DEMATEL-based model to improve the performance in a matrix organization

DEMATEL-based model to improve the performance in a matrix organization

Wei-Chih Wang

, Yueh-Hua Lin

, Chia-Li Lin

, Chu-Hsuan Chung

, Ming-Tsung Lee

a

Department of Civil Engineering, National Chiao Tung University, 1001, University Road, Hsin-Chu, Taiwan b

Department of Resort and Leisure Management, Taiwan Hospitality & Tourism College, No. 268, Chung-Hsing Street, Feng-Shan Village, Shou-Feng Township, Hualien, Taiwan

a r t i c l e

i n f o

High-tech facility design project

a b s t r a c t

Carrying out a design project for a high-tech facility requires a large amount of special knowledge in order to deal with the specific requirements from the client. Often the client forms a matrix organization, structured with several interrelated functional and project divisions to dictate the various requirements and needs. However, an ill-defined matrix organization is likely to result in late and inadequate decisions. This in turn may result in a poorly performing design. Identifying those divisions within the organization that perform poorly in the way they provide information for the project is critical. Therefore, this study proposes an innovative model to identify those divisions that may be responsible for the poor perfor-mance of the design project. The proposed model integrates the satisfied importance analysis (SIA) and the decision making trial and evaluation laboratory technique (DEMATEL). The SIA evaluates the per-formance of each division, while the DEMATEL captures the causal relationships among divisions to gen-erate an influence-relations map. The proposed model is applied to a real-world high-tech facility design project in Taiwan to demonstrate the strengths of the model.

Ó 2011 Elsevier Ltd. All rights reserved.

1. Introduction

Carrying out a design project for a high-tech facility requires a large amount of special knowledge in order to deal with the spe-cific requirements from the client. Thus, the client must allocate the necessary staff to review all the specification requirements and user needs so that the project architect/engineer can design the project.

Many researchers as well as project managers/engineers/archi-tects have long recognized that the way a construction project is organized will have a significant impact on the successful

comple-tion of the project (Thomas, Keating, & Bluedorn, 1983). Prior to the

design and construction of semiconductor facility projects, the as-signed staff members are often organized in a project task forces, or core teams, guiding the project from the design stage to start-up (Allen, 2000). Sometimes, a project client may form a matrix organization structured with several interrelated functional and project divisions to support the project when its complexity is high and the size of the project is large in scope. However, a matrix organization structured with ill-defined authorities/responsibili-ties for the various managers is very likely to result in late and inadequate decisions. As a result, the design is unlikely to be com-pleted in time or in accordance with the required specifications.

Efficient design management in ensuring the smooth running

of a project is being increasingly appreciated (Austin, Baldwin,

Li, & Waskett, 2000; Bogus, Molenaar, & Diekmann, 2005; Chua, Tyagi, Ling, & Bok, 2003; Karniel & Reich, 2009; Luh, Liu, & Moser, 1999; Sanvido & Norton, 1994; Senthilkumar, Varghese, & Chandran, 2010; Wang & Dzeng, 2005; Wang, Liu, & Liao,

2006). Those divisions within a company that are responsible

for the poor client input into the project (such as late and ambiguous decisions regarding requirements/needs) must be identified. This will allow the client to rectify the situation so that delays can be eliminated or, at least be minimized. Never-theless, identifying the cause-effect relationships among many divisions and sections is difficult since their relationships are interrelated.

This study proposes an innovative model to identify the divisions responsible for the poor performance of the design project. The core of the model combines the satisfied importance analysis (SIA) with the decision making trial and evaluation labora-tory technique (DEMATEL). This DEMATEL-based model is then applied to the real-world design project of a high-tech facility in Taiwan.

The remainder of this paper is organized as follows. Section2

reviews the latest related studies. Section3introduces the steps in

the proposed model, while Section4elucidates the background of

the case project. Section5presents the details of the proposed model

with reference to a case study. Section6discusses the application

results. Finally, in Section7 we draw our conclusions, and offer

recommendations for future research.

0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2011.10.016

⇑Corresponding author. Tel.: +886 3 5712121x54952; fax: +886 3 5716257. E-mail addresses: weichih@mail.nctu.edu.tw (W.-C. Wang),b92501064@ntu. edu.tw (Y.-H. Lin), linchiali0704@yahoo.com.tw (C.-L. Lin),takigawa.cv96g@g2. nctu.edu.tw(C.-H. Chung),greglee@nsrrc.org.tw(M.-T. Lee).

Contents lists available atSciVerse ScienceDirect

Expert Systems with Applications

2. Literature review

The client in the design and the construction of a new facility, often must make two types of decisions related to organizational structure. The first one is to determine which type of contractors (e.g., architect/engineer, consultants, project management, civil/ structure/architect contractor, and mechanical/electrical/plumbing contractor) should be involved in the project and the contractual relationships between these participants. The organization struc-ture is usually decided based on the project management approach adopted, such as the traditional design/bid/build, professional

con-struction management, or turnkey operations (Hendrickson & Au,

2003). In the first type of decision, for example, Cheng, Su, and

You (2003)applied a model to investigate various alternative ways of coordination among the participants in a construction project. They defined a total resistance index to evaluate the coordination efficiencies of each alternative organization in order to identify the optimal structure for the project.

