Repertory grid technique in the development of Tacit-based Decision Support System (TDSS) for sustainable site layout planning

Automation in Construction 20 (2011) 818–829

Contents lists available at ScienceDirect

Automation in Construction j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / a u t c o n

Repertory grid technique in the development of Tacit-based Decision Support System (TDSS) for sustainable site layout planning Hamzah Abdul-Rahman, Chen Wang ⁎, Khe Siong Eng Faculty of Built Environment, University of Malaya, 50603 Kuala Lumpur, Malaysia

a r t i c l e

i n f o

Article history: Accepted 26 February 2011 Available online 25 March 2011 Keywords: Tacit-based Decision Support System (TDSS) Project management Sustainability Site layout planning Repertory grid techniques

a b s t r a c t Site layout planning is a significant but relatively ignored work on construction site, which has been treated improperly as somewhat routine. It is known that the complex interrelationship of material, equipment, laborers, space, environment, assess road, surrounding buildings, and building types affect the productivity and efficiency of a construction process. This complexity was inhibiting the smooth flow of resources especially when many trade contractors were working simultaneously on site. The study aimed to extract a set of core factors in site planning focusing on the tacit knowledge acquisition process to develop a Tacit-based Decision Support System (TDSS). A combination of the repertory grid technique and open-structured interviews was conducted for the tacit knowledge acquisition process. Cluster analysis and repertory grid analysis on the extensive responses from a structured interview were conducted. A computer program entitled “TDSS” was developed as a flexible tool to assist both senior and junior site layout planning engineers. © 2011 Elsevier B.V. All rights reserved.

1. Introduction Construction project consists of many different functions of work packages. The combination of these activities has made the construction industry a dynamic and competitive business environment. The site layout is always the important focus for every construction company to consider. Construction site layout planning has been a relatively ignored area in the construction management science, compared to other studies such as the cost estimation and the schedule planning [1]. In research, the site layout planning is certain to rise up again to a more important position in the whole construction process. Construction space is always the first place to study in visual 3D construction process research [2]. Construction site layout planning should be one of the areas for research, for instance, how big a place, when, what places should be allocated and arranged during a construction period has been a challenge to evaluate. Construction site layout planning in this paper reports on how to provide a good layout within a construction site and around the end product (building) by eliciting the principles that lie inside the construction expert's mind. Besides the objective in finding the principles on the construction layout planning, there was another objective in this research, which was to use the repertory grid, a psychology tool developed by George A. Kelly [3], to perform the knowledge acquisition process. This is the first time that such a technique has been applied in construction site layout planning.

⁎ Corresponding author. Tel.: + 60 3 7967 3203; fax: + 60 3 7967 5713. E-mail address: [email protected] (C. Wang). 0926-5805/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.autcon.2011.02.004

Common sense and experience are always used by the construction experts all over the world. These types of knowledge are often implicit, hidden and lie inside the experts' minds. Most of them are lost when the expert retires or changes his job to another field. This kind of loss is no longer affordable to the construction industry in this modern age. To be a competitive construction company, the first stage to stay competitive is to find a knowledge acquisition tool to extract the tacit knowledge from the company's experts and any other sources. The next stage is to build up a distribution network for all employees, including setting up the learning-based expert system. Therefore, the main objectives in this research are 1) to retrieve the implicit knowledge of the high-rise building site planning experts; 2) to determine the principles that lay behind the planning of construction site layout; and 3) to develop a Tacit-based Decision Support System (TDSS) to support construction planning engineers. As one result of this study, a knowledge framework on construction site layout planning is preserved. For the knowledge acquisition process, the repertory grid technique provided a scientific and reliable method to elicit that important implicit knowledge. Main factors and principles of construction site layout planning have been revealed. Finally, a Tacit-based Decision Support System (TDSS) has been developed for the sustainability in site layout planning. 2. Overview of Key Concepts Knowledge can be grouped into two main types, namely: explicit and tacit forms. Explicit knowledge is the knowledge that can be recorded through database, while tacit knowledge is only stored inside a human head, such as those common sense or experience [4].

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

Nomenclature CIDB DBKL MPPJ MPSJ MBSA MPKj MPAJ MPS MPK PCA VBA

Construction Industry Development Board Malaysia Dewan Bandaraya Kuala Lumpur (in Malaysian Language); Kuala Lumpur City Hall (in English) Majlis Perbandaran Petaling Jaya (in Malaysian Language); Petaling Jaya Municipal Council (in English) Majlis Perbandaran Subang Jaya (in Malaysian Language); Subang Jaya Municipal Council (in English) Majlis Bandaraya Shah Alam (in Malaysian Language); Shah Alam City Council (in English) Majlis Perbandaran Kajang (in Malaysian Language); Kajiang Municipal Council (in English) Majlis Perbandaran Ampang Jaya (in Malaysian Language); Ampang Jaya Municipal Council (in English) Majlis Perbandaran Selayang (in Malaysian Language); Selayang Municipal Council (in English) Majlis Perbandaran Klang (in Malaysian Language); Klang Municipal Council (in English) Principal Component Analysis Visual Basic for Applications

