A HYBRID QUALITY FUNCTION DEPLOYMENT AND CYBERNETIC ANALYTIC NETWORK PROCESS MODEL FOR PROJECT MANAGER SELECTION

. Identifying appropriate decision criteria and making optimal decisions in a structured way is a complex process. This paper presents an approach for doing this in the form of a hybrid Quality Function Deployment (QFD) and Cybernetic Analytic Network Process (CANP) model for project manager selection. This involves the use of QFD to translate the owner's project management expectations into selection criteria and the CANP to weight the expectations and selection criteria. The supermatrix approach then prioritises the candidates with respect to the overall decision-making goal. A case study is used to demonstrate the use of the model in selecting a renovation project manager. This involves the development of 18 selection criteria in response to the owner's three main expectations of time, cost and quality.


Introduction
Every owner, consultant, and contractor is alert to the need for good project managers, making the task of finding the right project manager for a construction project a critical one for all concerned (Ogunlana et al. 2002). For construction companies, project managers play a challenging and dynamic role (Ochieng et al. 2013) and successful construction organizations focus on ensuring they have project managers with sufficient core competencies (Hwang, Ng 2013). There is a growing recognition, though, that different types of projects require different management approaches, different project management skills and capabilities, and with procedures tailored to the needs of the project (Müller, Turner 2007). Thus, the selection of project managers needs careful consideration. However, it is not easy to identify a suitable person (Kerzner 2001;Rashidi et al. 2011) and current project manager selection methods are known to be inefficient and carry little validity (Rashidi et al. 2011). Although the selection process can be based on the knowledge, preferences and experiences of the decision maker, a more systematic approach is needed (Hsu, Hu 2009). The development of a structured system would be extremely beneficial.
On the other hand, a review of the literature shows that, while many decision models are available for supplier, contractor and sub-contractor selection, considerably less attention has been paid to project manager selection. The traditional interview based project manager selection has been criticized because of its time consuming process and possibility of overlooking a qualified candidate (Jazebi, Rashidi 2013;Rashidi et al. 2011). In addition interviewing is highly subjective and the interviewer may not be sufficiently experienced in interview based decision making to make an accuracy final decision.
In response to these issues, Hadad et al. (2013) propose a decision support system based on Data Envelopment Analysis and the Mann-Whitney-Wilcoxon U test to select candidates according to past performance. Lau (2006) has defined a set of project manager selection criteria and developed an AHP-based project manager selection model. Afshari et al. (2013) develop a decision model-based Fuzzy Simple Additive Weighing (FSAW) approach to select construction project managers. Table 2 lists a number of other studies carried out to identify the characteristics of successful project managers. In summary, the review of the literature revealed very few decision models have been developed for project manager selection and the available decision models that focus on handling subjectivity and complexity of project manager selection lack in two major aspects; the inner dependency of decision elements and accommodating project client demands during model development.
This paper develops a decision model based on Quality Function Deployment (QFD), a customer oriented design tool that enables decision makers to systematically identify project client demands and translate their needs into project manager selection criteria while also taking into account the inter and inner dependencies among decision elements. In addition, Cybernetic Analytic Network Process (CANP) is employed to derive the weights of the client's needs, prioritizing project manager selection criteria, the degree of correlation between decision elements (client's needs and selection criteria) and the overall ranking of project manager candidates.

Quality Function Deployment
QFD is a quality management tool that focuses on customer satisfaction. It is a cross-functional planning tool that ensures customer needs are meaningfully translated into design requirements through a systematic process. QFD was initially conceptualized in the late 1960s and first implemented in the Kobe shipyards of Mitsubishi Heavy Industries Ltd (Akao 1972). A complete QFD process contains a set of four interconnected tables for (1) product planning or house of quality (HoQ), (2) part deployment, (3) process planning and (4) production planning (Liu, Wang 2010). Although QFD has been widely used in manufacturing for quality improvement, it has recently been adopted by the construction industry as a decision making tool for variety of purposes, including supplier selection (Bevilacqua et al. 2006;Taghizadeh 2013), selecting best marketing strategies (Dikmen et al. 2005), contractor selection (Juan et al. 2009) and better architectural design (Gargione 1999;Serpell, Wagner 1997;Yang et al. 2003).
Sorting and focus group discussions. Here, ISM is used as a structured way of understanding the interactions between decision elements in a complex system. The application of ISM during the problem decomposition stage helps to define all possible relationships between elements -first in the second level (criteria) and then in the third level (sub-criteria) of the decision model. Note that in the second QFD table (fourth level of ANP model) there is no correlation between candidates. This is because the candidates are regarded as decision alternatives and are independent of each other therefore no arc is drawn Fig.1.