The second type of decision is to determine whom or what group of people from the client should oversee the project. That is, should a functional, a matrix, or a project-oriented organization be applied? For instance, in a functional organization, the project is fully controlled by various functional divisions of the organization. In a project-oriented organization, the project manager takes full control and responsibility of the project. There are many variations of management style between these three organizational forms, depending on the objectives of the client’s organization and the

nature of the project (Thomas et al., 1983).

A few researchers have discussed the second type of decision. For

example,Thomas et al. (1983)described different organizational

forms of project management and outlined the principles to guide the organization in establishing their authority and responsibility. They suggested guidelines to improve the chances for success in the matrix organization. Different goals and situations result in

un-ique organizational requirements for each project. Thus,Tatum and

Fawcett (1986)outlined a logical process for structuring an organi-zation by presenting seven criteria that managers can use to evalu-ate their organizational requirements for the engineering and

construction of a project.Tatum (1986)applied his expanded view

of organizational structure to arrive at a systematic design for the organizational structure for a specific project.

The present study is more related to the second type of decision, which is focused on the organizational structure of the project cli-ent itself. To date however, there is scant literature available on identifying the key divisions of a client’s organization as an effi-cient strategy for improving a design.

3. Proposed model

Fig. 1presents the steps of the proposed model to identify the divisions and sub-divisions of a client’s organization that are responsible for the poor performance in a design project. The steps in the model are as follows.

Step 1: Define the organizational divisions and sections of a client that affect the performance of the design project. Step 2: Use the ‘‘satisfied importance analysis (SIA)’’ to evaluate

the degree of importance and the degree of satisfaction with each division. A division that results in poor perfor-mance is considered to be unsatisfactory.

Step 3: Apply the DEMATEL technique to construct a cause-effect impact-relations map (IRM) for the different divisions. Step 4: Integrate the SIA and DEMATEL. The SIA specifies those

divisions that are most responsible for the poor perfor-mance of the design project, and the DEMATEL traces these divisions.

Step 5: Further evaluation of the problematic sections within these key divisions is carried out by repeating steps 2–4. Step 6: Improve the performance of these unsatisfactory key

divi-sions and sections.

4. Case project

This case project involves the design and construction of a high-tech facility for a national research center (i.e., the project client) located in northern Taiwan. The case project is comprised of three main components; (1) civil and building construction (Civil); (2) mechanical, electrical and plumbing (MEP) works; and (3) special equipment construction (SPE). Designing SPE requires particular domain knowledge, such as synchrotron accelerators and resis-tance against micro-vibrations. The total floor area of the facility

is approximately 53,000 m2_{. The total budget for both the design}

and construction was approximately USD 229.3 million.

At the beginning of the design phase of the project (called de-sign project hereafter), the project client went through several dis-cussions with several consultants on how to structure the organization of the group that to manage this design project. In the end, the client set up a matrix organization because the project size was too large (insufficient staff available to form a project-based organization) and the required expertise too diverse

(involv-ing too many functional divisions).Fig. 2shows the client’s matrix

organization for managing this design project, including seven first-level divisions (three functional divisions and four project divisions) and 10 second-level sub-divisions (called ‘‘sections’’).

It should be noted that the design of the Civil and the MEP phases of this project was contracted out to an architect, and that the construction management (CM) division of the project client managed the architect. In addition, the aforementioned SPE was designed by both the acceleration design & operations (AD) and the acceleration engineering (AE) divisions.

Unfortunately, as the design project progressed, the design was significantly delayed and the design was unable to meet the needs of the user. One of the main reasons for this delay was due to the fact that the client made several decisions that came too late, and the fact that they made frequent changes. As a result, the relation-ship between the different divisions and sections of the client was very tense when it came to the question of responsibility. Since all these divisions and sections were interrelated it was hard for top management to determine the cause and effect relationships among them, making it nearly impossible to take the appropriate corrective action. It was at that time that we began to implement our proposed model to help identify the divisions/sections that

Step 2: SIA analysis - Assessing the degrees of importance and satisfaction with each division Step 1: Defining the functional and project divisions and

sub-divisions (sections)

Step 3: DEMATEL-based analysis - Constructing a cause-effect influence-relations map of divisions Step 4: Integrating the SIA and DEMATEL analyses to

identify key divisions

Step 5: Further evaluating the problematic sections within the key divisions by repeating steps 2~4 Step 6: Improving the performance of the unsatisfactory

key divisions and sections Fig. 1. Steps of the proposed model.

were problematic and to suggest guidelines that hopefully would prevent any further poor performance.

5. Applying the proposed model to the case project

This section will first present the required input data for con-ducting SIA and DEMATEL analyses, and then the analyses will be discussed in detail.

5.1. Collection of the input data

The required input data for performing the SIA and DEMATEL were obtained using two questionnaires. Thirty-five experts (including engineers and division/section managers that were in-volved in this case project) were asked to fill out each

question-naire. Table 1 shows an example of the questionnaire for

conducting the SIA.Table 1also shows the degree of importance

and the degree of satisfaction using a scale ranging from 10 (high-est importance or satisfaction) to 0 (low(high-est importance or satisfaction).