Knowledge acquisition is a process of collecting the valued and useful information from both kinds: tacit and explicit. In this new information era, knowledge acquisition would be one of the first missions for a construction company to determine where their precious knowledge is. Construction site layout planning is a highly dynamic process. The tacit knowledge acquisition in this research was to structure and formalize the concepts and rules that lie inside the minds of site planning experts. The repertory grid technique was used for the acquisition of tacit knowledge. Repertory grid is a special technique based on the personal construct theory. Originally it had been used in psycho-therapeutic treatment. After four decades of evolution, the repertory grid has been widely applied to various fields such as business, computer science, social science, and engineering. In this section, the reviews on repertory grid, personal construct theory, and construction site layout planning are conducted. The repertory grid technique was used in this research to retrieve the tacit knowledge of site planning experts. 2.1. Overview of Personal Construct Theory Every human is a personal scientist for himself, which is the origin of the personal construct theory developed by George A. Kelly in 1955 [3]. In personal construct theory, every person makes his own unique world by using his own value system. The value system of each person is a mixture of that person's life experience, knowledge, emotion and personality attributes. Every person is unique. Even more, in a philosophical position the personal construct theory belongs to the constructive alternativism. And the basic statement in constructive alternativism is that “we assume that all of our present interpretations of the universe are subject to revision or replacement” [3]. That causes the changing status of each person's value system from time to time. It is in motion and not static. People change their assessments, their decisions, even their choices when times pass or something slips in. People evaluate the world with a limited number of constructs which is becoming one's personal view [12]. Besides that, constructive alternativism has one scientific and pragmatic principle — “the systematic analysis of the conceptions employed by ordinary and scientific thought in interpreting the world, and including an investigation of the art of knowledge, or the nature of knowledge as such” [3]. The personal construct theory contains the principle well within its theory. A person's processes are psychologically channelized by the ways in which he anticipates events [3].

819

The statement above is the fundamental postulate of personal constructs. It can be interpreted another way as every person builds his own “value system” or “construct” by using his own unique interpretation framework. How someone interprets an event is different from person to person. For example, how good is the car? The definition of good is different for every person and that definition that is built inside a person's mind is a special interpretation framework. However, this interpretation framework is not in a static model but in a moving state. The person is not an object which is temporarily in a moving state but is he a form of motion [3]. Although the interpretation framework keeps changing for a human, some principles can still be obtained. It is like the patterns of a human mind set. For this pattern and principles, Kelly [3] made a few propositions to categorize them. And those propositions are called corollaries. A summary of corollaries is shown to better understand how the personal construct theory is developed. Some of the corollaries which relate to this research are elaborated [3]: a) Construction Corollary: A person anticipates events by construing their replications. b) Individuality Corollary: Persons differ from each other in their construction of events c) Organization Corollary: Each person characteristically evolves, for his convenience in anticipating events, a construction system embracing ordinal relationships between constructs. d) Dichotomy corollary: A person's construct system is composed of a finite number of dichotomous constructs. e) Choices Corollary: A person chooses for himself that alternative in a dichotomized construct through which he anticipates the greater possibility for extension and definition of his system. f) Range Corollary: A construct is convenient for the anticipation of a finite range of events only. g) Experience Corollary: A person's construction system varies as he successively construes the replications of events. h) Modulation Corollary: The variation in a person's construction system is limited by the permeability of the constructs within whose ranges of convenience the variants lie. i) Fragmentation Corollary: A person may successively employ a variety of construction subsystems which are inferentially incompatible with each other. j) Commonality Corollary: To the extent that one person employs a construction of experience which is similar to that employed by another; his psychological processes are similar to those of the other person. k) Sociality Corollary: To the extent that one person construes the construction processes of another; he may play a role in a social process involving the other person. The eleven propositions above are applied as to how an interpretation system or “Constructs system” of humans is developed and how it changes. There are some corollaries closely related to this research which heavily affect the site experts' decision, namely: construction corollary, organization corollary, dichotomy corollary, experience corollary and commonality corollary. 2.2. Overview of Repertory Grid Repertory grid is a part of the personal construct theory. The repertory grid is the diagnostic instrument designed by George A. Kelly [3] in his book “The Psychology of Personal Constructs”. The repertory grid was first designed as the diagnostic tool to test the personality of humans, which called “role constructs”. After decades of evolving, thousands of researchers have applied the repertory grid technique into various domains such as business management, computer science, education, and marketing. The design of repertory grid is based on the theory of personal constructs. The grid provided a method to assess the degree of structure and organization in

820

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

construing without bias [5]. A full repertory grid comprises three components, namely: “elements”, “construct” and “linking mechanism” [6]. 2.2.1. Elements In the original design of repertory grid, element is the “role constructs” of a person. In this research, the site planning experts' tacit knowledge is the “elements”. There are two general principles that have to be emphasized in selecting the elements. First, elements should be homogeneous [7]. Second, the elements should provide representative coverage of the area to be investigated [7]. This research focuses only on the construction site layout planning for high-rise buildings. A research done by Mohd Nazir Bin Sarah in year 2005 [8] on decision making tools for facilities refurbishment tender transactions in Malaysia Airline System applied the two general principles in his element selection. It provided a good result in identifying the differences and similarities among various characteristics of experts. 2.2.2. Constructs In the original design of repertory grid, the construct is the “value”, “judgment”, and “interpretation framework” of elements [9]. The constructs in this research are the attributes of the experts' knowledge. The definition of “construct” in this study has been refined to fulfill the research objective, which is to elicit the knowledge of site experts in construction site layout planning. They are then designed as the “factors that affect the site layout planning”. In this way, the judgment and decision making process of site experts in construction site layout planning could be elicited. When the repertory grid technique was used to analyze the data, the similarities and differences of decision making of different site experts in construction site layout planning were compared. The researchers arranged a confluence and comparison on different patterns of site experts. In the original constructs design, Kelly stated six assumptions to build a valid set of constructs [9], namely: a) the permeability of the constructs elicited (the site experts must be able to apply the constructs to the applicable situation or site layout); b) the preexisting constructions should be elicited (the existing commonly used technical term should be placed inside the constructs, in order for the site expert to easily understand); c) the verbal message attached to the construction should be communicable (the words or “technical terms” in the constructs should be understandable); d) the constructs should represent the subject's understanding (the constructs should represent the common value system of the site experts); e) the subject should not dissociate himself entirely from the constructs (the site experts should be able to see something from the constructs rather than seeing the strange thing); and f) the constructs elicited should be explicitly bipolar (the constructs should be stated in bipolar basis. The good or bad, have or don't have types of factors are placed inside the constructs). 2.2.3. Linking Measurements The linking mechanism in the definition by Easterby-Smith [7] is the way the constructs are used in relation to the elements, which indicates the meaning of the labels given to each pole [7]. To extend this component further, it includes the analysis of the links. Repertory grid consists of three types of links, namely: the links between constructs and constructs, the links between elements and elements, and links between constructs and elements. The “link” means the relationship between the components (constructs and elements). In the classical approach, the essential objective is to find the attributes of the elements and the constructs, and to determine what are they and how they match between each other for the repertory grid. The linking measurement is the measurement of the links between constructs and elements.

