The first two stages of the research are delineated in this section, and the last stage, in the Results and Recommendations section.
With regard to the reliability, it should be mentioned that several measures have been suggested to assess research reliability, which could be found in the previous literature [
30,
31]. In this study, all of the data, which gathered through each stage of research, were mechanically recorded and transcribed to provide a chance for others to assess the consistency and repeatability of the results [
31]. Additionally, multiple data sources, including previous literature and three groups of industry-related experts, and multiple research methods, including literature review, nominal group technique, and the questionnaire survey, were employed to ensure that the study’s findings would be reliable. Yin [
30] noted that meaningful parallelism of findings across multiple data sources could increase research reliability. Previous studies also emphasized that triangulation, simultaneously using multiple data sources or research methods, can go a long way toward enhancing the reliability and validity of results [
29,
32]. To this aim, in this research, previous studies were first rechecked to ensure that all critical factors were identified. Second, industry-related experts were surveyed to verify the accuracy of factor identification. Third, the opinions of another panel of experts were assessed using the nominal group technique to construct the causal relationships among the factors. Finally, the results and a transcription of the meeting were sent to them to examine the reliability of the results. Amendments were made in the model based on the comments received. The mentioned process, which has also been employed in previous studies [
33,
34], is believed to satisfy the reliability of the findings.
With regard to the validity, it should be noted that since the developed model is the basis for making decisions and committing resources, its validity has vital importance for model builders and users [
35]. To this purpose, prior to beginning each stage, the validity of the previous stage was assessed. Certain types of validity, including conceptual validity, experimental validity, operational validity, and external validity were deployed to evaluate the conceptual model, the ISM results, the proposed strategies, and the generalizability of the overall findings, respectively [
36]. In
Table 1, each type of validity is described, and the designed questions to evaluate them are shown. These questions have concordance with the questions designed in the previous studies [
37,
38,
39]. For each of the questions, the experts had to rate their level of agreement based on a 5-point Likert scale (from 1 as strongly disagree to 5 as strongly agree). The final value of the experts’ agreement was obtained by averaging the degrees of agreement of the experts with each question. The fourth column of
Table 1 shows the mean value of the experts’ agreement with each question. The final value of the experts’ agreement is finally presented in the fifth column.
With regard to the appropriateness and sufficiency of the experts employed to assess the reliability and validity of the research, it should also be noted that in this study, three panels of experts were deployed. The first panel, including twenty professionals, was responsible for identifying and verifying the critical causes of design changes. The second panel, including five professionals, was in charge of constructing the model and evaluating the validity of each stage, and the third panel, including eight professionals, was responsible for evaluating the external validity of the overall research findings. It is worth mentioning that although the number of experts in the first panel was limited to twenty, as the data gathered during the first stage were triangulated using literature review, questionnaires, and interviews with the second panel, the results and the number of experts are believed to be reliable and adequate, respectively. Referring to previous studies which employed the same approach, the number of experts in the second panel is also believed to be adequate [
40,
41]. In order to identify the right experts for the third panel of experts, purposive sampling was employed. The selection criteria were defined, namely, (1) having related publication(s) in established journals in the field of design changes, design errors, or design deficiencies; (2) having more than 15 years of experience with building construction projects; (3) holding a senior position in a professional organization. The experts were identified basically from the author lists of related publications, regional and country affiliates of International Project Management Association (IPMA), and Co-operative Network for Building Researchers (CNBR) Yahoo group. Invitations were sent to the experts via email, and they were requested to fill and return the questionnaire designed based on the four questions regarding the external validity of research findings, which are depicted in
Table 1. Five of the responses received were inappropriate, leaving only eight valid responses for the analysis. The respondents were from the United States, United Kingdom, Australia, Poland, Jordan, Ethiopia, China, and Malaysia, and all of them met at least two out of the three proposed criteria. This meets the need of having experts from both developed and developing countries. While a small sample is employed, the proper geographical balance of experts, who were positive of research results, is believed to satisfy the credibility and generalizability of the findings.