Table 2presents an example of the questionnaire for conducting the DEMATEL. Each respondent was asked to evaluate the impact (i.e., the degree of controlling the decision-making process and the resource allocations) of a division on each of the other divisions

using an integer scale (from 0 to 4).Table 2shows that if division i

(AE) has a weak direct influence on division j (LS), then a score of ‘‘1’’ is given to represent this weak influence. Conversely, if the AD division has a strong direct influence on the CM division, a score of ‘‘3’’ is assigned. A high score represents the belief that an improvement in the CM division depends strongly on an improve-ment in the AD division.

Cronbach’s

a

data collected from the questionnaires in this case project. The test results revealed that the questionnaires used in the SIA and

DEMA-TEL analyses are reliable (since

a

5.2. Evaluation of the SIA

The SIA method is based on the importance-performance

anal-ysis proposed byMartilla and James (1977). In the SIA, the input

data (degree of satisfaction and degree of importance of each divi-sion/section) collected from the questionnaires are normalized to a

single measuring scale. Eqs.(1) and (2)yield the initial degree of

satisfaction (IDS) and the standardized satisfaction value (SS). The number of respondents in the case study was 35

IDS ¼Sum of the degrees of satisfaction from all respondents Number of respondents ;

ð1Þ

SS ¼ ðIDS-Average of the initial degrees of satisfaction in all divisionsÞ

Standard deviation of the initial degrees of satisfaction in all divisions:

ð2Þ Eqs.(3) and (4)calculate the initial degree of importance (IDI) and the standardized importance value (SI)

IDI ¼Sum of the degrees of importance from all respondents Number of respondents ;

ð3Þ

SI ¼ ðIDI-Average of the initial degrees of importance in all divisionsÞ

Standard deviation of the initial degrees of importance in all divisions:

ð4Þ

Fig. 2. The matrix organization of the client for managing the case project.

Table 1

Example of questionnaire for executing the SIA.

Divisions/sections Degree of importance Degree of satisfaction

1. LS 7.7 5.2 1.1 LS1 10 8 1.2 LS2 8 6 1.3 LS3 7 4 1.4 LS4 2 1 1.5 LS5 9 7 1.6 LS6 10 5 2. ID 7.8 7.5 2.1 ID1 8 7 2.2 ID2 8 9 2.3 ID3 8 8 2.4 ID4 7 6 3. RO 10 8 4. AD 10 8 5. AE 9 7 6. EC 9 6 7. CM 9 7

Note: Degrees of importance and satisfaction range between 10 (highest impor-tance or satisfaction) and 0 (lowest imporimpor-tance or satisfaction).

Table 4shows the evaluations made using SIA in the case study. The evaluations are classified under the following four categories (SS, SI):

(1) s (+, +): a division with high satisfaction and high impor-tance. This category is labeled as ‘‘keep up the good work’’. (2)  (+, ): a division with high satisfaction and low impor-tance. A division in this category requires no further improvement. However, it is likely that the resources invested here may be better diverted elsewhere.

(3) . (, ): a division with low satisfaction and low impor-tance. This category is labeled as ‘‘low priority’’.

(4) (, +): a division with low satisfaction and high importance. This category is labeled as ‘‘concentrate here’’, indicating that the division requires urgent corrective actions.

In this case study, only the AE division (acceleration engineer-ing) nearly falls in the fourth category. That is, the AE division is

considered to be highly important, but is providing a low degree of satisfaction. Therefore, the AE division should be improved

immediately.Fig. 3shows the results of the SIA evaluation.

Divi-sions (including the LS, RO and CM diviDivi-sions) that are in the third quadrant are easily identified as requiring improvement, but as having a low priority.

5.3. Evaluation of the DEMATEL

In the proposed model, the end product of the DEMATEL process is an influence-relations map (IRM) that is a visual representation of the interdependencies of the divisions within the organization. The Battelle Memorial Institute of Geneva developed the DEMATEL method for a Science and Human Affairs Program to solve complex

and interrelated problems (Gabus & Fontela, 1973; Li, 2009; Lin &

Tzeng, 2009). The DEMATEL method enables management to solve problems visually and to isolate the related variables (e.g., divi-sions or sections) into cause and effect groups in order to improve the understanding of the causal relationships among these

vari-ables (Li, 2009). This method has been employed in numerous

fields, including the assessment of the structure of a software

Table 3

Reliability tests for the questionnaires.

Tests Cronbach’s

alpha

Results

In SIA analysis:

Data on the degree of importance 0.9225 High Data on the degree of satisfaction 0.9520 High In DEMATEL analysis:

Data on first-level divisions 0.9838 High Data on the second-level sections of the LS

division

0.9843 High

Data on the second-level sections of the ID division

0.9584 High

Table 4

Degrees of satisfaction and importance of the first-level divisions.