2.2.4. Mathematical Model of Grid Analysis The traditional methods of grid analysis used to be the D² method of factor analysis, the principal component analysis, and the multidimensional scaling [10]. Recently, cluster analysis has been used to identify the pattern of grids. In this section, the mathematical model of principal component analysis and cluster analysis are introduced. Actually in RepIV, principal component analysis and cluster analysis are the two main analyses. Principal component analysis (PCA) is a technique to simplify a group of data by reducing the dimensionality. It is powerful to identify the patterns in the data of high dimension. PCA is able to indicate the similarities and differences of the dataset. The basic idea of principal component analysis is to describe the relationship between two variables with a factor. On the other hand, cluster analysis Cluster analysis is a way to find the different patterns of the clustering of datasets, which is the grouping of similar objects [11]. Using the different perspectives to differentiate the dataset would produce different subgroups of the data and produce different explanations on the dataset. However, it does not mean that the researchers could randomly make the groupings. The cluster analysis provides a way to calculate the distance between two single data or the distance between two subgroups to identify the patterns of the cluster. Cluster analysis comprises a few different computation methods and concepts to do the analysis. The objective of cluster analysis is to identify the cluster in an unknown or unclassified group. Since it is an unknown group, the method of how to classify the dataset and how to identify the group becomes quite flexible. The computation techniques of principal component analysis are used to perform the cluster analysis.

2.3. Overview of Construction Site Layout Planning Site layout is considered as the space on site, which is available for temporary or general construction equipment, material layouts, and flows of all resources involved in adding to the end product. In simple terms, it is the site area minus the end product (although in some projects parts of the end product are used as part of the site area) and in others there are off-site marshalling storage areas [13]. Good site layout planning assists in minimizing the traveling time and movement costs of plant, labor, and materials, activity interference during construction work, and site accidents, and ensures that work on buildings and other construction positions is not impeded by the thoughtless storage of materials on these locations [14–17]. The highly complex nature of site layout has emerged from some different study focuses. Various topics on different aspects of site layout that were studied in the recent past included the space conflict, the planning sequence of temporary facilities, the positioning of material and equipment, the intelligent equipment selection system, the intelligent materials routing system, multi-constraint site planning, the genetic algorithm for construction site layout, the genetic optimization of site layout, routing system for large vehicles, GIS-based site layout planning

Table 1 Company position of interviewees (elements). Company position

Co. Director (Subcontractor) Head of Department/Department Manager/General Manager/Project Director Construction Manager/Project Manager/Sr. Project Manager/Sr. Technical Manager/Site Manager/Project Coordinator Project Engineer/Assistant Construction Manager/Site Agent/Resident Engineer/Project Planner/Sr. Project Engineer

Number of interviewee (elements)

Percentage (%)

1 5

3.85 19.23

13

50.00

7

26.92

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829 Table 2 Working experience in site layout planning. Experience (years)

Number of elements

Percentage (%)

1–5 6–10 11–15 16–20 20 above

3 3 3 7 10

11.54 11.54 11.54 26.92 38.46

system, work space in multistory buildings and work space generically in a construction method model [2,18–24]. Some focused on the quantitative factors and some focused on the qualitative aspects, but the combination of quantitative and qualitative aspects of the site layout problems has made it hard to create a reasonable comparison between each of these. For example, the routing system for large vehicles and multi-constraint site layout planning might not be able to create in exactly the same site conditions because of the different focuses. 3. Data Acquiring and Analysis Procedures The repertory grid method attached with an open-structured interview survey was applied in this research. The repertory grid was applied because it had a well established system to perform the knowledge elicitation process from design to analysis. Recently, it has extensively been used to study different disciplines of expertise yet it has rarely been used in the civil engineering decision making process. Construction site layout planning has always been conducted using the tacit knowledge. The repertory grid could retrieve this kind of knowledge efficiently. Repertory grid technique allows the researcher to get a mental map of interviewees, and to write this map with the minimum observer bias. The open-structure interview was applied because it could provide extensive and rigorous interview system for qualitative research. In this study, both methods proved that they