Fig.1.
A two stage QFD model and its representative ANP network 3.1.1 Interpretive Structural Modelling ISM was first proposed by Warfield (1974). It is an effective tool for analysing a complex system and identifying how an element of a complex system can interact. The foundation of this method is in graph theory and mathematics and it generally helps to have a better understanding of a complex system by dividing the system into several sub-systems. A complete ISM involves five main steps: (a) Identifying the variables that make the system complex -in this case, the client's real demands (second level) and selection criteria (third level).
(b) Forming a contextual relationship matrix to show the relationship between the variables. For every pair of variables, the group of experts are asked to respond to the question: "Is there any relationship between element A and B? If the answer is yes, please select the correct proposition: 1-A helps to achieve B? 2-B helps to achieve A?" The literature offers no guidance on the number of experts (sample size) that should participate in ISM. However ISM is a qualitative technique, the aim of which is the development of a hypothetical structural model which can be further validated statistically with the aid of structural equation modelling (SEM) when generalization is the main objective (Singh, Kant 2008;Chandramowli et al. 2011;Faisal et al. 2006). In addition, 3-7 experts are usually consulted in ISM work (Kumar et al. 2013;Gorane, Kant 2013;Azevedo et al. 2013). Clearly for every pair of variables K (the number of experts involved in the process) answers will be received. In order to form a consensus, a simple algorithm is proposed as follows: It is assumed a consensus has been reached when at least 80% of the answers agree on the existence or nonexistence of a relationship between two variables. Where there is less than 80% agreement, the experts are required to provide at least one reason why they made their judgment. A group discussion then takes place after which the experts are asked to change their judgment if they found it necessary. Consensus is taken to have been reached if there is at least 60% agreement on the existence or non-existence of relationship between the pairs of variables. The process is then repeated with the remaining pairs of variables until a consensus exists for every pair.
Once the relationships between the variables are established in this way, the aggregated qualitative group judgments are transferred into a contextual relationship matrix A ij . For each two elements in row i and column j of this matrix the quantitative value of 1 is assigned if the experts agreed upon the existence of relationship (R ij ); zero otherwise.
(c) Forming the transitive reachability matrix. Transitivity means that when A has relationship with B and B has relationship with C then A should necessarily have relationship with C. To form transitive matrix formula (1), Boolean multiplication and addition is applied (1+1=1*1=1+0=0+1=1, 1*0=0*1=0). The transitive matrix is achieved when two consecutive matrixes become identical. where W ij denotes the contextual relationship matrix of variables, V denotes the variable, R ij denotes the relationship between variables i and j (i,j=1,2,3…m).
where: n=1,2,3…m and I is the identity matrix (d) Depicting the network of interactions (ANP model) in the second and third level; using the transitive reachability matrix, which is a binary matrix comprising 0 and 1, an arrow pointing from V 2 (column element) to V 1 (row element) is drawn if the corresponding cell R 12 =1.

