] used SD for strategic management of an earthmoving project. Park et al. [50] proposed qualitative SD model to explore and test design-build (DB) alternatives for enhancing DB performance. Ford and Bhargav [51] studied the application of flexible strategies for project management quality improvement. Ogunlana et al. [52] used SD to explore and enhance overall performance in an organization. Tang and Ogunlana [53] similarly employed SD to study and improve an organization’s performance behaviour using SD to suggest organizational performance improvement strategies.

Studies regarding performance have focused more on forecasting performance of construction projects as part of monitoring and control of projects to achieve their objectives. With the concept of strategic management as a recurring theme, these studies worked towards mitigating the effects of dynamic parameters that affect project performance. Leon et al. [54] used SD to simulate dynamic complexities between system variables and forecast project performance. They simulated intervention scenarios to improve project performance indicators, such as considering the interrelated structure and interaction of performance indices including cost, schedule, quality, profitability, safety, environment, team satisfaction, and client satisfaction. Nasir and Hadikusumo [55] used SD for performance assessment by modelling the owner-contractor relationships in construction projects. Ecem Yildiz et al. [56] used SD to develop a strategy map to manage performance in construction by assessing the impact of different strategies on aggregated performance measures. Kim et al. [57] used an SD modelling approach to assess construction project behaviour, by studying the dynamic interrelationship between the causes and effects of skilled labour shortage on construction project performance indices. Wu et al. [58] used SD to gain better understanding of labourers’ behavioural diversities and the associated impacts on project performance. Vahabi et al. [59] proposed a dynamic simulation model to evaluate the impact of project briefing clarity on the impact of project performance. Soewin and Chinda [60] developed a dynamics model of construction performance indices to examine and improve these measures in the long term. Luo et al. [61] investigated the impact of leadership dynamics on project performance by using SD to simulate the variation of leadership on the evolution of project performance. Tang et al. [62] used SD to carry out dynamic performance measurement and simulation of a public-private partnership project to construct a unified project performance measurement indicator system.

In future, more studies should be conducted to capture dynamic relationships between key performance indicators (KPIs). This can be regarded as a two-part challenge: first, to be able to include more KPI parameters within the SD model, which can better assist in representing the construction environment; and second, properly capturing the dynamic relationships between these parameters to determine overall performance. Furthermore, factors affecting performance that have not investigated in detail, such as out-of-sequence work [12], should be duly studied.

3.2.3. Rework and Change

In construction, errors leading to rework and changes usually have ripple effects impacting several project-performance measures (e.g., schedule, cost, and quality). The dynamic relationship between planned activities (e.g., established operation programme and planned site operations) and causes of rework or changes (e.g., unexpected events, errors, design changes, and omissions) makes SD research a suitable candidate for managing rework and change. The importance of studying causal relationships between rework/change and other variables within the construction system (e.g., activity, project, organizational, human, and environmental factors) [63] has inspired more research in the area of dynamic planning and construction rework, especially during the late 1990s and early 2000s [22]. In this regard, previous works have focused on determining causal structure of rework influences on construction in order to propose effective strategies for limiting rework and its impact on project performance [6365]. Others have studied different characteristics and patterns of construction changes compared with construction rework to propose model-based project management solutions [6670]. Studies have since emphasized the need to capture causations of rework and changes. Love et al. [71] proposed a systematic model to capture the dynamics of rework causes. Li and Taylor [72] studied the feedback mechanisms between undiscovered rework and project performance during design and construction phases. Similarly, causes of rework and their dynamic impacts were studied by Forcada et al. [73] for highway projects and Forcada et al. [74] for urban renewal projects.

Recent studies looked at utilizing SD to study causal relationships and simulate multiple scenarios of change management strategies in order to propose improvement solutions. Ansari [75] used SD to improve project planning by simulating change management policies for an Iranian construction project. Saad et al. [76] used SD to simulate the ripple impact of design changes in healthcare construction projects, which can be used to allow proactive decision making for better design change management. Porwal et al. [77] used SD to analyse the dynamic behaviour of system variables due to changes in the scope of works in order to minimize construction waste generation. Moreover, recent literature indicates a stagnation of research in the area of rework causation, which can be the result of using tools such as questionnaire surveys for identification and ranking of causal factors [78] and relying on the abundance of research previously performed. In this regard, there is potential for more research to identify and study the qualitative and quantitative aspects of rework causation for different construction system contexts. Moreover, there is potential to study the dynamic impact of changes on the recursive nature of rework during both the design and construction stages.