821

could appropriately be linked together to form a system. The grid questionnaire and the open-structured interview questions were designed after 23 conversations with site supervisors and engineers as a pilot study. In this research, there were 125 main factors related to construction site layout planning and 13 factors related to personal psychology. In the initial setting, these 13 psychological factors were part of the analysis to identify the reliable interviewees which were taken into account in developing the Tacit-based Decision Support System (TDSS). 3.1. Sampling for Repertory Grid In the initial trial, the local authorities were consulted to get such well defined samples. The researchers aimed at all the Klang Valley local authorities, namely: DBKL, MPPJ, MPSJ, MBSA, MPKj, MPAJ, MPS and MPK. These local authorities were located in the Klang Valley area where more than 50% Malaysian high-rise buildings are concentrated. Klang Valley is an area in Malaysia comprising Kuala Lumpur and its suburbs, and adjoining cities and towns in the state of Selangor. An alternative reference to this would be Kuala Lumpur Metropolitan Area or Greater Kuala Lumpur. It is geographically delineated by Titiwangsa Mountains to the north and east and the Strait of Malacca to the west. The conurbation has a total population of over 6 million, and is the heartland of Malaysia's industry and commerce. Klang Valley is home to a large number of migrants from other states within Malaysia and foreign workers largely from Indonesia, India and Nepal. The researchers visited each local authority to retrieve the data of those projects related to this study. Only those buildings above 16 storeys were selected and recorded. The total amount of projects found from all these authorities was about 550 in number. The initial expectation was to find 30 qualified interviewees from these 550 projects. There were total of 26 valid interview data sets obtained eventually. The interview period fluctuated from one and a half hours

Fig. 1. The distribution of elements in the first and second principal component.

822

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

to three hours. Table 1 classified the 26 interviewees' job positions into four levels. The first level was the director of company, in this research there was only one company director who was the owner of a contractor. The decision making for this level would included the construction process, technical planning, profit of company, resource planning, and all types of business strategies and policies. The second level comprised of heads of department, the general managers, and project directors. This level of decision making would be included in the construction process, resource planning, problem solving, and compliance with the company policies, person-in-charge selection and contractor management. The second level was usually set in large-size companies and people in this level always came across several projects at a time. The third level included the construction managers, project managers, site managers, and project coordinators. This level usually means the manager for a single project. They were the person-in-charge of a particular project. Job functions in this level included resource planning, construction process, and cost of construction. They need to meet every goal set by their headquarters for one single project. The final level consisted of the project engineers, assistant managers, and residential engineers. This level of engineers would go through a specific task in a single project. Those tasks included the planning, progress inspection, and management of workers. Table 2 shows the interviewees' experience in site layout planning. There were more than 76.92% interviewees who had at least a 10-year experience in site layout planning. In the original design of repertory grid, element is the “role constructs” of a person. In this research, the site planning experts' tacit knowledge is the “elements” as mentioned in Section 2.2.1.

After the considerations for potential users be determined, the second step was to design forms which would be created in TDSS. The process was supposed to answer questions such as “how the presentation should look like in the TDSS” and “how it should guide the decision maker to answer the questions and make the comparison”. The third step was to select a proper program to develop

3.2. Data Analysis Using Rep IV Data analysis was conducted after 26 sets of data were collected. The first step required in this stage was to choose appropriate repertory grid software. Eleven repertory grid software found in the market were taken into consideration, which included: a) Indiogrid, b) Webgrid III, c) Rep IV, d) Enquire Within, e) Grid Stat and Grid Scal, f) Circumgrids III, g) Omnigrid, h) Grid Cor, i) Grid Suite, j) Ingrid X, and k) Flexigrid. The reason for choosing Rep IV was because it contained various grid analysis programs in RepIV, for instance the PLANET, KSS0, NEXTRA and WebGrid. The second reason was that compared to the others, RepIV had been tested widely around the world so that it was more reliable to use. This software cost about USD 600. Assisted by RepIV, the 26 sets of data were analyzed. There were a total of 138 factors in the repertory grid. The initial step was to seek for the differentiation of different types of experts. Hence, there were thirteen psychological factors that had been included. Each interviewee had 125 technical (related to construction site layout) factors keyed into the RepIV. There were 3250 factors for 26 interviewees. Two grid algorithms in the RepIV were used, namely: the principal component analysis and hierarchical (tree) cluster analysis. 3.3. TDSS System Development The system design of TDSS started after the analysis stage. The first consideration was to build a decision support system which contained the tacit knowledge from experts. The second consideration was to build a user-friendly TDSS. Site planning engineers were not expected to take a lot of time sitting in front of a computer to test for the quality of their decisions. Actually, they usually need an immediate answer when the job came up. Hence, the developed TDSS should not be difficult to use and to understand. Third, it should be able to provide a better decision making flow compared to the common way. Therefore, the potential users were expected to be guided through a process, and then they could be able to find their own special set of factors which were suited to the particular project that they were working on.

Fig. 2. The distribution of constructs in the first and second principal component.

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

Fig. 3. The cluster analysis of the 26 elements and 125 constructs.

823

824

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

these forms. In the selection of such software, the Microsoft Visual Basic, Microsoft Access, and Adobe Designer were tested and compared. In the midst of testing, the research team found that in the RepIV version1.12R, the functions of data import and export could be created. With these import and export functions, two formats could be imported or exported to the RepIV, namely: the spreadsheet format and text format. The research team realized that the most compatible program with the RepIV was the Microsoft Excel. Excel is a program initially created in the spreadsheet format and it supports the text format as well. The VBA function was built in the Excel. VBA is a simpler version of Visual Basic, which enables many small programs to run simultaneously inside the Excel. In addition, the compatibility of Excel was much higher than other programs, because Microsoft Windows and Microsoft Office were both the most widely used systems and programs in the world. With this kind of high compatibility, the program can be easily applied to most computers. It came to the programming stage, which was to develop the TDSS, after the design program was decided. In this stage, the research team employed a business intelligence software developer to create the buttons inside the Excel by using the VBA. Researchers started to create buttons such as “option button” and “command button” with VBA to fulfill every specific task that was assigned to them. For example, the “Reset Button” was programmed to empty all existing data in the form and the “Option Button” was to key in the number into the form in order to make the analysis later, and the “Process Button” was to categorize the factors into the boxes that were previously set. TDSS was developed as a system to acquire and analyze the tacit knowledge from experts. This kind of information was to be typed into the computer in order to figure out the patterns of the data, and then theorized into checklists and guidelines for all other site planners to learn from this pattern. In this study, site experts were interviewed using the repertory grid technique attached with open-structure interviews to record the critical factors of construction site layout planning. The TDSS could provide valuable support to solve ill-structured decision problems. The Microsoft Excel and RepIV were used to develop a human–computer interface. When the interface was built on the web, the RepIV provided an analysis engine for users. The TDSS was assessed and validated by selected site planning experts.