Knowledge acquisition under a Pair-Wiser framework
In this step, since the network and its interactions have been depicted, pairwise comparisons are performed to estimate (a) the local priority vectors ω 1 in which client's interests are compared pairwise with respect to the overall goal. (b) Following the concept of a quality chart, HoQ looks like a house, correlation matrices (W 4 and W 2 ), the roof and porch of HoQ are formed. Note that the relationships between the variables of the two matrices are identified with the aid of ISM. (c) Since two stages of QFD are being modelled in ANP two relationship matrices should be formed. The first is the relationship matrix between the client's interests and selection criteria (W 3 ) and the second is the relationship matrix of candidates and selection criteria (W 5 ). In order to form these matrices, in both QFD, the relative dominance of element i over j with respect to a control criterion are estimated. However, ANP/AHP data collection is always an arduous task, especially when a large number of pairwise decision elements are involved. Recently, however, Chen (2010) has proposed a new cybernetic model of ANP in which only N comparisons instead of N(N-1)/2 pairwise comparisons are performed, which considerably eases the situation.
The concept of pairwise comparisons is to measure the dominance of two elements or two clusters at the same time. Experts are asked to answer the generic question of: given a control criterion (sub-criterion), a component (element) of the network, and given a pair of components (elements), how much more does a given member of the pair influence that component (element) with respect to the control criterion (sub criterion) than the other member? (Saaty, Vargas 2012). In contrast, with the cybernetic model, experts are asked to assume that all decision criteria to be compared in pairs are in a table and to score all the elements on a scale of 1-9 with respect to their common properties. Following the rules given in Table 1, which Chen (2010) calls Pair-Wiser rules, for every two elements the difference between given scores for each pair of elements represents the amount of dominance of element i over element j.
Once judgments based on Pair-Wiser rules are made and then translated into pairwise comparisons, the principal eigenvector is the priority vector of a consistent matrix (Saaty 2003), and the relative dominance or priority vector of the decision elements is estimated by solving . .
(2) where A is the Pairwise comparison matrix, W is the eigenvector and λ max is the largest eigenvalue of A.
After estimating the priority vectors, the consistency of judgments should be checked. This is because the priority vector values in ANP decision theory are estimated, rather than precise, values. The inconsistency in each comparison matrix should to be no greater than 10% to be acceptable (Saaty 1990), which means a consistency ratio (CR) of less than 0.1. Therefore, if CR>0.1 judgments are considered to be inconsistent and the experts should be asked to revise their judgments. This is an iterative process and terminates when all the matrices have an acceptable consistency ratio.  Table 1. Pair-Wiser Rules* *Where PC and PJ denote paired comparison and pairwise judgments. Scores (i and j) are given on the scale of 1 to 9 where 1= Not important, 2= Not to moderately important, 3= Moderately important, 4= Moderately to strongly important, 5=Strongly important, 6= Strongly to very strongly important, 7= Very strongly important, 8= Very strongly to extremely important, 9= Extremely important.

Supermatrix formation and Analysis
In conventional QFD, each HoQ is quantified separately. The priorities of the HOWs obtained in the first HoQ is then transferred to the next quality chart (HoQ) and are regarded as priorities of WHATs. When the number of HoQ is increased, in practice the number of computations is also increased -making the quantification process more subjective and problematic. The supermatrix of ANP, however is able to compute and facilitate quantification of multiple HoQs in one single matrix. The supermatrix generally shows the influence priority of the decision elements on the left of the matrix over an element at the top of matrix (Saaty 2004). Obtaining the overall priorities of decision alternatives with respect to goal is a quite difficult. After forming the supermatrix, the priority vectors and pairwise comparison matrices are transferred into corresponding cells in order to form an unweighted supermatrix. Note that, decision network of QFD is not a pure ANP network (a network wherein the goal does not exist and the priorities of decision criteria are driven by pairwise comparison of the criteria with respect to alternatives or other criteria-feedback) but is a special type of hierarchical thinking structure wherein inner dependence exist between decision elements. When a hierarchical structure is solved with ANP, the identity matrix of I is added in the last row and column of the unweighted super matrix (Saaty 2004). Three steps are then followed in order to estimate the final priority of decision alternatives, (1) normalizing each column of the unweighted supermatrix in order to obtain a column stochastic matrix and (2) raising this matrix to the power of a large arbitrary number to obtain the limit supermatrix with steady status (3) According to Saaty (2009), since the priorities of intermediate decision elements after raising the weighted supermatrix power shifts to zero, to display the priorities of decision criteria, sub-criteria and alternatives with respect to the overall goal, sum the 1 st , 2 nd … and N th power of supermatrix then the final matrix should be column normalized to one again. This matrix is where the priorities of decision criteria, sub-criteria and alternatives with respect to the overall goal are displayed.
Based on the network illustrated earlier, supermatrix W is given by:

Case study
In Malaysia, refurbishment and renovation work are categorized as demolition projects based on the definition in the MS2318:2012 Malaysian Code of Practice for Building Demolition. Although its contribution to the economy in terms of turnover is almost half that of construction (Davidson, Leather 2000), it is labour intensive, very technical, complex in nature and involves many risky activities (Egbu 1999). The Jaya Supermarket and Terengganu Stadium collapse, for example, are two tragic demolition accidents which resulted in a number of fatal and non-fatal injuries (Michael et al. 2013). Compared with the equivalent in developed countries, the demolition industry in Malaysia is considered to be immature; however, it is a fast growing industry. Because of the challenging nature of this industry wherein demolition and construction activities are performed concurrently, the project environment is more complex and uncertain. The case study of a large scale renovation project was therefore selected to demonstrate the application of the proposed model. The case study involved the renovation of the internal layout of Johor Baharu City Square. This landmark was completed in 1998 and consists of a 36 story office building (the tallest building in Johor city), a five story shopping complex and an underground car park. It is a modern, convenient location close to the Singapore-Malaysia customs and immigration checkpoint, with ease of access to public transport, containing more than 200 retailers (local and international brands) and with many food outlets attracting local and international shoppers. The renovation included creating additional lifts, staircases and a new glass roof. The major renovation work was located in the centre of the building associated with structural changes. The slabs, beams and columns of the fifth floor had to be completely removed to create an open space. Most of the shops were to be temporarily closed during the renovation process.
On the other hand, the renovation, maintenance and upgrade of any facility usually involves full or partial service interruptions that result in user dissatisfaction. This issue has several negative impacts on business and can ruin reputations if shoppers are affected. These are the main concerns of every renovation project client, and especially in the case of Johor Baharu City Square, which is an important and iconic Johor State business and tourism attraction.
Ownership of Johor Baharu City Square is shared among the Johor Baharu City Council (MBJB), Johor Baharu Central Municipal Council (MBJBT) and the government of Singapore Investment Corporation (GIC). GIC is the investment arm of the Singapore government that owns 70% of the shares. The other 30% of shares are divided between MBJB and MBJBT, with similar responsibilities such as public health and sanitation, waste removal and management, town planning, environmental protection and building control, social and economic development and the general maintenance functions of urban infrastructure but with different coverage areas. The shareholders have their own office of real estate and asset development and the directors of these offices were directly in charge of the renovation of City Square and appointing the project manager and therefore they are effectively the project clients.
Since the aim is to show how a QFD-CANP model can be used for project manager selection, judgment sampling (a non-probability sampling method) is used (Sekaran 2010;Chang et al. 2007;Kang et al. 2011;Lam, Zhao 1998;. A team of experts was formed comprising one project manager with over 12 years of active experience, holder of M.Sc degree in construction management and P.M.P certification (E1), one associated processor of civil engineering specialising in demolition with a Ph.D and B.Eng in civil engineering and M.Eng in engineering management (E2), and three industrial Ph.D students with postgraduate and undergraduate degrees in project management and civil engineering, together with 4, 6 and 6 years of industrial working experience including project management consultancy, safety audition, quality control, environmental monitoring (E3, E4, and E5). Having an in-depth understanding of project management knowledge and relevant working experience were the two main basic criteria for team membership.

Phase I: Decision Model
As discussed in the "Model" section above, ANP decision network of a QFD based project manager selection comprises four levels. Level one represents the decision goal; the overall objective of model and prioritizing decision alternatives. The second level is the main criteria of the decision model (what the clients need). The third level or selection criteria are the way in which the clients' needs are to be answered and final level is the candidates or alternatives.
Three representatives (directors) were individually interviewed. During the interview session they were asked to respond this generic question: "What would you expect from a renovation project manager with respect to characteristics of this project?" Twenty eight vague demands were identified and then summarized. This list was given to the panel of experts and they were asked to first identify and then exclude any similar demands by sorting the demands with the aid of the card sorting technique. Before starting the card sorting session, the panel of experts were briefed on the card sorting technique. In a 90 minute session, the panel of experts grouped the demands into three main categories. Later they were asked to assign a name to each category (translate the vague terms into meaningful criteria). The categories were named time, cost and quality. It is noted that the term quality is a broad term and all environmental, safety and health related demands were sorted under this name. The outcome of this session identified the decision model criteria (2 nd level of Fig.3).
Next, the second translation process was started by conducting a focus group discussion in which the team of experts were asked to determine the measures (project manager competency criteria) of clients' time, cost and quality related demands. The outcome of this stage, 18 selection criteria, shown in Table 2 forms the 3 rd level of Fig.3. Note that although in broad terms these 18 factors are called selection criteria, they are considered as sub-criteria level in the model.