3.2.4. Scheduling

SD has been extensively applied in the area of project scheduling. Many early studies related to SD modelling in scheduling focused on the effect of scheduling delays. Williams [79] used an SD model to mitigate the effect of delay and disruption in projects by optimizing project duration extensions. Howick and Eden [80] studied delay caused by compressing large projects for earlier delivery. Similar studies on mitigating effects of delay were also performed by Peña-Mora and Li [81], Peña-Mora and Park [82], Howick [83], and Ford et al. [84]. Subsequent studies utilized SD to perform dynamic planning and study the impact of scheduling in multiple projects. Yaghootkar and Gil [85] studied the practice of schedule-driven management and behaviour of multiproject organizations to show that planned project milestones cannot easily be attained by simply focusing on schedule. Recent studies tried to quantify the impact of multiple factors on project schedule [86, 87].

In the area of scheduling, there is potential for future studies to analyse the impact of exogenous factors on delays. Multiple projects can be studied to generalize the impact of different factors affecting scheduling. Causalities between such factors (e.g., change orders and fast-tracking) can be further investigated in relation to their impact on scheduling across several project delivery methods. Moreover, there is potential for more studies to consider multiple performance measures and their dynamic interrelationships during the scheduling process.

3.2.5. Resource Management

The application of SD in the area of resource management has mostly focused on the organizational level. Park [88] used SD to propose a model-based approach for construction resource management. Laslo and Goldberg [89] studied resource allocation using SD to improve the efficiency of flow of resources in organizations. Lê and Law [90] proposed an SD model to simulate experience transfer at the organizational level. Other researchers have studied project cash flow [91], evolution of workforce skill [92] and workforce planning [93], and information and knowledge management [9496].

Potential for future research includes studying the effect of different strategies for better resource management schemes at the project and organizational levels. The dynamic impact of resource management on several measures of performance (e.g., cost, schedule, and productivity) should also be further investigated. Moreover, there is potential for integrating SD concepts into cloud computing platforms and building information modelling (BIM), which facilitate sharing of information resources and collaboration between different parties within CEM.

3.2.6. Productivity

Several researchers have used SD to study productivity problems. Earlier studies utilized SD to capture construction systems and observe the impact of one or multiple factors on productivity. Chapman [97] studied how changing key personnel impacted design productivity. Prasertrungruang and Hadikusumo [98] studied how downtime resulting from equipment failure impacted productivity. SD has since been applied to shape management strategies aiming to increase productivity. Alvanchi et al. [99] used SD modelling tool to investigate the effects of different working-hour arrangements on productivity.

Recent studies focused on simulating the construction process in order to observe in-depth interrelationships between different factors and the productivity measure. Nasirzadeh and Nojedehi [100] used SD to model the complex relationships between different factors affecting labour productivity. Researchers have hybridized SD with other modelling approaches to propose predictive models of productivity, which can also be used to improve the productivity measure. Gerami Seresht and Fayek [101] developed a fuzzy SD (FSD) predictive model for productivity of equipment-intensive activities using fuzzy logic principles to capture subjective variables within the SD model. Khanzadi et al. [102] used a hybrid SD-ABM approach to predict and improve the labour productivity measure.

Review of the literature on productivity indicates that no unified definition of productivity exists. Hence, SD models solving productivity problems are specific to problem context and the definition of productivity used in the model. In addition to capturing the impact of factors that contribute to the productivity loss, the objective of productivity improvement at the project or organizational level entails also studying the dynamic effect of factors that positively affect productivity. In this regard, future research needs propose informed solutions based on studies of the dynamic impact of best practices and their contribution to the overall improvement of productivity measures. Furthermore, opportunity exists to further investigate how some dynamic factors affect productivity using the increasingly popular approach of integrating SD with other modelling techniques. Some research potential includes further investigation into the impact of workplace congestion and worker motivation on productivity at the project and organizational levels using hybrid modelling methods such as SD-ABM.