Divisions Degree of satisfaction Degree of importance (SS, SI) Initial value SS Initial value SI

1. LS 6.790 1.218 8.200 0.970 .(, ) 2. ID 7.179 0.243 8.200 0.970 (+, ) 3. RO 7.057 0.214 8.286 0.799 .(, ) 4. AD 7.571 1.722 9.514 1.654 s_{(+, +)} 5. AE 7.114 0.001 9.029 0.685 s_{(+, +)} 6. EC 7.257 0.539 8.943 0.513 s(+, +) 7. CM 6.829 1.074 8.629 0.114 .(, ) -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Standardized satisfaction value (SS)

Standardized importance value (SI)

5. Acceleration engineering (AE) 4. Acceleration design &

operation (AD) 6. Electronics & control (EC) 7. Construction management (CM) 1. Light source (LS) 2. Instrumentation development (ID) 3. Radiation &

operation safety (RO)

Fig. 3. SIA analysis of the first-level divisions. Table 2

Example of questionnaire for performing the DEMATEL.

system design for supervisory control systems (Hori & Shimizu,

1999), the reprioritization of failures in a system failure mode

(Seyed-Hosseini, Safaei, & Asgharpour, 2006), the development of

competencies of global managers (Wu & Lee, 2007), the evaluation

of the performance of e-learning programs (Tzeng, Chiang, & Li,

2007), the reconfiguration of policy portfolios in the field of

semi-conductor intellectual property (Huang, Shyu, & Tzeng, 2007), the

evaluation of airline safety (Liou, Tzeng, & Chang, 2007), the

assess-ment of the interrelated ranking criteria in the ranking problem of

a real estate agent service quality (Tseng, 2009), and the evaluation

of value-creating industrial clusters in a science park (Lin & Tzeng,

2009).

The steps in the DEMATEL method are as follows (Lin & Tzeng,

2009); (1) Step D1 – find the average matrix, (2) step D2 – calculate

the direct influence matrix, (3) step D3 – calculate the indirect influence matrix, (4) step D4 – derive the total influence matrix, and (5) step D5 – obtain the influence-relations map.

5.3.1. Step D1: Finding the average matrix

Suppose h experts are available to solve a complex problem with n organization divisions being considered. The scores assigned by each expert yield an n  n non-negative answer matrix

Xk_{, with 1 6 k 6 h. Hence, X}1_{, X}2_{, . . . , X}h_{are the resulting matrices}

for each of the h experts, and each element of Xkis an integer,

de-noted as xk

ij. The diagonal elements of each resulting matrix X

k

are all set to zero. The n  n average matrix A can then be com-puted by averaging the h experts’ value (or score) matrices. The

(i, j) element of the average matrix A is denoted as aij(the average

influence), aij¼ 1 h Xh k¼1 xk ij: ð5Þ

Table 5presents an initial average matrix A of the divisions in

the case project. The value of h in Eq.(5)is 35 (35 respondents).

For example, the initial average value of the effect of the AE divi-sion on the CM dividivi-sion (AE ? CM) is calculated to be 2.086, indi-cating a medium direct influence (since 4.0 is the highest value). 5.3.2. Step D2: Calculating the direct influence matrix

A direct influence matrix D is obtained by normalizing the aver-age matrix A. That is,

D ¼ sA; ð6Þ

where s is a constant, which is calculated as follows (Tzeng et al.,

2007); s ¼ Min 1 max16i6nPn_j¼1jaijj ; 1 max16j6nPn_i¼1jaijj " # ; i;j ¼ 1;2;...;n: ð7Þ

InTable 5, the sum of the fourth row is the maximum value (11.943) of max16i6nPnj¼1jaijj and max16j6nPni¼1jaijj. Thus, the value

of s equals 1/11.943. Then, based on Eq.(6)and the values of the

average matrix A (Table 5), a direct influence matrix D is obtained

and presented inTable 6. For example, the value of the direct

influ-ence AE ? CM is calculated to be approximately 0.175(=2.086/ 11.946).

Let us suppose that the (i, j) element of matrix D (denoted as dij)

is the direct influence by division i on division j. Then,

limn?1Dm= [0]nn, where D = [dij]nn (Goodman, 1988). In

addi-tion, 0 <Pn_j¼1dij;Pni¼1dij61, and the sum of only one row or

col-umn equals one.

5.3.3. Step D3: Calculating the indirect influence matrix

The indirect influence of division i on division j declines as the

power of the matrix increases, as in D2_{, D}3_{, . . . , D}1_{(Lin & Tzeng,}

2009). This fact guarantees convergent solutions to the matrix

inversion, similar to an absorbing Markov chain matrix. The inrect influence matrix ID can be obtained from the values in the di-rect influence matrix D. That is,

ID ¼ D2þ D3þ    ¼X

1

i¼2

Di¼ D2ðI  DÞ1; ð8Þ

where I is the identity matrix.Table 7shows the calculated indirect

influence matrix ID of the divisions in the case project, using the

calculation functions fromMATLAB (2009).

5.3.4. Step D4: Deriving the total influence matrix

The total influence matrix T is also an n  n matrix, and is given byLi (2009), T ¼ D þ ID ¼ D þ D2 þ D3þ    ¼X 1 i¼1 Di ¼ DðI  DÞ1: ð9Þ

Let tijbe the (i, j) element of matrix T; the sum of the ith row and

the sum of the jth column, diand rj, respectively, are obtained as

follows: di¼ Xn i¼1 tij ði ¼ 1; 2; 3 . . . ; nÞ; ð10Þ rj¼ Xn j¼1 tij ðj ¼ 1; 2; 3 . . . ; nÞ: ð11Þ Table 5

Initial average matrix A of the divisions.