Selection Criteria
Description Reference Job Experience (JE) Job experience reviews an applicant's job background that may help the applicant perform well. Job background may help make the applicant familiar with the working environment and the skills and methods needed to improve their performance. Rashidi et al. 2011;Ling, Loo 2013 Academic Achievement (AA) This can give an overview of applicant's talents and performance and involves reviewing some internal or personal characteristics such as hardworking, responsible, systematic, and intelligent/ knowledge that is important in reaching the owner's goals. Rashidi et al. 2011 Communication Skills (CS) A project manager needs to be equipped with effective written and verbal communication skills. In order for an organization/firm to gain a competitive advantage, managers must strive to increase levels of efficiency, quality, responsiveness to customers/clients, and innovation. Rashidi et al. 2011;Goodwin 1993;Ceran, Dorman 1995 Microsoft Project Software (PMSS) Increasing complexity and demand for better performance in managing a project has forced the players in the construction industry to equip themselves with effective tools. Planning software allows the project manager to generate solutions to complex problems using algorithms, modelling, simulations, data processing, and 'what-if' analysis. Goodwin 1993;Ceran, Dorman 1995;Hegazy et al. 2005 Planning Skill (PS) Planning is an important process used by managers to identify and select appropriate goals and courses of action for an organization. Planning skill is needed for scheduling activities in order to achieve a project's goals on time. Hwang, Ng 2013;Goodwin 1993;Meredith, Mantel 2011;Ceran, Dorman 1995;Ogunlana et al. 2002 Organizing Skill (OS) The project manager creates the structure of working relationships between organizational members that best allow them to work together and achieve goals. Good project managers often devise a productive and effective organizational structure. Hwang, Ng 2013;El-Saba 2001;Meredith, Mantel 2011;Ogunlana et al. 2002 Directing/ Leading (DL) In leading, managers determine the project's direction; state a clear vision for employees to follow, and help employees understand the role they play in attaining goals. The outcome of the leading function is a high level of motivation and commitment from employees in the organization. Goodwin 1993;Ogunlana et al. 2002 Controlling/ Monitoring (Ctrl) In controlling, the project manager evaluates how well the organization is achieving its goals and takes corrective action to improve performance. Good project manager often monitor individuals, Reference departments and the organization to determine if the desired level of performance has been reached.

Conducting Meetings (CM)
This is a basic skill needed for project managers to coordinate their workforce's knowledge of the current project execution and to discuss any problems. Hwang, Ng 2013;Ceran, Dorman 1995 Record Keeping (RK) It is necessary to develop a systematic filing system to ease the search for important documents such as contract documents, plans, daily working records, purchase orders, submittals, drawings, sketches, letters, transmittals, change order proposals, etc. Hegazy et al. 2005;Levin 1998;Hamilton 1991 Time Management (TM) Project managers need to maintain a meaningful schedule to monitor the contractor's work progress and to use as a tool in resolving disputes caused by delays and acceleration. Multitasking is important to same time. Time management is also important because a primary objective of construction work is to minimize time and cost while maintaining project quality. Hwang, Ng 2013;Yasin et al. 2002 Property Management (PM) Property management refers to the performance of the ongoing maintenance or repair of a construction project after the completion of the construction phase. Arditi, Nawakorawit1999;Horner et al. 1997;Chanter, Swallow 2007 Worker welfare Management (WWM) The project manager needs to make sure that workers' welfare is considered in order to engender a harmonious working atmosphere and take action to protect workers being injured during construction. They have to bear in mind that workers and employees are the main asset in their project. Ogunlana et al. 2002;Hwang, Ng 2013;Roethlisberger, Dickson 2003;Ofori et al. 2004 Rules and Regulation (RR) It is necessary for a project manager to have a wide knowledge of law and contract regulations. It is important to be familiar with standard construction practice and methods and keep abreast of new developments and changes in the field. The project manager needs to know the rules in pertaining to a contract including the contract clauses, Occupational Safety and Health Act (OSHA) regulations and Code of Conduct. Yasin et al. 2002;Czuchry, Yasin 2003 Problem Solving Skills (PSS) The most common issues related to construction projects are those originating from delays, suspension, or acceleration of the times established in the contract for the completion of work. Site conditions may present as a problem, especially subsurface conditions. If the project manager cannot handle the volume of work involved, this is likely to lead to submittal delays, scheduling conflicts, and a negative impact on the project. Ability to multi-task is therefore a useful skill to have.