3.2.7. Risk and Contingency

In the area of risk and contingency management, SD has been widely applied and is becoming a preferred approach, because it provides an advantage over traditional methods for risk analysis such as sensitivity analysis, expected value, Monte Carlo simulation, and decision trees [32]. These methods do not easily account for the unique characteristics of risks stemming from interdependencies between factors, feedbacks between system components, and the impact of indirect effects caused by other risk factors [5]. Thus, SD has been used mostly at the project level for risk analysis and response process. Previous studies utilized SD to quantitatively allocate risks between owners and contractors in construction projects [103] and to closely study the dynamics and impact of individual risk factors. Nasirzadeh et al. [5] used a fuzzy-based SD approach for integrated risk management process to evaluate alternative response strategies. Alasad and Motawa [104] used SD to assess demand risk in toll road projects. Wang and Yuan [105] used SD to study the dynamic risk interactions contributing to schedule delay in infrastructure projects. The effectiveness of passive and aggressive contingency management strategies for a real estate development project on timeliness, cost, and facility value was tested by Ford [106] using SD. De Marco et al. [107] also developed a dynamic contingency management model for a DB project using SD to simulate various decision-making scenarios to effectively manage the contingency budget during the project lifecycle.

Recent studies on risk and contingency focused more on identifying impacts of one or more risk factors for mitigation purposes. Siraj and Fayek [108] used SD to analyse the impacts of integrated and interacting risk and opportunity events on work package cost in order to determine work package and project contingencies using expert judgement and subjective assessment. Etemadinia and Tavakolan [109] proposed a hybrid method extending the structural systematic approach of interpretive structural modelling (ISM) method with the dynamic capabilities of SD to analyse uncertainties in the design phase of construction projects.

In future, there is potential to further investigate the combined effects of more risks and assess the dynamic interrelationships and causalities between such risks. This would enable better analysis of cumulative and concurrent impacts of risk events on multiple project objectives. Furthermore, there is potential for research in the area of risk response for critical risks, which provides a more comprehensive detail on dynamic risk management. Development of integrated SD frameworks for modelling different stages of risk management processes is another venue for further investigation. Moreover, there is potential to explore the use of user-friendly software and simplify the incorporation of SD concepts within the construction industry.

3.2.8. Health and Safety

Construction health and safety is one of the most fundamental requirements for a successful project. This stems from the ramifications of health- and safety-related incidents that may cause loss of life or property. However, most studies that utilized SD for investigating health and safety were done since 2010, so relatively recently. In this regard, SD has been utilized to study the dynamics of construction safety and safety culture at different levels (i.e., individual, project, and organization). Mohamed and Chinda [110] developed a causal model that simulates interactions between safety culture enablers and their impacts on safety goals at the organization level. Jiang et al. [111] studied individual and environmental conditions to model causations of construction workers’ unsafe behaviours. Lingard and Turner [112] proposed a CLD to improve construction workers’ mental and physical health by identifying multi-level (i.e., individual to industry-level) determinants of workers’ health and safety behaviours. Another multi-level study of safety culture affecting organizational safety performance was conducted by Qayoom and Hadikusumo [113]. Two studies published in 2021 focused more on individual safety behaviour [114] and an individual safety index [115] to understand and improve the safety climate at construction sites.

Future studies can better assess model components for capturing the safety behaviours of individuals and their interactions. This challenge is evident in that SD captures systems at the macro-level. Thus, there is promising potential for further research on hybridization of SD models with other modelling approaches, such as ABM, that are capable of capturing individual behaviours and internal interactions among individuals that lead to emerging behaviours. There is also potential to integrate SD with emerging technologies that better track and analyse safety in construction, such as computer vision [116] and exoskeletons [117, 118].

