Information Science (accept with minor revision

)

Vendor selection by integrated fuzzy MCDM techniques with independent and interdependent relationships

Jiann Liang Yang a, Huan Neng Chiub, Gwo-Hshiung Tzengc,1, Ruey Huei Yeha

a

Department of Industrial Management, National Taiwan University of Science and Technology, 43 Keelung Road, Second 4, Taipei 106, Taiwan jlyang@mail.lit.edu.tw

b

Department of Business Administration, Asia University, 500, Liufeng Road, Wufeng, Taichung 41354, Taiwan; E-mail: hnchiu@asia.edu.tw

c

Department of Business and Entrepreneurial Management, Kainan University, No. 1, Kainan Rd., Luchu, Taoyuan 338 Taiwan; Institute of Management of Technology, College of Management, National Chiao Tung University, 1001, Ta-Hsuch Rd., Hsinchu 300, Taiwan; E-mail: ghtzeng@cc.nctu.edu.tw; ghtzeng@mail.knu.edu; Fax: 886-3-5726749

Abstract Vendor selection decisions have been long considered one of the most important functions of the purchase department. In real-world systems, selecting appropriate vendors should be considered and evaluated in terms of many different criteria resulting in a vast body of data that are often inaccurate or uncertain. Furthermore, the number of evaluation criteria in very complex problems is often too large to determine dependent or independent relationships. However, most conventional decision models cannot be considered for clarifying the interrelations among the

1

Corresponding author: G.H. Tzeng (Distinguished Chair Professor), ghtzeng@cc.nctu.edu.tw ; ghtzeng@mail.knu.edu.tw ; Tel: +886-3-5712121 ext 57505; Fax: +886-3-5753926. 1

sub-criteria of a criterion by virtue of additivity and independence assumptions. In the proposed integrated fuzzy multiple criteria decision making (MCDM) techniques for solving vendor selection problems, we utilize triangular fuzzy numbers to express the subjective preferences of evaluators with respect to the considered criteria, and also use the criterion gauges to evaluate a well-known high-tech manufacturing company. In addition, a relationship map to identify the independence or interdependence of the sub-criteria of a criterion is constructed by using interpretive structural modeling (ISM). Furthermore, the fuzzy analytical hierarchy process (AHP) method is used to derive the relative weights of each criterion, and the fuzzy synthetic performance of each common criterion can also be obtained by applying a non-additive fuzzy integral technique. The best vendor can then be determined according to the overall aggregating score conforming to each candidate vendor by the fuzzy weights with fuzzy synthetic utilities. Finally, an empirical case is illustrated to demonstrate that the proposed method is more suitable than the traditional method, especially when the sub-criteria are interdependent in real situations. Further, the results could provide some valuable opinions on how we can improve each sub-criterion to reduce the gaps between real performance values and aspired/desired values to achieve the best vendor.

Keywords:

vendor

selection,

multiple criteria

decision

making

(MCDM),

independence, interdependence, non-additive fuzzy integral

1. Introduction In today¶s highly competitive environment, it is impossible for a company to produce low cost and high quality products successfully without satisfactory vendors. The selection of appropriate vendors has long been regarded as one of the most
2

important functions to be performed by a purchasing department. In essence, vendor selection is an unstructured, complicated, and multi-criteria decision problem. Over the past two decades, many studies have pointed out that the key to success is to set effective evaluation criteria for the vendor selection problems (VSP). The earlier works on vendor selection identified 23 criteria (i.e., price, delivery, quality etc.) for evaluating and selecting appropriate vendors, and then deciding on the quantities to be ordered [13]. In 47 of 76 articles, the vendor selection decisions used more than one criterion (i.e., multi-criteria) [39]. A vendor selection problem usually involves more than one criterion and criteria often conflict with each other. In multiple criteria decision making (MCDM), it is widely assumed the criteria are independent. A considerable number of decision models have been developed based on the MCDM theory, such as the preference ranking organization method (PROMETHEE) [2], the analytical hierarchy process (AHP) [27,17], discrete choice analysis (DCA) [35], total cost ownership (TCO) [12], and data envelopment analysis (DEA) [40,26]. However, the available information in a multi-criteria decision process is usually uncertain, vague, or imprecise, and the criteria are not independent in the real world. To efficiently grip the vagueness in the available information and the essential fuzziness in human judgment/preference, fuzzy sets theory was first proposed by Zadeh in 1965 [42] and a decision making method in a fuzzy environment was developed by Bellman and Zadeh [1]. Later, a number of studies used this theory to deal with uncertainty in vendor selection problems. Holt [19] applied seven decision methods to contractor selection decisions. The design process pointed out the advantages and disadvantages in the vendor selection model by Morlacchi in 1999 [25]. De Boer et al. [11] provided a comprehensive review of the literature concerning supplier selection. In these papers, fuzzy sets theory was suggested for solving vendor selection problems. Mikhailov [24]
3