Phase II: Knowledge Acquisition
The decision network depicted in phase I (Fig.3) indicates how the knowledge of the experts under the framework of CANP should be elicited in order to satisfy the two main objectives of prioritizing selection criteria and evaluating candidates against weighted criteria. In order to achieve the first objective, three sets of questionnaires were designed to acquire knowledge concerning two different target individuals; the client team's representatives and the panel of experts. The decision criteria (time, cost and quality) are the translation of client vague demands into a meaningful needs. Hence, it is necessary for the client team's representatives to prioritize their needs. The first questionnaire, following the cybernetic ANP framework, was given to client team and they were asked to answer this generic question: "On the scale of 1-9 rank time, cost and quality according to your preference." Thereafter, the second set of questionnaires was given to experts in order to identify the degree to which the selection criteria were able to satisfy the client's needs, since the client team are not sufficiently well informed to judge on this issue. At this stage, the expert panel was asked to respond the question: "Assume the 18 selection criteria are on the table, on a scale of 1 (not important) to 9 (extremely important) rank the degree to which these 18 criteria are able to satisfy the client's time demands." This survey was repeated for quality and cost because the elements of sub-criteria level were connected to all three time, cost and quality criteria.
The final set of questionnaires following the road map shown in Table 3 and Fig. 2 was designed to measure the degree to which criteria and sub-criteria correlate to each other (W 2 , W 4 ). This type of question was asked: "Given the control criterion Crp, which one of the CM, AA, CS and JE is more helpful in achieving Crp? Assume that CM, AA, CS and JE are on the table, on the scale of 1-9 rate these factors according to their degree of influence."

Fig.3. ANP decision network of project manager selection
The second objective was achieved through conducting structured interviews with the six applicants. This involved a set of open-ended questions carefully designed to match the selection criteria. During a 25 minutes interview session with each candidate, they were asked to answer this type of question: "How much the ability to use project planning software helps project manager and how do you describe yourself in this regard?" (question matched with MPSS criterion). The experts were asked to evaluate the candidates based on the given answers on the scale of 1 (not qualified) to 9 (extremely qualified).
To show how the calculations under CANP are performed, Crp was selected as a control criterion. Initially, as shown in Table 4, the five experts were asked to score all four selection criteria correlated with Crp. With the aid of Pair-Wiser rules, the given scores were then translated into pairwise comparisons. The aggregated pairwise comparison matrix of the five experts obtained is shown in Table 5. Next, formula (2) was applied to estimate the relative importance of each of the ten selection criteria. The influence of Crp on each selection criteria was then estimated by solving this equation. Finally, CR=0.00029 as measure for consistency of judgment was computed. Since the value was less than 10%, the estimated priority vectors for AA, CM, CS and JE were considered to be acceptable (Table 6).   The above process should be repeated for all control criteria. After estimation of all the priority vectors, the normalized vectors were transferred to the corresponding cells of supermatrix W in order to create the unweighted supermatrix. The column stochastic matrix was computed and as explained earlier, raised to 15 th power where the steady limit supermatrix was obtained. The 15 matrices were summed and then renormalized to obtain the final limit supermatrix where the priorities of the 27 elements with respect to the overall goal were obtained (Table 7).