3.3. Past and Present Trends of SD Application Based on Research Application Areas

To analyse research trends, the selected articles were categorized based on the previously defined research application areas and were grouped as articles published in five-year intervals between 1995 and 2019. Publications from 2020 to 2021 were also included in the analysis, and the relatively fewer number of publications for this two-year period was taken into consideration. As illustrated in Table 4 and Figure 9, a significant increase in SD application occurred between 2012 and 2021. Figure 9 shows trends in SD-based CEM research areas. Each year range in the figure shows the article count for five-year intervals for 1995 to 2019 and a two-year interval for 2020 to 2021. The top five CEM research areas where SD was used as part of the modelling process were: Decision making and policy analysis (27%), Performance (16%), Rework and change (11%), Scheduling (8%), and Productivity (7%). The application of SD for the purpose of Decision making and policy analysis is the most discussed topic in the literature. The area of Decision making and policy analysis also has the most intersection with other research areas, because providing solutions to a problem related to another research area can be phrased as a decision-making problem (e.g., improving performance, reducing rework, and improving project schedule). The research area with the second most focus is Performance. This stems from the various ways to define performance, which can encompass the discussion of one or multiple construction metrics (i.e., performance indicators), mostly at the project or organizational levels. Rework and change ranked third, with decreasing interest shown since 2015.

Analysis of trends in the literature indicates clusters of research areas that researchers have shown interest in and those with a decreasing trend in publications. Despite the fewer number of articles, Scheduling and Health and safety have garnered more interest relative to previous periods in their respective areas. For Scheduling, researchers have capitalized on SD’s potential for modelling delays to address delay-related scheduling problems. Performance is another area of CEM research that has seen increasing publications since 1999. Conversely, Cost planning, estimation, and control, Claim and contract administration, and Bidding and procurement have received much less interest since about 2009.

3.4. Integration of SD Modelling with Other Methods
3.4.1. Integration of SD Modelling with Traditional Methods

Traditional methods have been used at different stages of CEM processes and are widely known among construction stakeholders. Examples include work breakdown structure, responsibility matrixes, Gantt charts, project network techniques such as PERT and CPM, and cost scheduling. Traditional tools are important for their use in incorporating data during input modelling. Depending on the magnitude of data for input, preferred methods can range from spreadsheets and MS Excel files to more sophisticated software such as Statistical Package for the Social Sciences (SPSS), Primavera, and MS Project.

Because traditional methods have mostly used a linear approach for system representation, SD’s ability to more efficiently and comprehensively capture dynamic CEM systems makes it ideal for integrating with these approaches for wider applicability. The role of SD in complementing traditional methods becomes more pronounced during the SD model implementation phase. SD can thus complement traditional methods to improve theory and practice by offering a strategic and tactical perspective [18] at different phases of CEM. For example, SD can be used to capture the construction system for traditional risk management practices in PMI’s Project Management Body of Knowledge [32]. It can also be used to implement different strategies related to resource allocation, scheduling, and project control for improved project management. Ansari [75] provided a comprehensive SD model that measured the impact of change management policies on scheduling, cost, quality, resources, and cash flows that also utilized traditional approaches such as CPM, Gantt chart, and resource allocation. However, little research has been conducted on integrating traditional methods, which has partly contributed to the diminished visibility of SD method with planners and decision makers in the wider construction industry. Therefore, research focused on practical applications of SD implementation with traditional methods is needed, which would allow SD to appeal to more experts within the CEM field.

3.4.2. Integration of SD Modelling with Other Modelling and Simulation Techniques

Hybridization of SD with other modelling techniques has become an increasingly preferred practice by researchers, as it enables modellers to use the potential benefits offered by the model components. In this regard, researchers have used SD to complement different simulation and modelling techniques, such as DES, ABM, and BIM.

In the context of construction research, hybridization of SD with DES was an early form of hybrid modelling approach and goes back to the early 2000s [119]. Although DES is suitable to analysing the stochastic nature of construction parameters at the tactical level, DES is not capable of modelling construction systems at the holistic level and also falls short in capturing dynamic feedback processes between system variables [7]. Hybrid SD-DES combines the sequential modelling benefits of DES with the dynamic modelling capabilities of SD. Xu et al. [120] used a hybrid SD-DES model in which DES captured construction activities’ microlevel variables, such as resource allocation and predecessor–successor relationships, and SD represented the construction environment as a macrolevel phenomenon and captured feedback relationships between different model subsystems (i.e., construction process, resource, project scope, schedule target, and project performance subsystems). Alzraiee et al. [7] used SD-DES modelling for dynamic planning in which operational-level parameters (i.e., duration and activity sequence) were modelled using DES, and the dynamics arising from interactions between the model’s variables were modelled using SD. Similarly, other researchers have used DES to capture sequential processes and other operational model parts at the microlevel and SD to capture feedback relationships and other dynamic aspects of the model at the macrolevel. Some areas of this research include productivity [121], performance [49, 122], cost estimation, planning, and control [49], claim and contract administration [123], and resource management [92].