proposed the fuzzy AHP method for determining the weight of each criterion and scoring each alternative for each criterion. Kumar et al. [22] presented a fuzzy goal programming approach for solving vendor selection problems with three objectives. To select a suitable partner for strategic alliance, fuzzy sets theory can be applied to analyze and consider a multiplicity of complex criteria in an MCDM environment [14]. Moreover, Shyur and Shih [9] developed a hybrid MCDM method for strategic vendor selection by using both the ANP and TOPSIS techniques. For solving the measurement of qualitative items, an approach was developed using both quantitative and qualitative data for supplier selection intended to create value for the customer [16]. In the real word, this usually exists when the purchase situation is full of uncertainty and the imprecision of human subjective judgment. Thus, in this paper fuzzy sets theory is used to cope with these situations. Basically, an MCDM problem includes five primary elements: alternatives, criteria, outcomes, preferences and information (see Table 1). The multiple criteria decision issue [33] focuses mainly on the distinguishability of the evaluation criteria and on the determination of the preference structure (i.e., weights). Regarding the first issue, previous research concerning the determination of the evaluation criteria used in vendor selection has usually depended on products, services and purchase situations [39,15]. However, the number of evaluation criteria in very complex problems is often too large to determine dependent or independent relationships among these criteria. Thus, a complex system can be divided into many sub-criteria to allow differences to be judged or scores measured more easily. On the other hand, the analytical framework for vendor selection processes in traditional decision models fails to consider the interdependence of sub-criteria. To distinguish the interrelations among sub-criteria in a complex system, a suitable approach called interpretive structural modeling (ISM) [36,37,38] is utilized to decompose the elements in
4

i=1. fij : the outcome of each choice. The process of ISM is based upon Boolean operations of one-to-one correspondence between a binary matrix and a graphical representation of a directed network. Then. ISM may help to identify the structure of related criteria in a system. f1 C1 w1 L L Ci wi M L L Cn wn L fij M (Performance scores) L L fi  L f n 5 . In the above decision problems.large-scale complex systems.n.«. wi : the weights/preference structures of the decision maker as criterion. elements can change and evolve with time and by situation. The definitions of five primary elements in an MCDM problem Criterion Alternative A1 M Aj M Am Aspired level Note that A j : the set of alternatives from which we will choose our decision. the ³information´ of the dynamic change and the evolution of four elements above will be treated in the decision process.«. Furthermore. j=1. The ISM method for identifying relationships among criteria in a vendor selection problem and the interrelations of different criteria and their levels of relative importance was described by Mandal and Deshmukh in 1994 [23]. Table 1. in this paper these concepts are used to build a relation map to identify the independence or dependence of the sub-criteria of a criterion. thus. measured in terms of the criteria. Therefore. Ci : the set of criteria with which we are concerned in order to make a good decision. we can integrate MCDM techniques with additive and non-additive models to evaluate vendor selection.m.