Divisions 1. LS 2. ID 3. RO 4. AD 5. AE 6. EC 7. CM Sum 1. LS 0 1.657 1.229 2.343 1.686 2.286 1.200 10.400 2. ID 1.543 0 1.143 1.686 2.571 1.571 1.514 10.029 3. RO 1.486 1.371 0 1.571 1.571 1.486 1.800 9.286 4. AD 2.286 2.086 1.543 0 2.143 2.114 1.771 _11.943 5. AE 1.686 2.657 1.257 1.829 0 1.657 2.086 11.171 6. EC 1.971 1.629 1.314 1.800 1.657 0 0.943 9.314 7. CM 0.914 1.429 1.371 1.171 1.771 1.057 0 7.714 Sum 9.886 10.829 7.857 10.400 11.400 10.171 9.314 11.943 Table 6

Direct influence matrix D of the divisions.

Divisions 1. LS 2. ID 3. RO 4. AD 5. AE 6. EC 7. CM Sum 1. LS 0 0.139 0.103 0.196 0.141 0.191 0.100 0.871 2. ID 0.129 0 0.096 0.141 0.215 0.132 0.127 0.840 3. RO 0.124 0.115 0 0.132 0.132 0.124 0.151 0.778 4. AD 0.191 0.175 0.129 0 0.179 0.177 0.148 _1.000 5. AE 0.141 0.222 0.105 0.153 0 0.139 0.175 0.935 6. EC 0.165 0.136 0.110 0.151 0.139 0 0.079 0.780 7. CM 0.077 0.120 0.115 0.098 0.148 0.089 0 0.646 Sum 0.828 0.907 0.658 0.871 0.955 0.852 0.780

Here direpresents the sum of the direct and indirect influences

of division i on the other divisions, and rjdenotes the sum of the

direct and indirect influences on division j by the other divisions.

Table 8presents the total influence matrix obtained using Eq.(9).

The table also provides the values of diand rj.

When j = i, di+ riis an index of the strength of the influences by

and on a division, and is a measure of the importance of that

divi-sion. The term di ri(also called ‘‘Relation’’) disaggregates the

divi-sions into a ‘‘cause group’’ and an ‘‘affected group’’. If di riis

positive, then division i influences other divisions more than it is influenced by them, and thus it belongs to the cause group.

Con-versely, if di riis negative, then division i is influenced by more

divisions than it influences and thus it belongs to the affected

group (Liou et al., 2007; Tzeng et al., 2007).Table 9shows the

val-ues of di+ riand di rifor each division. The results demonstrate

that the values of di riare positive for the LS, RO and AD divisions,

which therefore fall into the cause group. In general, if manage-ment wishes to improve the divisions that belong to the affected group, then they must take great care to also control the cause group divisions.

5.3.5. Step D5: Obtaining the influence-relations map (IRM)

To visualize the complex causal relationships among divisions using a visible structural model, one can develop an IRM from the values of d + r and d  r, represented on the x-axis and the

y-axis, respectively (Lin & Tzeng, 2009).Fig. 4presents the IRM of

the first-level divisions in the case project. Using this map, man-agement can visualize the difference between the cause divisions

(LS, RO and AD) and the affected divisions (ID, AE, EC, and CM). Fur-thermore, the following net influence matrix N can be used to eval-uate the strength of the effect of one division on another:

N ¼ Netij¼ tij tji: ð12Þ Table 10shows the net influence matrix of the divisions in the case project. For example, the net influence of the CM division on

the AE division is 0.108(=0.708–0.816;Table 8).

5.4. Integration of SIA and DEMATEL

Integrating the SIA and DEMATEL generates various manage-ment strategies to improve the performance of the divisions. These strategies are as follows:

(1) Strategy A (SS > 0 and SI > 0): the division requires no further improvement;

Table 7

Indirect influence matrix ID of the divisions.

Divisions 1. LS 2. ID 3. RO 4. AD 5. AE 6. EC 7. CM Sum 1. LS 0.678 0.721 0.535 0.674 0.746 0.664 0.625 4.643 2. ID 0.636 0.715 0.518 0.662 0.703 0.652 0.602 4.487 3. RO 0.584 0.642 0.484 0.609 0.663 0.598 0.546 4.126 4. AD 0.723 0.797 0.592 0.787 0.826 0.746 0.689 5.158 5. AE 0.687 0.734 0.560 0.715 0.804 0.705 0.641 4.846 6. EC 0.593 0.654 0.484 0.623 0.678 0.631 0.573 4.237 7. CM 0.506 0.547 0.401 0.523 0.560 0.515 0.484 3.534 Sum 4.407 4.810 3.574 4.592 4.979 4.511 4.159 Table 8

Total influence matrix T of the divisions.