Hybridization of SD with ABM in CEM research is a relatively new topic, following the increasing popularity of ABM within the research community. ABM is not best suited to modelling policies, investigating which processes dominate in aggregated systems, and investigating aggregated system-level dynamics [124]. Nasirzadeh et al. [125] highlighted the main limitations of SD, which include difficulty in modelling heterogeneous environments, as SD works mainly on aggregate variables, and difficulty in modelling systems that evolve through time, as the system’s structure is fixed in SD. Thus, hybrid SD-ABM uses SD to capture higher-level dynamics, complex feedback relationships, and continuous factors and ABM to model microlevel variables, complexities, and emerging behaviours arising from agent interactions as well as the spatial nature of agent behaviours [126]. Some examples include the hybrid SD-ABM by Khanzadi et al. [102], in which SD was used to simulate continuous factors affecting labour productivity and their dynamic feedback relationships and ABM was used to model congestion, which results from interactions between different agents. Nasirzadeh et al. [125] used hybrid SD-ABM to model construction workers’ safety behaviour, where SD was used to capture multiple governing feedback relationships of several continuous variables and ABM was used to capture the complexity that emerges from agent interactions.

Other hybrid modelling approaches include hybrid SD-ISM, which can be performed by identifying variables and their interrelationships in the context of the problem to be modelled using ISM to reveal the intrinsic structure of the complex system and SD to model different process subsystems and quantify dynamic relationships [109]. SD-BIM can also be performed by using BIM to utilize and simulate field data and using SD to map interactions and feedback relationships [77, 127].

3.4.3. Integration of SD Modelling with Fuzzy Logic

The fuzzy logic approach, introduced in 1965 by Zadeh [128], is an extension of classical Boolean logic to handle real-world parameters by enabling mathematical translation of linguistic variables. Fuzzy logic theory is applicable in modelling CEM problems whose variables exhibit subjectivity or vagueness and require reasoning with ambiguous, incomplete, and/or imprecise data [128].

Researchers have integrated fuzzy logic with SD to produce FSD models. Fayek [129] discussed the importance of incorporating fuzzy logic to model CEM problems and summarized the aspects of CEM problems that are best suited for fuzzy logic modelling and fuzzy hybrid techniques. Some of these aspects include: when there is a reliance on experts for decision making based on subjective information and experience; when variables are imprecise or unstructured and there is a need to capture the complex relationship between these variables; and when the need arises to facilitate experts’ decision making using linguistic terms instead of strict numerical terms. In this regard, FSD models are able to capture real-world systems with nonprobabilistic (i.e., systems with subjective variables or linguistically expressed information) and probabilistic uncertainties [130].

In the CEM literature, fuzzy logic has been used in FSD modelling for two purposes. First, in the qualitative stage of SD modelling, FSD is used to define model variables whose nature cannot be expressed using crisp values or probabilistic terms and to qualitatively define causal relationships between these variables [131], which include factors such as crew motivation, haul road condition, adequacy of maintenance program, and familiarity with new techniques. Fuzzy logic can be used with other methods to identify variables and capture causal relationships within the FSD model. Siraj and Fayek [108] and Rostamnezhad et al. [132] used expert inputs in their FSD models, and they used fuzzy decision-making trial and evaluation laboratory (FDEMATEL) to capture uncertainty and vagueness arising from human judgements. Palikhe et al. [133] used fuzzy analytical hierarchical process to identify critical factors and underlying relationships for their FSD model. Second, fuzzy logic is also used in the quantitative stage to quantify fuzzy system variables and quantitatively define causal relationships between variables. In this regard, FSD has been implemented to quantify claims [125], model productivity [101, 134, 135], and model quality management [136]. FSD models have been most common in the area of risk and contingency [5, 10, 103, 108, 137], which may be due to fuzzy logic’s ability to capture subjective uncertainties and the imprecise nature of risks.