3.1. Data Collection
Through an intensive literature review, a primary list of causes of design changes, consisting of 38 causes, was prepared by the authors, which is presented in
Table A1. In order to narrow down the causes to the most critical ones and identify causes, which might be overlooked, the first panel of experts, consisting of twenty experts working in the Iranian building construction industry, was employed. All of them had at least 20 years of experience participating in building construction projects and have been holding very senor position in their representative organizations (five project managers, four construction managers, four senior designers, two senior engineers of technical office, two contract managers, two project control managers, and one site administrator). Out of these twenty experts, six experts were working for consultants, and eight and six experts were serving contractor and client firms, respectively. They all had a doctoral degree from a reputable university. It is also worth mentioning that heterogeneous experts, which were chosen from the community of consultants, contractors, and clients, reduce the adverse effects of biases and improve the generalization ability. Experts’ opinions with regard to causes’ significance were surveyed using a questionnaire consisting of three sections as follows.
In the first section, after providing a paragraph outlining the objectives of this study, the respondents were asked to clarify their general information. Respondents’ years of experience, qualifications, and position in their company were gathered in this part. This section was designed to ensure that all experts are appropriate for this study.
In the second part, the initial list of causes was provided for experts in a table, and they were asked to determine the level of importance of the causes using an ordinal scale of 1 to 5, which 1 indicated the lowest significance and 5 indicated the highest significance. Their opinions, then, were aggregated using the Relative Importance Index (RII) and based on Equation (1). This method has been repeatedly utilized in similar previous studies [
4,
23,
25] to rank and identify main factors.
where
Wi denotes the rating given to each cause by the respondents (ranging from 1 to 5),
A is the highest weight (i.e., 5 in this case), and
N is the total number of respondents.
In the third section, the respondents were asked to state other important causes which might have been neglected in the study. However, no further causes were identified using this part. Finally, causes, which had an RII of more than 60, were selected for the next step to investigate their interrelationships using the ISM approach. These causes are provided in
Table 2. The main causes were also categorized based on the stakeholders associated with them, and the management areas from which they arise. In terms of their stakeholders, five categories, namely contractors, clients, consultants, government, and all parties, and in terms of their relevant management areas, nine categories, namely policy, scope, time, cost, quality, communication, procurement, integration, and external were assumed for the causes.
3.2. Data Analysis
Having determined the most critical causes of design changes, in this stage, their interrelationships were investigated using the ISM. Interpretive structural modeling is defined as a process that transforms unclear and poorly articulated mental models of systems into visible, well-defined models [
43]. It provides a better understanding of how various variables of a complex system are interrelated and facilitates analyzing the direct and indirect effect of each variable on the others [
40]. This approach has been repeatedly applied to investigate construction management issues. Prakash and Phadtare [
41], for example, developed a hierarchical structural framework of verifiable drivers in project marketing through the application of the ISM. Iyer and Sagheer [
34] used the ISM to investigate the interrelationships of risks in Public–Private Partnership (PPP) projects. Tavakolan and Etemadinia [
44] incorporated fuzzy logic and proposed the fuzzy weighted interpretive structural modeling to determine and trace the interactions among project risks. Yu et al. [
33] also used the ISM approach to identify factors that significantly impact the utilization of big data and to investigate how these factors influence each other.
According to Janes [
28], three steps of the ISM approach are as follows.