4. Data Interpretations and Results There were 125 construction site layout factors and 7 psychological factors in each set of data in this repertory grid analysis. There were 26 sets of data each of which had 125 factors that had been analyzed. Therefore, a total of 3250 variables were keyed into the RepIV (repertory grid software) and STATISTICA 6 (Statistical software) for analyses. There were two main types of analysis that were included in the RepIV, namely: principal component analysis and cluster analysis. The original Kelly's repertory grid was used in factor analysis. Factor analysis is considered as an extension of principal component analysis. Basically, these three types of analyses have similar objectives though with different algorithms.

4.1. First Trial — The Principal Component Analysis The first analysis trial in this study was the principal component analysis. The result is shown in Fig. 1. In the initial stage, clusters were made to differentiate the elements. For example, group 1 includes M25, M21, M3, M19, M20, and group 2 includes M14, M11, M6, M12, and M15. In similar way, a few clusters with different elements within were created. The main problems came up as what were the criteria and principles for grouping. In the initial argument, the researchers tried to differentiate the experienced engineers and fresh engineers. However, in this diagram, this argument seems not consistent with all groups, which might be caused by the insufficient number of the junior engineers compared to the experienced engineers to make the differentiation. The errors found in this diagram were the first principal component and second principal component consisted of the variance percentage at 29.6% in all factors, which was not enough to reflect the overall picture. In the bottom of Fig. 1, there was a list of numbers, which indicated the variance percentage of components. Fig. 1 consists of the first component at 18.9% and the second component at 10.7%, respectively. Hence, the figure was drawn with the pair of the second and fourth component or the third and seventh component so that the positioning of elements in the figure would be changed accordingly. It found that principal component analysis was consistent with the K-Mean clustering analysis in some clusters. The distribution of constructs in the first and second principal component is shown in Fig. 2.

4.2. The Second Trial — Hierarchical Cluster Analysis Cluster analysis is a classification technique which is used to cluster the similar events, data, and structures. It was simple to apply yet powerful. In the RepIV, hierarchical cluster analysis was included in a subprogram — “Focus”. Fig. 3 shows the cluster analysis of the 26 elements and 125 constructs. As illustrated in Fig. 3, the cluster of data was shown by the colors of grey and white. Cluster analysis had no sense of order compared to the principal component analysis. Hence, the cluster analysis was to find out the difference between two constructs or elements. Difference was the main concern of the cluster analysis. In Fig. 3, there was a white cluster in the upper part and grey color in the lower part, which showed clear differences. The remaining part was quite mixed, which meant little difference. The box above showed the clustering of constructs. The clustering of elements was shown in the line above the box. Fig. 4 shows a captured screen of the elements clustering. In Fig. 4, each line connected to the box was an element. The number on the right side is the percentage of match between the elements. The lines from the right hand side, lines number 3 and 4 were matched with about 80%. With these matches, the cluster could be made. With the clustering of elements, the research team started to make a few groups of clusters as that had been attempted to in the principal component analysis. To conduct grouping in the cluster analysis was actually more accurate compared to that in the principal component analysis, since the cluster analysis reflected the data information objectively. The cluster analysis in the RepIV was called the “hierarchical

Fig. 4. Captured screen of the clustering of elements.

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

cluster analysis”. Fig. 5 shows a captured screen of the clustering of constructs. Fig. 3 (right side) and Fig. 5 show the similar pattern of constructs clustered, which meant that the average mean of the constructs was similar and the standard deviation of constructs was small so that they would be clustered. The complexity in the clustering of constructs in Fig. 3 was that the constructs had exchanged their (right/left) side to make them clustered. The right and left side of the constructs made a big difference. The research team had to make sure in the constructs clusters the constructs were normal or reversed. Besides Fig. 3, the RepIV had a set of matrix to show the elements and constructs matches. Table 3 shows that the elements were matched close to each other. For example, the changes between 60% and 75% matches could be considered as reasonable for the researchers to change the elements in each cluster. Hierarchical cluster analysis is open for researchers to make a fine-adjustment. 4.3. The Third Trial — The K-Mean Cluster Analysis K-Mean cluster analysis was conducted after the research team conducted a few email communications to the founder of RepIV, Dr. Brian R Gaines, who is the pioneer of repertory grid in the world. Dr. Gaines suggested the researchers to use the discriminant analysis. The discriminant analysis was conducted after the groups or clusters were determined to provide a reasonable answer for the repertory grid. Exploratory statistical analysis was also focused at this stage assisted by the software “STATISTICA 6” to compare different types of analysis including factor analysis, two-step cluster analysis, regression analysis, and K-Mean cluster analysis. K-Mean cluster analysis was proved to be able to provide the most reasonable and valid result. The result of the K-Mean cluster analysis for this study is shown in Table 4. Different from hierarchical clustering, the K-Mean clustering allows researchers to decide how many clusters are needed. Though the hierarchical clustering also compared data and drew matching lines, the K-Mean clustering provides a stronger mathematical base. The analysis results for the clusters in K-Mean are shown in Table 5. Table 5 shows the distance among the members of each cluster in the 3 groups of clustering in the K-Mean analysis. The first group of clusters has the members of M1, M2, M6, M11, M12, M13, M14, and M15. There were three optimal points chosen by the K-Mean algorithm in the center of the clusters as the guide. The “distance” was the distance between each member and the center of the cluster. For example in Table 5, the element M1 has the distance at 0.9938 to the center of the 1st cluster. To make sense of the data, the superior and the peer group assessments by combining the results of qualitative data and personal assessments of the elements provided reliable data set to key in TDSS. 4.4. Different Cluster Groups Appeared in K-Mean Clustering The K-Mean clustering algorithm is composed of four steps, namely: a) Initialization: to create K (K= number of clusters wanted to create) number of initial cluster points randomly from rows of data in the dataset; b) Close distance data searching: to start search for the closest data nearby the initial cluster points and start form the cluster; c) Recalculate the mean: to update the center of clusters in each cluster by recalculating the arithmetic mean of the cluster in the dataset; and d) Repeat step 2 and step 3: to update the center of cluster until the stable cluster was found. In the algorithm procedures, K-Mean clustering was sensitive to the initial cluster points setting. It has been reported that solutions obtained from the K-means are dependent on the initialization of the cluster center. In STATISTICA 6, three different options of initial cluster center settings were offered as follows: 1) Choose observations to maximize the initial between-cluster distance; 2) Sort distances and take observations at constant interval; and 3) Choose the first N (number of clusters) observations.