3.5. Modelling Aspects of SD
3.5.1. Issues in Qualitative and Quantitative Modelling

The initial SD modelling step of defining the model boundary and level of aggregation is crucial to system understanding. Systematically structuring the problem to be modelled can lead to a better boundary definition and can be done using model boundary charts [14] or cognitive maps. Siraj and Fayek [108] identified model boundaries and aggregation level for risk analysis using a model boundary chart to define the model scope and define the model subsystem at the work package level. Defining model boundaries and system abstraction can also be done at higher levels. Mostafavi et al. [138] studied interdependencies between policy metrics at project, regional, and national levels for policy analysis of infrastructure systems. However, a review of the selected articles in CEM literature indicates that most studies have not discussed their process for defining model boundaries, including defining endogenous and exogenous variables and aggregation level.

Following the system understanding and problem articulation phase, the SD modelling process can be summarized as consisting of qualitative and quantitative stages. This study found that in the qualitative modelling stage, most researchers identified system variables and established the qualitative relationships between them using one or a combination of existing knowledge, literature review, and expert inputs. However, extracting knowledge from experts (e.g., using interviews) alone is insufficient and should be supplemented with other forms of data [13]. Some system variables can also be “soft” (not measurable), making it impossible to always use numerical data [139]. Moreover, the quantitative stage deals with formulating the model by building quantitative relationships between model elements and variables [101]. This is achieved by using numerical values or probability distribution functions for defining system variables and using table functions or mathematical equations to define causal relationships between system variables [13].

Construction systems whose causal relationships involve subjective variables do not have numerical metrics and are linguistically expressed. This study found that utilizing approaches such as probabilistic and analytical methods to capture these systems can be problematic owing to lack of sufficient historical data and that the use of fuzzy logic concepts has been widely utilized in these contexts [10, 11, 135]. An important aspect of fuzzy logic application is fuzzy arithmetic, which replaces classical arithmetic to perform algebraic operations involving fuzzy variables. Hence, the type of fuzzy arithmetic method selected significantly impacts the accuracy of the results; implementation of fuzzy arithmetic in the mathematical equations of FSD models can result in overestimation of uncertainty, reducing users’ ability to accurately predict system output [101]. Of the two methods for carrying out fuzzy arithmetic operations, the α-cut method and the extension principle, analysis of the published articles in CEM literature shows a lack of research in the implementation of the extension principle in fuzzy arithmetic operations.

There is a lack of a systematic method for qualitatively capturing system variables, developing stock-and-flow and causal relationships, and performing quantitative modelling. In the presence of data, relationships between system variables can be captured using artificial intelligence-based approaches [140], such as machine learning, such as artificial neural networks (ANN) and fuzzy logic, to learn system rules from historical data. When data exhibits subjectivity, fuzzy logic–based methods such as neuro-fuzzy inference systems (NFIS) [131] and data-driven fuzzy rule base systems [108] can be used to facilitate SD model development. However, the potential of these methods to capture system complexity in SD modelling is not yet fully explored. Very few articles explored the use of other methods to elicit relationships between system variables in the absence of data. Procedures in methods such as FDEMATEL can be improved by incorporating weights to account for experts’ profiles and their disparity in capabilities. The FDEMATEL methods proposed by some researchers [108, 132] and the structural equation modelling (SEM) proposed by others [61, 141] involve lengthy and more complicated algorithms and may necessitate computer tools or software for a wider audience.

3.5.2. Delays in SD

Some researchers have incorporated time-delayed response systems in their models. Alvanchi et al. [99] used a feedback delay element in their FSD model to signify the delayed effect of increased working hours to signify the adverse effect of set overtime on the productivity ratio, which occurs a week later. Prasertrungruang and Hadikusumo [142] used delay elements in their SD model to capture the time-delayed occurrence of severe equipment breakdown when quality maintenance and new equipment is provided. Delays are critical sources of dynamics in almost all systems [13], and their impacts become more pronounced in dynamic models that capture complex construction systems for the purpose of decision making. However, a review of the selected articles in CEM literature found that the use of delays in the SD models is underutilized, as few studies have incorporated delay concepts in their models.