34. The mean of maximal. Because the COA method is a simple and practical method. the weight and the grade of importance are the general forms used to demonstrate the preference structure. 6 . it can be used in nonlinear situations.41. the defuzzification of the fuzzy weight of criteria is used to calculate the best nonfuzzy performance (BNP) value of final weights. the importance-assessing methods used to demonstrate the importance of criteria are often based on the assumptions of addition and independence. However.Furthermore. Even in an objective sense. From the above literature review. we adopt the AHP method [29] to determine the fuzzy weight of each independent criterion in a fuzzy environment.5. any two criteria may be dependent from the subjective viewpoint of decision-makers. If the importance of a criterion can be captured properly. it can be seen that the MCDM process. this paper further applies the Sugeno integral for fuzzy measure and uses a non-additive (µsuper-additive¶) fuzzy integral technique [31. the overall score and ranking of each vendor will determine the choice of the appropriate vendor. Whereas the fuzzy integral model does not need to assume independency of one criterion from another.32. These methods have been successfully applied in various circumstances [18.8]. center of area (COA). and there is no need to introduce the preferences of any evaluators. in fact. it cannot be used for comparison.28] are the three most common defuzzification methods. definite value. we will employ it to determine the BNP weight of each criterion in this study. the quality of the decision-making will be enhanced correspondingly. Conventionally. Finally. Since the fuzzy value is not a clear. people have found that using such an additive model is not always suitable because of the interactions among the criteria to somewhat different degrees. Thus. Therefore. and the E -cut methods [43. Subsequently.7.21] to evaluate the synthetic performance of alternatives to reflect in reality that the decision criteria are inevitably correlated to others with different degrees.

The rest of this paper is organized as follows.particularly in a fuzzy environment. some fundamentals associated with the proposed approach are addressed. We validate the integrated techniques as an effective method for selecting an appropriate vendor. This paper is different from previously published research in three ways. In Section 3. which is a non-additive method. Sugeno¶s fuzzy integral with fuzzy measures. the overall scores for the criteria for each vendor can be obtained to enable the selection of appropriate vendors. Therefore. First. we adopt Warfield¶s ISM method to build a relation map to clarify the interrelations among the sub-criteria of a criterion and to apply this relation map to a complex vendor selection problem. can be used to achieve the goals of practicality. An empirical study of the vendor selection problem in Taiwan is presented to show that the proposed method is useful and effective in real cases. Thus. In Section 5. and objectivity. 7 . is used to calculate the synthetic utility of the weights of the interactive sub-criteria to fit with the patterns of human perception. problems in a fuzzy environment are discussed in detail. The proposed methods are used in this paper to establish an integrated MCDM model with consideration of interrelations and synthetic utility among the sub-criteria of a criterion to evaluate vendor selection problems. and the results are discussed and compared with those obtained using the simple additive weight (SAW) method in Section 4. in this paper fuzzy MCDM techniques were used to evaluate the performance of the candidate vendors for a well-known high-tech manufacturing company in Taiwan as an empirical case. Finally. we conclude this paper with some suggestions for future research. especially when independent or interdependent sub-criteria exist in a complex hierarchical system. accuracy. the weights of criteria can be determined using the fuzzy AHP method and the BNP weights of each criterion can be calculated simultaneously using a defuzzification method. In Section 2. Second.

has a significant impact on a company¶s competitive priorities. The P decision-makers are used to define n criteria through brainstorming. supporting services. the weights of all the criteria and the performance scores of each vendor are equivalent to linguistic variables. ³essentially important´. from which the appropriate vendor is chosen. The vendor selection problem in fuzzy MCDM environments The vendor selection decision process. delivery.m. ³very strongly important´. For these reasons.«. and to distinguish the interrelations among the sub-criteria of each criterion.P. 1. which is complex and usually involves vague information. denoted by Ek. for j=1. denoted by Vj.«. The integrated evaluation hierarchy of the vendor selection process is shown in Fig. It is assumed that each criterion is independent of the others (see Appendix 1. The K evaluators are used to evaluate the performance scores of criteria for each candidate vendor.e. are recruited.«.1). who are outsiders (i.. the linguistic effect values of vendors found in this paper are primarily used to assess the linguistic scorings given by evaluators. and innovation. for k=1. We use this kind of expression to compare two evaluation criteria by linguistic variables in a fuzzy environment as ³absolutely important´. A committee is constituted with P+K evaluators. Similarly. and K experts. In this paper. denoted by Mp. ³weakly important´.K. we consider an MCDM vendor selection problem in a fuzzy environment. such as price. quality. 2). To achieve a company¶s purchasing objective. the vendor selection problem may involve m candidate vendors.2. for p=1. The linguistic variables are used as a way 8 . in which P decision makers from cross-functional divisions of the company. to determine the weights of criteria. consultant companies). and ³equally important´ on a fuzzy five-level scale (see Fig. which can be measured by means of several linguistic terms represented by triangular fuzzy numbers (TFNs).