Divisions 1. LS 2. ID 3. RO 4. AD 5. AE 6. EC 7. CM Sum (d) 1. LS 0.678 0.859 0.638 0.870 0.888 0.855 0.725 5.514 2. ID 0.765 0.715 0.613 0.803 0.918 0.783 0.728 5.327 3. RO 0.708 0.757 0.484 0.740 0.795 0.723 0.697 4.904 4. AD 0.914 0.972 0.721 0.787 1.005 0.923 0.837 6.158 5. AE 0.828 0.957 0.665 0.868 0.804 0.844 0.816 5.781 6. EC 0.758 0.791 0.594 0.774 0.817 0.631 0.652 5.017 7. CM 0.582 0.666 0.515 0.621 0.708 0.604 0.484 4.180 Sum (r) 5.234 5.717 4.232 5.463 5.934 5.363 4.939 Table 9

Degree of total influence of the divisions.

Divisions d r d + r d  r Group 1. LS 5.514 5.234 10.748 0.280 Cause 2. ID 5.327 5.717 11.043 0.390 Affected 3. RO 4.904 4.232 9.135 0.672 Cause 4. AD 6.158 5.463 11.622 0.695 Cause 5. AE 5.781 5.934 11.715 0.153 Affected 6. EC 5.017 5.363 10.380 0.345 Affected 7. CM 4.180 4.939 9.119 0.759 Affected 9 9.5 10 10.5 11 11.5 12 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 d+r d-r 1. Light source (LS) 4. Acceleration design & operation (AD) 3. Radiation &

operation safety (RO)

5. Acceleration engineering (AE) 2. Instrumentation development (ID) 6. Electronics & control (EC) 7. Construction management (CM) 0.094 0.07 0.044 0.06 0.097 0.143 0.144 0.169 0.039 0.007 0.062 0.019 0.13 0.129 0.181 0.137 0.216 0.149 0.027 0.108 0.049

Fig. 4. Influence-relations map of the first-level divisions.

Table 10

Net influence matrix N of the divisions.

Divisions 1. LS 2. ID 3. RO 4. AD 5. AE 6. EC 7. CM 1. LS 2. ID 0.094 3. RO 0.070 0.144 4. AD 0.044 0.169 0.019 5. AE 0.060 0.039 0.130 0.137 6. EC 0.097 0.007 0.129 0.149 0.027 7. CM 0.143 0.062 0.181 0.216 0.108 0.049 Table 11

Suggested strategies for improving the performance of the divisions.

Divisions SIA DEMATEL Strategies

SS SI (SS, SI ) d + r d  r Group 1. LS 1.218 0.970 .(, ) 10.748 0.280 Cause E 2. ID 0.243 0.970 (+, ) 11.043 0.390 Affected B 3. RO 0.214 0.799 .(, ) 9.135 0.672 Cause E 4. AD 1.722 1.654 s(+, +) 11.622 0.695 Cause A 5. AE 0.001 0.685 s(+, +) 11.715 0.153 Affected A 6. EC 0.539 0.513 s(+, +) 10.380 0.345 Affected A 7. CM 1.074 0.114 .(, ) 9.119 0.759 Affected F

(2) Strategy B (SS > 0 and SI < 0): the division requires no further improvement, but its resources may be better used elsewhere;

(3) Strategy C (SS < 0, SI > 0 and d  r > 0): the division must be improved directly, and with top priority;

(4) Strategy D (SS < 0, SI > 0 and d  r < 0): the division must be improved indirectly, and with top priority;

(5) Strategy E (SS < 0, SI < 0 and d  r > 0): the division must be improved directly, and has a low priority;

(6) Strategy F (SS < 0, SI < 0 and d  r < 0): the division must be improved indirectly, and has a low priority.

Table 11 presents the evaluation obtained using the SIA and DEMATEL methods. In the SIA analysis, the LS, RO, and CM divi-sions have an unfavorable performance (SS < 0) and must be im-proved. In addition, the DEMATEL analysis indicates that the LS and RO divisions are in the cause group, while the CM division is in the affected group. Thus, the LS and RO divisions should be im-proved directly, but with low priority based on strategy E (SS < 0, SI < 0 and d  r > 0) while the CM division should be improved indirectly with low priority according to strategy F (SS < 0, SI < 0 and d  r < 0). The AE division may be improved indirectly with top priority (strategy D: SS < 0, SI > 0 and d  r < 0) because its SS value is almost negative.

Fig. 5shows the evaluation. The left of the figure presents the SIA analysis, whereas the right shows the DEMATEL analysis. It shows that the RO division affects all other divisions. Thus, deter-mining the specifications/needs of the radiation and operation safety (RO) in a timely and positive manner is critical to the design management.

5.5. Identifying the second-level sections

Improving the LS division also depends on improving those sub-divisions (sections) that are responsible for the poor performance. The input data collected from the aforementioned questionnaires and the same modeling steps that are used in applying the SIA and DEMATEL methods yield the following results.

 The performance of the following three sections, LS3, LS4, and LS6 is not satisfactory.

 LS3 (power supply) and LS4 (accelerator operation) must be improved indirectly and with low priority as suggested by strat-egy F (SS < 0, SI < 0 and d  r < 0).

 The LS6 (injector) section must be improved indirectly with top priority based on strategy D (SS < 0, SI > 0 and d  r < 0).