Fig. 5. Captured screen of the clustering of constructs.

825

826

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

Table 3 The match of elements. 1* 2* 3* 4* 5* 6* 7* 8* 9* 10 * 11 * 12 * 13 * 14 * 15 * 16 * 17 * 18 * 19 * 20 * 21 * 22 * 23 * 24 * 25 * 26 *

100 76 63 76 68 77 74 71 77 72 73 56 75 78 66 73 80 65 64 68 72 72 67 76 73 72

76 100 59 70 61 74 66 63 69 66 67 60 71 78 65 69 73 66 59 63 68 68 65 72 66 68

63 59 100 65 62 58 58 55 65 59 55 55 58 64 54 62 64 58 69 64 68 61 62 61 65 60

76 70 65 100 72 73 74 65 73 73 71 63 66 73 60 73 76 57 68 65 72 73 61 68 70 66

68 61 62 72 100 66 74 53 66 80 63 64 55 68 58 66 74 49 69 56 62 78 48 59 65 59

77 74 58 73 66 100 75 68 71 70 70 61 69 75 66 67 75 64 57 64 62 71 62 71 62 65

74 66 58 74 74 75 100 63 67 76 70 61 65 72 64 67 78 56 63 63 65 78 55 67 68 66

71 63 55 65 53 68 63 100 69 59 64 47 66 64 57 63 65 60 56 64 63 59 63 68 64 66

77 69 65 73 66 71 67 69 100 68 69 58 68 71 62 72 72 63 65 67 68 68 63 73 67 68

72 66 59 73 80 70 76 59 68 100 68 66 61 70 61 69 75 53 67 61 68 80 56 67 70 62

73 67 55 71 63 70 70 64 69 68 100 58 68 69 61 65 71 58 57 64 64 71 60 70 68 70

56 60 55 63 64 61 61 47 58 66 58 100 54 60 56 62 60 50 59 59 60 65 51 59 56 56

75 71 58 66 55 69 65 66 68 61 68 54 100 73 61 63 73 64 56 64 63 65 65 71 65 68

The first option was to maximize the distance between the clusters when it chose for its initial cluster centers. The second option was to choose the initial cluster centers in a relatively average situation. The third option was to randomly choose whatever the first observations had been set. The statistical software provided different options for the researchers to choose before the initial cluster center was automatically set. The second option (sort distances and take observations at constant interval) was used to create the groups of clusters. The reason for this option to be chosen was because it showed the most significant pattern which could be explained in words. In this study, the different groups of clusters with different elements in a cluster were presented. In the clustering algorithm, for example, the tree clustering (existed in the RepIV) was designed to select groups. As long as the clusters had a significant pattern that could be illustrated for certain phenomena, the cluster was valid. After all those cluster groups were compared, the M19, M21, and M25 remained consistent in their data pattern. The focus group had appeared in “four-group clustering” in both the first option and the third option. Likewise, members of M3, M16, M19, M21, and M25 were valid to be taken as the focus group in the TDSS system.

5. TDSS System and Validation 5.1. The Appearance of TDSS in RepIV There was a special function in the RepIV that enables users to compare two different grids, which was called “SocioGrid”. In the developed TDSS, this function was designed to compare the inputs of the current site planning engineer (User) and the solutions of the focus group (Experts). After the inputs of the current site planning engineers were exported as a text file in RepIV format, it could be opened side by side with the solutions of the focus group for comparison. After the current site planning engineer's text file was created in the Excel, it could be saved in a hard disk. A captured screen of TDSS in Excel is presented in Fig. 6. The next step was to open the RepIV by click on the “New” button in the SocioGrid Column. Open the focus group file and the current site planning engineer text file, click on the button “Composite”. These two grids could combine as shown in Fig. 7.