³bad´. ³good´. 2 Membership functions for the linguistic variable defined in this paper (an example)  qually important Weakly important ssentially important ery strongly important 5 9  " ad Fair ood bsolutely important 7  !  Q . and ³very good´. ³fair´.to measure the scores of vendors for each criterion as ³very bad´. ub-criteria lari y the interrelations o sub-criteria by the I method alculate the synthetic utility o sub-criteria using the uzzy integral technique uij ! ´ hij dg P i etermine the eights o criteria rom the method spired level Fig. Table 2 is a format for identifying a TFN to present the membership functions on a five-level scale. 1 The integrated evaluation model of vendor selection 1 1 3 Fig.

.C1t C21.Cnt .Cn2. .C22. . wn verall score i ij §wu ap minimize ery good x 9 . . w1 w2 . .C2t .C12. Cn1. .x ery bad    ¥ £ ¦  ©¨    ¤ £   ¡ £  § £ ¢ ¢  oal The best vendor riteria C1 C2 . . . . . . Cn C11.

Table 2 Linguistic expression for fuzzy scale of relative weights of criteria and performance values of alternatives (an example of one judge) Intensity of fuzzy Definition of linguistic variables for scale (example) relative weights of criteria Equally important Weakly important Essentially important Very strongly important Absolutely important Definition of linguistic variables for performance values of alternatives Very bad Bad Fair Good Very good % 1 ! .

1. 3 % 3 ! .1.

3. 5 % 5 ! .1.

5.3. 7 % 7 ! .

5.9 % 9 ! . 7.

4. 9.7.8 Intermediate values between two adjacent judgments Used to represent compromise between the priorities listed above The weight of criterion i given by the evaluator Mp is denoted by % wip ! . 9 %%%% 2.6.

P. for i=1. a TFN is defined as . uip . and p=1. For criterion i.lip . the fuzzy performance k % score sij of candidate vendor j in terms of TFNs given by evaluator Ek is denoted by k ij .«.U analysis.n. k ij . mip .«.

mip . the real number can be represented as intervals with equal lower and upper bounds. That is.mip  V . where mip is the mean of the TFN and V is the spread and is a positive number. sup Q . mip  V . Its membership function is assumed to be normal (i.e..

Note that the value of V depends on the characteristics of the criterion. within the real interval «lip . and the evaluator Mp can subjectively ­ ½ give his/her range of the linguistic variable ³weakly important´=(1. On the other hand. the expression ³quality´ represents a linguistic variable in the company and it may take on values such as ³weakly important´. for the performance scores of vendor j. The membership functions can denote the degree of truth that a TFN is equal to x. uip » . a TFN is defined as 10 # $ # # k % sij ! .3. For example.5).x ! 1 ).

«.K.m.«. for j=1. to simplify the .L k ij . and k=1. For each criterion.

the decision-maker¶s range for the linguistic variable ³very good´=(85. k ij . k ij . Some fundamentals of the proposed integrated approach In this section. 3.k ij I. To use a TFN to represent a linguistic variable. any interference with one of the sub-criteria affects all the others. These fundamentals include the methodology used to clarify the interrelationships among the sub-criteria of a criterion. respectively.. for example. directly or indirectly. 3.90. Thus. all sub-criteria of the systems are mutually related. some important fundamentals that are used in the proposed method (see Section 4) are addressed. a reachability matrix is derived from the adjacency matrix by adding the identity matrix then raising the resulting matrix to successive powers until no new entries are obtained. the left and right spreads and both are real numbers which depend on the subjective perception of the decision-maker.1 Clarifying the interrelationships among sub-criteria of a criterion In a completely interdependent system. a TFN (i. if a company uses the yield rate to measure the quality and the linguistic variable ³very good´ is given for the yield rate by a decision-maker. the concept of determining the fuzzy weight for each criterion. To clarify the interrelationships among the sub-criteria of a criterion.100) can be determined.U values of I and H as 5 and 10.e. For the sub-criteria of criterion i. Thus. an adjacency 11 ' ( ' ' k % sij ! . and the principle of calculating the synthetic utility with the interactive sub-criteria.