Table 12 shows the suggested strategies for improving the

second-level sections in the LS division.Fig. 6shows the evaluation

results of the SIA and IRM analyses of these sections. 6. Discussion

The above evaluation results were presented to two top managers of the project client. Their main feedback was as follows.

 The modeling results are useful in systematically identifying the key divisions/sections that perform poorly.

9 9.5 10 10.5 11 11.5 12 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 d+r d-r 1. LS 4. AD 3. RO 5. AE 2. ID 6. EC 7. CM 0.094 0.07 0.044 0.06 0.097 0.143 0.144 0.169 0.039 0.007 0.062 0.019 0.13 0.129 0.181 0.137 0.216 0.149 0.027 0.108 0.049 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Standardized satisfaction value (SS)

Standardized importance value (SI)

5. Acceleration engineering (AE) 4. Acceleration design &

operation (AD) 6. Electronics & control (EC) 7. Construction management (CM) 1. Light source (LS) 2. Instrumentation development (ID) 3. Radiation & operation safety (RO)

Fig. 5. Integration of the SIA and the DEMATEL for the first-level divisions.

Table 12

Strategies for improving the performance of the sections of the LS division.

Sections SIA DEMATEL Strategies

SS SI (SS, SI) d + r d  r Group 1.1 LS1 0.775 0.873 s(+, +) 7.796 1.125 Cause A 1.2 LS2 0.927 0.436 s(+, +) 6.871 0.140 Cause A 1.3 LS3 1.348 0.604 ._{(, ) 7.171 0.443 Affected F} 1.4 LS4 0.792 1.745 ._{(, ) 8.730 0.517 Affected F} 1.5 LS5 0.927 0.738 s(+, +) 8.019 0.244 Cause A 1.6 LS6 0.489 0.302 X(, +) 7.117 0.549 Affected D

 A visual diagram (Figs. 5 and 6) is useful for visualizing and clearly communicating the cause-effect relationships among the divisions/sections.

 The top managers appreciated the evaluation results and admit-ted that many of their late decisions as well as the inefficiencies in supervision were caused by an ineffective organizational structure and an excess of conflicting user opinions raised by different divisions or sections.

 The relatively poor performance caused by the LS division might be due to the fact that this division needs to assign a much lar-ger number of staff members to be highly involved in this design project. Hence, some of the staff in the ID division may have to be transferred to support the LS division.

 The top managers considered taking the following three actions to improve the performance of the above mentioned key divi-sions/sections. First, they agreed to clearly define the authori-ties and responsibiliauthori-ties of each division/section manager. The project client actually had no previous experience with a matrix organization. A few division managers had difficulties in work-ing cooperatively with each other. Second, the project client decided that the final decision-maker for the research center must direct and monitor the matrix effort in a firm and decisive manner. That is, he should take full control of the project until the various staff members are familiar with their duties and responsibilities within the matrix organization, and until the project is on the right track. Third, they would recruit additional engineers to alleviate the manpower shortage in some of the key divisions/sections.

 Finally, regular updating of the evaluation of the model is pre-ferred. That is, any additional information obtained by manage-ment should be fed into the model as soon as possible to enable corrections to be made in a timely fashion.

7. Conclusion

A matrix-based organization consisting of many divisions/sec-tions is often used for managing a design/construction project. In a large-scale design project, these divisions/sections are complex and interrelated. Controlling the key divisions that perform poorly is a key to the efficient management of the project. This study pro-posed an innovative model that helps to assess the performance and the cause-effect relationships among the matrix-based divi-sions of a high-tech facility with their design project.

In the model, the SIA method was adopted to evaluate the per-formance of each division, while the DEMATEL analysis was em-ployed to analyze the cause-effect interrelationships among the divisions. Then, the SIA and DEMATEL were combined to determine the key divisions that most strongly affected the design perfor-mance. Top management in the case study were pleased with the results of the presented model.

Future research will computerize the proposed model in order to expedite the evaluation steps, so that the appropriate actions can be taken sooner. In addition, the SIA and DEMATEL schemes may be applied to solve different decision-making problems, involving various interrelated divisions.

Acknowledgements

The authors thank the National Science Council of Taiwan (Con-tract No. NSC99-2221-E-009-132-MY2) and the Ministry of Educa-tion of Taiwan (via the Aim for the Top University program) for financially supporting this research. The valuable assistance and collaboration of the respondents and experts in the case study are greatly appreciated.

References

Allen, B. L. (2000). Wafer fabrication facilities. In Y. Nishi, & R. Doering (Eds.), Handbook of semiconductor manufacturing technology (pp. 1047–1066). Austin, S., Baldwin, A., Li, B., & Waskett, P. (2000). Analytical design planning

technique (ADePT): A dependency structure matrix tool to schedule the building design process. Construction Management and Economics, 18, 173–182. Bogus, S. M., Molenaar, K. R., & Diekmann, J. E. (2005). Concurrent engineering approach to reducing design delivery time. Journal of Construction Engineering and Management, 131(11), 1179–1186.

Cheng, M. Y., Su, C. W., & You, H. Y. (2003). Optimal project organizational structure for construction management. Journal of Construction Engineering and Management, 129(1), 70–79.