3.5.3. Validation

This study found that a wide range of validation methods were used in several SD models. These methods can be categorized as structural and behavioural validation tests [13], performed to assess whether qualitative and quantitative models have contradicted the structure of or closely captured the real system. These tests are performed with the understanding that it is impossible to prove that a model is right [139] and that efforts are made towards building trust in the method followed during modelling [143]. In this regard, most studies have used different variations of the structural validation test including boundary adequacy, rationality of qualitative relationships or parameters, dimensional consistency, and extreme conditions. Ruiz and Guevara [47] used dimensional consistency, integration error, and anomaly tests. Nojedehi and Nasirzadeh [135] and Hou and Wang [43] used structure assessment tests, boundary adequacy, dimensional consistency, and extreme conditions. Luo et al. [61] used structural validity and dimensional consistency tests. Very few researchers performed behavioural validation tests, which can be attributed to the absence of historical data. Articles that used behaviour reproduction to assess the model’s capacity to reproduce historical data include Xu et al. [120], Qayoom and Hadikusumo [113], Li et al. [114], Luo et al. [61], and Ruiz and Guevara [47].

3.5.4. Future Trends in SD Modelling

Analysis of the selected articles on application of SD for CEM research indicates that SD modelling has transformed into different forms of hybrid SD. Such hybridization has been performed to either improve modelling capabilities featuring SD itself (i.e., improving the qualitative and quantitative modelling stages) or capture more of the problem context in CEM research; that is, to better capture CEM problems not effectively captured by SD modelling alone.

In this regard, there is potential to further explore the application of the fuzzy logic approach in SD qualitative and quantitative modelling stages in order to improve fuzzy arithmetic implementation in FSD modelling and increase the accuracy of FSD models. This can be performed by incorporating different types of fuzzy numbers (i.e., triangular, trapezoidal, and Gaussian) and experimenting with several t-norms (Yager t-norms, Hamacher t-norms, and Schweizer-Sklar t-norms) [108]. Moreover, FSD application in CEM is still limited owing to its low accuracy in capturing nonlinear and highly dimensional relationships among system variables [131]. Hence, there is potential to explore integration of FSD models with data-driven approaches such as neuro-fuzzy systems [131], which are able to better define relationships between such system variables.

Moreover, there is potential for future research to complement the modelling capabilities of SD by integrating it with other modelling approaches, which would enable development of more holistic hybrid models capable of capturing more complexities of given CEM systems. Further studies could focus on hybridizing fuzzy, SD, and ABM paradigms. This can enable modellers to quantify different types of uncertainties (i.e., probabilistic and subjective), understand the system’s governing dynamic relationships and feedback interactions, and capture complexities arising from the spatial nature of agents and the dynamic interactions between agents that give rise to emergent behaviours.

Despite the capabilities they add to modelling, hybridization approaches can add to model complexity, which will also directly impact the model validation phase. Hybrid simulation challenges owing to lack of modelling framework and absence of communication architecture between individual modelling paradigms [119] can also contribute to lagging interest of many researchers to implement different types of hybrid modelling approaches within different CEM research areas. In this regard, more work should be done to produce hybrid modelling frameworks [144] that clearly delineate the exchange of information between different modelling approaches.

SD application in CEM has mostly been confined to research purposes, owing to some underlying challenges in SD implementation. Although models are a very important part of communicating results and conclusions [145], more work can be done in communicating the modelling process to end users, because much of the learning comes from such processes. Building large models that are difficult to communicate and too complex to critically evaluate has also been a source of criticism of SD models [146]. Construction practitioners will not implement SD in their projects when they are unaware of the value of SD, which can stem from lack of knowledge about the concept, seldom use of SD in their organizations, or the misconception that SD is impractical [147].

4. Conclusions and Recommendations

In this paper, systematic review and content analysis of 213 articles obtained from 21 high-ranking peer-reviewed journals was performed to analyse the application of SD in CEM and derive directions for future research. The novelty of this study lies in its approach of covering articles spanning more than 25 years to get a comprehensive picture of SD research in CEM. The findings of this study indicate that the use of SD in the area of CEM research steadily increased from 1995 to 2021. This study used analytical and objective approaches to study research trends, contributions of authors and their affiliations, and provide a profile of CEM projects with SD applications. The main contributions of this study are (1) addressing the lack of a comprehensive systematic review and content analysis in the application of SD in CEM and (2) providing researchers and construction practitioners with the state-of-the-art in SD research and application within the construction industry. Furthermore, this study provides researchers and practitioners a focused resource on SD research because it incorporates different approaches to structuring the systematic review by defining major areas of CEM research areas and analysing the trends of SD research within those research areas. For researchers interested in the use of SD modelling in CEM, this study thus provides a comprehensive review to identify modelling issues related to the use of SD in CEM and assesses the potential for SD hybridization with other modelling paradigms.