Step 1: Relationship Identification
In this step, a Structural Self-Interaction Matrix (SSIM), which represents which causes directly affect other causes, is constructed. For this purpose, experts are asked to determine which causes influence a specific cause, and which causes receive influence from it. As experts’ opinions might differ, the nominal group technique is normally used, in which experts’ viewpoints are gathered independently and sent to other experts. This process continues until a consensus on causes’ interactions is achieved by experts. In this approach, respondents’ competency overrides their quantity, and there is no criterion to clarify how many respondents should be chosen. However, most previous studies have used three to five experts to develop the SSIM [
40]. For two causal factors such as
i and
j, each element of the SSIM located above the diagonal is assigned a letter based on the following rules:
V indicates factor i affects factor j;
A indicates factor j affects factor i;
X indicates factors i and j have a mutual impact on each other;
O indicates there is no relation between i and j.
In this paper, the second panel of experts, including five accessible experts who were willing to participate in this study, were deployed. Two of them were faculty members having more than 20 years of experience teaching and researching in the field of construction management. The other three experts had more than 25 years of experience working for different construction companies as senior project managers and also had a doctoral degree in the field of construction management. Providing the matrix of causes and the rules of filling the SSIM, they were asked to determine which causes are related to each other. Based on the nominal group technique, the completed matrices were collected and sent to other experts to be investigated. Finally, after four rounds, a consensus on the causes’ interactions was achieved.
Step 2: Model Calculation
In this step, the initial and final reachability matrices are generated, and the causes are partitioned into different levels. In the initial reachability matrix, based on Equation (2) and the SSIM, each element is assigned either 0 or 1, i.e.,
Having determined the initial reachability matrix, using the principle of transitivity the final reachability matrix is prepared. The transitivity states that if an element ‘
x’ is connected to an element ‘
y’ and ‘
y’ is connected to an element ‘
z’ then ‘
x’ is also related to ‘
z’. Thereafter, the reachability and antecedent sets can be obtained from the final reachability matrix. While the antecedent set for a cause like
C1 is composed of the cause itself and all other causes exerting influence on
C1, the reachability set for
C1 consists of the cause itself and all other causes receiving influence from
C1. Finally, comparing the reachability and antecedent sets, the intersection set, consisting of causes that are similar in both aforementioned sets, is defined for each cause. All the critical causes can be located on different levels by performing the process as follows. If the intersection set is the same as the reachability set, causes are located at the top level. The top-level causes meeting the above condition are eliminated from the element set, and this process is repeated iteratively till all the levels are determined [
40].
In this study, according to experts’ opinions and the SSIM, the initial reachability matrix was prepared, which is depicted in
Table 3. Following the rules of transitivity, the final reachability matrix was prepared, which is shown in
Table 4. This matrix indicates how a specific cause can directly or indirectly exert influence on other causes.
Step 3: Diagram Drawing
In this step, based on the final reachability matrix and the location of each cause in different levels, an ISM diagram, which represents the interrelationships of the causes, is drawn. The levels are indicative of causes’ significance, and the causes located at the lowest level of the diagram are the root causes of the considered problem. Therefore, more efforts should be put into managing them.
In this study, the final reachability matrix, the reachability set, and the antecedent set were determined for each cause, and after eight iterations, the causes were classified into eight levels. The hierarchical model of causes, which consists of eight levels, is illustrated in
Figure 2.
Having determined the driving and dependence power of each cause, based on the final reachability matrix, the MICMAC technique can be used to classify them. The MICMAC analysis works on the principle of multiplication properties of matrices [
45], and it aims to investigate the dependence and driving power of factors [
46]. According to Malone [
47], factors can be categorized into four clusters as follows.
Independent factors: this cluster is comprised of factors that have a high driving power and a low dependence power. These factors exert a profound influence on other factors and consequently have the capacity to drive the whole system.
Linkage factors: this cluster is comprised of factors that have high driving and dependence powers. These factors are unstable, and not only can any modification of them affect other factors but also it can result in a feedback effect on themselves [
47].
Autonomous factors: This cluster is comprised of factors that have low driving and dependence powers. As they are not connected to other factors, they are not of great importance in the system.
Dependent factors: this cluster is comprised of factors that have a low driving power and a high dependence power. As these factors receive an influence from others, they are mainly reliant on other factors and do not impact them.