78 78 64 73 68 75 72 64 71 70 69 60 73 100 71 67 77 64 62 67 68 73 62 73 70 68

66 65 54 60 58 66 64 57 62 61 61 56 61 71 100 59 66 56 53 59 57 63 54 63 56 60

73 69 62 73 66 67 67 63 72 69 65 62 63 67 59 100 72 64 69 66 74 72 67 66 68 64

80 73 64 76 74 75 78 65 72 75 71 60 73 77 66 72 100 62 71 68 70 77 61 71 73 69

65 66 58 57 49 64 56 60 63 53 58 50 64 64 56 64 62 100 56 65 61 54 66 65 62 61

64 59 69 68 69 57 63 56 65 67 57 59 56 62 53 69 71 56 100 67 72 68 62 63 72 64

68 63 64 65 56 64 63 64 67 61 64 59 64 67 59 66 68 65 67 100 70 65 68 74 67 71

72 68 68 72 62 62 65 63 68 68 64 60 63 68 57 74 70 61 72 70 100 68 69 68 72 66

72 68 61 73 78 71 78 59 68 80 71 65 65 73 63 72 77 54 68 65 68 100 55 67 68 63

67 65 62 61 48 62 55 63 63 56 60 51 65 62 54 67 61 66 62 68 69 55 100 68 66 67

76 72 61 68 59 71 67 68 73 67 70 59 71 73 63 66 71 65 63 74 68 67 68 100 67 70

73 66 65 70 65 62 68 64 67 70 68 56 65 70 56 68 73 62 72 67 72 68 66 67 100 69

72 68 60 66 59 65 66 66 68 62 70 56 68 68 60 64 69 61 64 71 66 63 67 70 69 100

M1 M2 M3 M4 M5 M6 M7 M8 M9 M10 M11 M12 M13 M14 M15 M16 M17 M18 M19 M20 M21 M22 M23 M24 M25 M26

The second analysis function in the RepIV was designed to figure out the percentage of matches between the focus group and the current site planning engineer. The RepIV could figure out the percentage of matches among elements in the focus group and the current site planning engineer. The percentage of matches between elements of focus group and current site planning engineer is shown in Fig. 8.

Table 4 The groups of clustering in K-Mean analysis. Three-Group Clustering Cluster 1 Cluster 2 Cluster 3

N 1, 2, 6, 11, 12, 13, 14, 15 N 3, 8, 9, 16, 18, 19, 20, 21, 23, 24, 25, 26 N 4, 5, 7, 10, 17, 22

Four-Group Clustering Cluster 1 Cluster 2 Cluster 3 Cluster 4

N 1, 3, 8, 9, 13, 16, 18, 19, 20, 21, 23, 24, 25 N 2, 6, 12, 14, 15 N 4, 5, 7, 10, 17, 22 N 11, 26

Five-Group Clustering Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5

N 1, 4, 5, 7, 10, 11, 17, 22 N 8, 9, 13, 18, 20, 23, 24, 26 N 3, 19, 25 N 16, 21 N 2, 6, 12, 14, 15

Six-Group Clustering Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6

N 3, 16, 19, 21, 25 N 18, 23 N 1, 8, 9, 11, 13, 20, 24, 26 N 4, 5, 7, 10, 17, 22 N 12 N 2, 6, 14, 15

Seven-Group Clustering Cluster 1 Cluster 2 Cluster 3 Cluster 4 Cluster 5 Cluster 6 Cluster 7

N 8, 13, 18, 20, 23, 24 N 5, 7, 10, 22 N 12 N 2, 6, 14, 15 N 1, 4, 9, 16, 17 N 3, 19, 21, 25 N 11, 26

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

827

Table 5 The distance of the member of the three-group clustering.

Distance

Distance

Distance

M1

M2

M6

M11

M12

M13

M14

M15

0.9938

1.2111

0.9596

1.4623

1.6909

1.1645

1.0001

1.4034

M3

M8

M9

M16

M18

M19

M20

M21

M23

M24

M25

M26

1.7135

0.9922

1.1519

1.4776

1.7002

1.4615

1.3523

1.3354

1.5220

1.2691

1.3089

1.4341

M4

M5

M7

M10

M17

M22

0.9229

1.1658

1.1137

1.0816

0.9843

1.0290

The percentage of matches between elements of each focus group and the current site planning engineer was shown in the chart on the left side of Fig. 7. Analyses could be performed in this composite grid as one normal grid. TDSS provided a professional repertory grid analysis for the engineers (users) to take a grip on how they perform when they are compared with focus group members. This analysis is available for a large composite grid for complex problem solving.

each process. After the testers finished the steps in TDSS, they were requested to fill up an evaluation form for the TDSS. The competed evaluation forms showed that TDSS could be used in the construction site layout planning to improve the speed and accuracy of decision making process.

5.2. The Validation of TDSS

This research team previously applied the Grounded Theory [25] to perform its analysis, which was later on proved to be prejudiced. The main difference between the Grounded Theory and the Personal Construct Theory is that the Personal Construct Theory has a clear objective when the research started, but the Grounded Theory does not. TDSS tackled a focused area of practice and produced a prototype format. The function of “Sociogrid” in RepIV enables the user to analyze between his/her inputs and experts' solutions. The TDSS was developed to be compatible with both Microsoft Excel and RepIV. Visual Basic for Application (VBA) was used in the Excel to create a user-friendly platform for TDSS. The Excel was able to export the users' inputs to the RepIV. The functionality that allowed the users to perform a complete analysis on repertory grid enabled the advanced users to conduct large scale analysis for multiple users' data entry. TDSS should be continuously evolving. There are many special

As a final product of this research, the developed TDSS serves as a guide for site planning engineers. A validation process for the TDSS was carried out with a total of six testers (site planning engineers). TDSS was expected to improve the speed and accuracy of their decision making process. Three testers were from the pool of interviewees and the other three were members from the Construction Industry Development Board (CIDB) Malaysia. All of them were site planning engineers for high-rise buildings. The researches ran the TDSS system in each tester's construction site. The testers were requested to test the TDSS system in front of all project members and the heads of sub-contractors. All witnesses were allowed to raise questions during the testing period. The research team had prepared a basic TDSS manual as the reference for users to ensure fairness on

6. Discussions and Limitations

Fig. 6. Captured screen of TDSS in Excel.