This study profiled the available SD literature in the 21 ranked journals, and found that the top three contributors to this field were JCEM, JME and ECAM. Analysis of top contributing authors and their affiliations was also presented. The top contributing countries to SD research were found to be the United States, United Kingdom, and China. Analysis of the profile of projects for SD application shows that infrastructure projects were used most in SD model applications, which indicates the significance attributed to such types of projects by different countries. The analysis also found that a significant number of articles only provided either qualitative SD models or SD models without application on real projects. Although SD has had relative success in terms of its application to project management compared with other CEM research areas, SD’s practical application in construction management was found to be relatively less and confined to individual projects, which confirms conclusions by Lyneis and Ford [18]. This highlights the significant challenge to use SD modelling for CEM problems, stemming from either lack of historical data or reluctance from construction stakeholders to apply SD methods. In this regard, there is a need to produce more SD models that can be generalized, particularly at the organizational level.

This study identified eleven major research areas within CEM and assessed the role of SD in abstracting and modelling problems in each. SD was mainly used in the research areas of Decision making and policy analysis, Performance, and Rework and change between 1995 and 2021. Scheduling and Health and safety acquired relatively more interest among researchers between 2017 and 2021, with the number of publications in these fields increasing relative to previous years. This study also identified some major potential areas of future research in different CEM application areas, which can be used to guide researchers to further SD’s application within these eleven research areas.

A critical review of the literature also identified the possible areas of improvement regarding SD hybridization with traditional methods and other modelling approaches. Analysis of the literature indicates that more work needs to be done in integrating SD with more traditional tools, which can help facilitate a better understanding of SD among construction practitioners and increase SD’s applicability and presence across a vast spectrum of projects. There is also potential for further research in SD hybridization with other methods, especially in the areas of (1) SD-ABM modelling to capture the spatial natures of construction environments and emerging nature arising from individual interactions, and (2) SD-BIM to facilitate a more collaborative decision-making process in dynamic construction environments. Moreover, there is a potential to improve the qualitative and quantitative modelling processes in SD using modelling approaches such as machine learning, ANN, NFIS, FDEMATEL, and SEM. This study also identified the added complexity that may result from hybrid SD modelling owing to system abstraction, aggregation, and model validation. Another identified SD modelling issue was the lack much research on incorporating feedback delay in SD models.

The selected 213 articles do not fully represent all published articles in the area of SD. The analyses and results are performed for articles written in English, so they do not represent journals written in other languages. Future work will entail scientometric analysis to map journal and author co-citation networks, article citations, and research themes to better visualize analysis outputs. Future work will also identify more construction management research areas in which to refine the focus of SD research and help guide stakeholders in understanding the state-of-the-art specific to their problems.

Data Availability

The data used to support the findings of this study are included within the article and the Supplmentary Materials file.

Conflicts of Interest

The authors declare that there are no conflicts of interest regarding the publication of this paper.

Acknowledgments

The authors wished to acknowledge Dr. Phuong Nguyen for his comments and feedback, and Renata Brunner Jass for her edits of the manuscript. This study was funded by the Natural Sciences and Engineering Research Council of Canada Industrial Research Chair in Strategic Construction Modeling and Delivery under Grant no. NSERC IRCPJ 428226–15, which is held by Dr. Aminah Robinson Fayek. As a part of the University of Alberta’s Future Energy Systems research initiative, this research was made possible in part thanks to funding from the Canada First Research Excellence Fund, under Grant no. FES-T11-P01, held by Dr. Aminah Robinson Fayek.

Supplementary Materials

In Supplementary Materials, Table S1 presents the complete list of selected articles used for the content analysis and critical review, and Table S2 presents the list of selected articles categorized based on application areas. (Supplementary Materials)