Chua, D. K. H., Tyagi, A., Ling, S., & Bok, S. H. (2003). Process-parameter-interface model for design management. Journal of Construction Engineering and Management, 129(6), 653–663.

Gabus, A., & Fontela, E. (1973). Perceptions of the world problematique: Communication procedure, communicating with those bearing collective responsibility. DEMATEL Report No. 1, Geneva, Switzerland, Battelle Geneva Research Center.

Goodman, R. (1988). Introduction to stochastic models. California, US: Benjamin/ Cummings Publishing Company Inc..

Hendrickson, C., & Au, T. (2003). Project management for construction – Fundamental concepts for owners, engineers, architects and builders.<http://www.ce.cmu.edu/ pmbook/>Accessed 05.10.10.

Hori, S., & Shimizu, Y. (1999). Designing methods of human interface for supervisory control systems. Control Engineering Practice, 7(11), 1413–1419.

6.5 7 7.5 8 8.5 9 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 d+r d-r 1.1 LS1 1.5 LS5 1.2 LS2 1.3 LS3 1.6 LS6 1.4 LS4 1.2 -1.2 0.139 0.26 0.299 0.162 0.265 0.074 0.113 0.015 0.108 0.006 0.125 0.022 0.137 0.025 0.129 -1.5 -1 -0.5 0 0.5 1 1.5 -2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Standardized satisfaction value (SS)

Standardized importance value (SI)

1.1 Beam dynamic (LS1) 1.2 Radio frequency (LS2) 1.5 Instrumentation & control (LS5) 1.6 Injector (LS6) 1.3 Power supply (LS3) 1.4 Accelerator operation (LS4)

Huang, C. Y., Shyu, J. Z., & Tzeng, G. H. (2007). Reconfiguring the innovation policy portfolios for Taiwan’s SIP Mall industry. Technovation, 27(12), 744–765. Karniel, A., & Reich, Y. (2009). From DSM-based planning to design process

simulation: A review of process scheme logic verification issues. IEEE Transactions on Engineering Management, 56(4), 636–649.

Li, C. W. (2009). A structure evaluation model for technology policies and programs. Ph.D. Dissertation, Institute of Management of Technology, National Chiao Tung University, Taiwan.

Lin, C. L., & Tzeng, G. H. (2009). A value-created system of science (technology) park by using DEMATEL. Expert Systems with Applications, 36(6), 9683–9697. Liou, J. H., Tzeng, G. H., & Chang, H. C. (2007). Airline safety measurement using a

hybrid model. Journal of Air Transport Management, 13(4), 243–249.

Luh, P. B., Liu, F., & Moser, B. (1999). Scheduling of design projects with uncertain number of iterations. European Journal of Operational Research, 113, 575–592. Martilla, J. A., & James, J. C. (1977). Importance-performance analysis. Journal of

Marketing, 41(1), 77–79.

MATLAB (2009). The Mathworks, Inc.,<www.mathworks.com/patents>, US. Sanvido, V. E., & Norton, K. J. (1994). Integrated design-process model. Journal of

Management in Engineering, 10(5), 55–62.

SAS (2007). Introduction to SAS. UCLA: Academic technology services, statistical consulting group. <http://www.ats.ucla.edu/stat/sas/notes2/> November 24.11.07.

Senthilkumar, V., Varghese, K., & Chandran, A. (2010). A web-based system for design interface management of construction projects. Automation in Construction, 19, 197–212.

Seyed-Hosseini, S. M., Safaei, N., & Asgharpour, M. J. (2006). Reprioritization of failures in a system failure mode and effects analysis by decision making trial and evaluation laboratory technique. Reliability Engineering & System Safety, 91(8), 872–881.

Tatum, C. B. (1986). Designing project organizations: An expanded process. Journal of Construction Engineering and Management, 112(2), 259–272.

Tatum, C. B., & Fawcett, R. P. (1986). Organizational alternatives for large projects. Journal of Construction Engineering and Management, 112(1), 49–61.

Thomas, R., Keating, J. M., & Bluedorn, A. C. (1983). Authority structures for construction project management. Journal of Construction Engineering and Management, 109(4), 406–422.

Tseng, M. L. (2009). A causal and effect decision making model of service quality expectation using grey-fuzzy DEMATEL approach. Expert Systems with Applications, 36(4), 7738–7748.

Tzeng, G. H., Chiang, C. H., & Li, C. W. (2007). Evaluating intertwined effects in e-learning programs: A novel hybrid MCDM model based on factor analysis and DEMATEL. Expert Systems with Applications, 32(4), 1028–1044.

Wang, W. C., & Dzeng, R. J. (2005). Applying cluster identification algorithm and simulation to generate probabilistic network schedules for design projects. Construction Management and Economics, 23(2), 199–213.

Wang, W. C., Liu, J. J., & Liao, T. S. (2006). Modeling of design iterations through simulation. Automation in Construction, 15(5), 589–603.

Wu, W. W., & Lee, Y. T. (2007). Developing global managers’ competencies using the fuzzy DEMATEL method. Expert Systems with Applications, 32(2), 499–507.