828

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

Fig. 7. The composite grid of focus group and current site planning engineer.

computer techniques to be taken into the development of TDSS in future research. For example, currently there are some simple tracking systems which could be used to track the decision pattern of every computer user. The decision support system can start updating the system knowledge according to the popularity of the knowledge used. In another way, it can also be used to track the errors

made by the users. This type of application is expected to create a more precise system compared to the TDSS developed in this research. Further, simulation of real conditions, predictive algorithms, and more comprehensive and user-friendly statistical analysis packages should be taken into consideration in future improvements of TDSS.

Fig. 8. The percentage of matches between focus group element and current site planning engineer.

H. Abdul-Rahman et al. / Automation in Construction 20 (2011) 818–829

7. Conclusions and Recommendations In the development of TDSS, repertory grid technique was applied to acquire experts' tacit knowledge in construction site layout planning for the first time. The repertory grid technique and openstructure interview survey were jointly combined in this research to produce a comprehensive and valid set of results. This research was an exploratory process in the development of the TDSS. The principles of construction site layout planning for high rise buildings (tacit knowledge of experts) are solidified in the developed TDSS system. TDSS is a decision support system that contains knowledge. The knowledge of the construction site layout planning for high rise buildings in the urban areas was loaded into TDSS in this research. The TDSS was mainly developed as a simple to use but powerful model which can support the decision making process of every site planning engineer. This can lead to better and faster decisions compared with the normal process. In future study, the repertory grid technique is advised to be combined with a qualitative research technique in order to develop an optimized TDSS because of the flexible nature of the repertory grid technique. More researches should be carried out in various fields such as in the cost planning and control, scheduling, management, contract planning and control, subcontracting, and quality control. The TDSS developed in this research is validated and is able to be commercialized. References [1] X. Ning, K.C. Lam, M.C.K. Lam, Dynamic construction site layout planning using max – min ant system, Autom. Constr. 19 (1) (2010) 55–65. [2] B. Akinci, M. Fischer, J. Kunz, R. Levitt, Representing work spaces generically in construction models, J. Constr. Eng. Manage. 128 (4) (2002) 296–305. [3] G.A. Kelly, The Psychology of Personal Construct, Routledge, London and New York, 1991. [4] I. Nonaka, H. Takeuchi, The Knowledge-Creating Company, McGraw Hill Company, 1995.

829

[5] D. Bannister, F. Fransella, Inquiring Man, Routledge Kegan & Paul, 1986. [6] M.L.G. Shaw, L.F. Thomas, Focus on education: an interactive computer system for the development and analysis of repertory grids, Int. J. Man Mach. Stud. 10 (1978) 139–173. [7] M.L.G. Shaw, Recent Advances in Personal Construct Technology, Academic Press Inc., London, 1981. [8] M.N. Sarah, A Knowledge Based Expert Systems Based Upon WebGrid-III, PhD Thesis, University Technology Mara (UiTM), 2005. [9] F. Fransella, D. Bannister, A Manual for Repertory Grid Technique, Academic Press Inc. Ltd., London, 1977. [10] M.L.G. Shaw, On Becoming A Personal Scientist, Academic Press Inc. Ltd, London, 1980. [11] J.A. Hartigan, Clustering Algorithms, John Wileys & Sons Inc, 1975. [12] C. Robson, Real World Research: A Resource for Social Scientists and Practitioner– Researcher, Blackwell Publisher Ltd, 1993. [13] H. Said, K. El-Rayes, Optimizing the planning of construction site security for critical infrastructure projects, Autom. Constr. 19 (2) (2010) 221–234. [14] P.S. Hans, Construction Management and P.W.D. Accounts, 6th edHans Publications, Ludhiana, India, 1984. [15] G. Foster, Building Organization and Procedure, 2nd ed.Longman's, London, 1986. [16] I.D. Tommelein, R.E. Levitt, B. Hyes-Roth, SightPlan model for site layout, J. Constr. Eng. Manage. 118 (4) (1992) 749–766. [17] I.D. Tommelein, R.E. Levitt, B. Hyes-Roth, Site-layout modeling: how can artificial intelligence help? J. Constr. Eng. Manage. 118 (3) (1992) 595–611. [18] J. Lee, S. Han, J. Yang, Construction of a computer-simulated mixed reality environment for virtual factory layout planning, Comput. Ind. 62 (1) (2011) 86–98. [19] M.Y. Cheng, S.C. Yang, GIS-based cost estimates integrating with material layout planning, J. Constr. Eng. Manage. 127 (4) (2001) 291–299. [20] M.Y. Cheng, J.T. Connor, Arcsite: enhanced GIS for construction site layout, J. Constr. Eng. Manage. 122 (4) (1997) 329–336. [21] P.P. Zouein, I.D. Tommelein, Dynamic layout planning using a hybrid incremental solution method, J. Constr. Eng. Manage. 125 (6) (1999) 400–408. [22] M.J. Mawdesley, S.H. Al-jibouri, H. Yang, Genetic algorithms for construction site layout in project planning, J. Constr. Eng. Manage. 128 (5) (2002) 418–426. [23] E. Elbeltagi, T.M. Hegazy, Site Layout Model on a Spreadsheet, American Association of Cost Engineers, US, 2000. [24] C.M. Tam, T.K.L. Tong, A.W.T. Leung, W.C. Chiu, Site layout planning using nonstructural fuzzy decision support system, J. Constr. Eng. Manage. 128 (3) (2002) 220–231. [25] B. Glaser, A. Strauss, The Discovery of Grounded Theory, Aldine Publishing Company, Chicago, 1967.