Validation of the model
There are many strategies suggested by scholars to validate a study. Fuertes et al. (2013) examined the comprehensiveness, reliability, suitability and diagnosticity of their model in order to test content, face and construct validity. The authors argue that construction management related models are costly to validate since they need a large amount of data or multiple implementation. Statistical analysis is a fundamental necessity behind their proposed validation strategy along with measurement of the experts' degree of agreement with the results. Another widely used validation strategy is to compare the output of the model with that of existing models (Gibb et al. 2006;Juan et al. 2009;Sun, Meng 2009). The second strategy was adopted to validate the model and the Serbian VIseKriterijumska Optimizacija I Kompromisno Resenje (VIKOR) method was selected to rank candidates and the final results were compared with the rank order derived by the QFD-CANP project manager selection model. VIKOR is a Multiple Criteria Decision Making (MCDM) tool that is used to rank alternatives (Opricovic, Tzeng 2004). The ranking results are based on the closeness to an ideal solution. This technique is usually incorporated with other MCDM techniques such as AHP/ANP or Delphi method since the criteria of the decision model should be weighted first. The project manager selection criteria had already been weighted with the aid of CANP. Hence, the same priorities were used. The steps according to Büyüközkan, Ruan (2008);Tavakkoli-Moghaddam et al. (2011) to find the most competent candidate were: Step1. A team of experts were formed. The same panel of experts were used since their experience and knowledge helps to generate more reliable results; Step2. The experts were weighted based on their experience and knowledge. This is because different experts have different level of expertise and knowledge. A simple individual AHP-based decision model was developed where knowledge and experience represent decision criteria and five panel members represent alternatives. The criteria with respect to goal and alternatives with respect to each criterion were pairwise compared in order to determine the weighting of each expert. Table 8 shows the results.
(c) Proposing a compromise solution: Sort the alternatives into descending value order based on S j , R j and Q j . Alternative A' (the alternative with the lowest value of Q j ) is proposed as a compromise solution if both C 1 and C 2 are satisfied.
, where m= number of alternatives and A" is the alternative with second lowest Qj. C 2 . "Acceptable Stability ˮ if alternative (A') is the best in the R j and/or S j ranking. A set of compromise solutions is proposed if one of the above conditions is not satisfied. If C 2 fails then the compromise solution is {A' and A"} or if C 2 fails then the compromise solution is {A', A"… A m } and A m is determined by for maximum (m). The propositions of these alternatives are "in closeness ˮ. Table 9 shows the candidates' ranking according to the S, R and Q index in decreasing order. The minimum amount of Q index belongs to the second candidate and since Q (C2) -Q (C1) = 0.572>0.2 the first condition is satisfied. In addition, the second candidate also has the best situation according to the S and R values and therefore the "acceptable stabilityˮ conditions have been also satisfied. Comparison of candidates' ranking in both Table 7 and 9 indicates that the VIKOR ranking results are similar to the QFD-CANP rank order (C3>C6>C5>C4>C2>C1). This similarity of results supports the validity of proposed model. Note that, due to the inability of VIKOR to prioritize client's needs and selection criteria, adding another decision-making technique to prioritize decision criteria and sub-criteria adds more complexity to the process. In contrast, the strong mathematics behind ANP is able to prioritize the client's needs, selection criteria and alternatives of the decision model, which lessens the computations and facilitates the quantification of the two stages of the QFD process if real experts make consistent judgments.

Conclusion and recommendations
Decision making for selection has always involved complexity and uncertainty especially when the decision maker deals with large number of decision criteria and alternatives. Project manager selection also has a complex and uncertain environment wherein the decision maker evaluates several candidates against a series of selection criteria. The more criteria and alternatives, the higher the risk of misjudgment, especially when the decision maker relies on unstructured decision making methods. In the renovation and refurbishment business, where the environment is more technical and dynamic compared with construction because of the simultaneous operation of construction and demolition activities, there is a need for appropriate decision making in the selection of a project manager -who will have considerable and crucial responsibilities for all aspects of project. With the aid of a quality management tool this paper first identifies what project clients need (WHATs) and then determines what specialities project manager should have in order to satisfy these needs (HOWs). A decision problem comprising four levels including goal, criteria, subcriteria and alternatives is formulated and a multiple criteria decision making tool developed with the ability to take into account inter and inner-dependencies. The contribution of this paper is creation of a two-stage QFD based decision making model that, with the aid of CANP, is able to select the most competent candidate in a structured way with less judgments and a stress-free complex computation process. In contrast with common project manager selection methods in which the selection of candidates is based solely on matching project manager competency criteria, this study highlights the direct influence of client demands (what the unique characteristics of the project imply) on selection criteria that indirectly influence the candidates' ranking. It is also concluded that correlation in criteria and sub-criteria level should be accommodated since this helps to obtain more reliable and actualized prioritisation of selection criteria.
A limitation of the model is not taking into account the subjectivity of judgments during deployment of QFD and acquisition of knowledge either in ISM or CANP. Future research is needed to incorporate fuzzy set theories to handle the associated subjectivity. More case studies are also needed to further develop the database of client demands so that more project characteristics are identified and the QFD's translation process to the database of refurbishment project manager competency criteria.

Short CV
Farahbod Mohammadi is currently Ph.D student at faculty of civil engineering Universiti Teknologi Malaysia (UTM). He obtained B.Sc in Civil Engineering and M.Sc in Construction Management from Islamic Azad University(IAU) and (UTM) respectively. He also serves as faculty member (lecturer) at IAU-Fouman&Shaft branch. His main research interests include application of multiple criteria decision making theories in construction domain, construction technology, demolition safety assessment and environmental aspect evaluation associated with demolition activities.