L &% &% k ij &% .

and I and H are k ij ) within the scale range of 1-100. decision-maker Mp sets the . . where MEij is the mean of the TFN. k ij k  H .

e. Next. l>1 stops (stable reachability). Subsequently. Consequently. To get the consensus opinion of all evaluators. furthermore. 0 or 1) of the comparisons among sub-criteria is called the mode.K . in this paper. otherwise.e. l denotes the powers. we use a mode method to calculate the value of the opinions of evaluator Mp for the relationships among sub-criteria in the adjacency matrix: if the majority evaluator answer is ³1´..matrix (i. the value of the relationship for the sub-criterion is ³0´. which represents the sub-criteria not being related. Otherwise. M¹ ¹ 0 ¹t vt º p ¨ 0 e12 © p A ! Ci 2 © e21 0 M © M M © Cit © etp etp2 ª 1 (1) p where err ' denotes the value of the relation between the rth row and the r th column sub-criteria given by evaluator Mp. respectively. which represents the sub-criteria being related. if the majority evaluator answer is ³0´. the mostly frequent value (i. P . The former includes the 12 ) 1 0 2 (2) (3) . relation matrix) can be constructed by evaluator Mp. The general form of the adjacency matrix A can be expressed by Ci1 C i1 Ci 2 L L L O L Cit e1p ¸ t p ¹ e2t ¹ . p ! 1. Then. Then. If the answer given by evaluator Mp for p sub-criterion Cir inflecting the sub-criterion Cir is ³Yes´. where I is the identity matrix. the stable reachability set (Rr) and the priority set (Ar) can be calculated based on Eqs. p the value of err ! 0 is given. when T l ! T l 1 . err ! 1 . the adjacency matrix A can be determined. t . the value of the relationship for the sub-criterion is ³1´. (4) and (5). this represents the intensities of different dependencies among sub-criteria.. the reachability matrix T can be calculated by T ! ( A  I ).K . r ! 1. and T l denotes the stable reachability matrix. then. if the majority evaluator answer is ³1´.

1+0=0+1=1. Thus. as follows: Ci1 C i1 A ! Ci 2 Ci 3 Ci 4 ¨0 © ©1 ©1 © ª0 Ci 2 1 0 1 0 Ci3 0 0 0 0 Ci4 0¸ Ci1 ¹ . * Rr ! Cir errp' ! 1 . _ _ a a * Ar ! Cir er 'p ! 1 .e. the adjacency matrix is added to the identity matrix to form a tentative reachability matrix T at l=1. (4) and (5). the values of the adjacency matrix A between sub-criteria given by evaluator Mp can thus be represented as below. the multi-level diagraph of the relationships among sub-criteria can be shown as follows: Rr ‰ Ar ! Rr (6) For example.element of Cir for all reachable sub-criteria. the levels and relationships among sub-criteria can be determined.. and Ci4. according to Eqs. Ci2. whereas the latter includes all sub-criteria of the reachable elements of Cir . r Then. 1+1=1. and 1¹ T ! A  I ! Ci 2 ¹ 0 Ci 3 ¹ Ci 4 0º Note that the reachability matrix is calculated under the operators of the Boolean multiplication and addition law (i. Ci1 Ci 2 Ci1 T ! . and 0+0=0). In addition. 0 v 0 ! 0 . 1v 1 ! 1 . Ci3. 1v 0 ! 0 v 1 ! 0 . let criterion i consist of four sub-criteria Ci1.

Then. the row and column 13 3 3 (4) (5) Ci1 ¨1 © ©1 ©1 © ª0 Ci 2 1 1 1 0 Ci 3 0 0 1 0 Ci 4 0¸ ¹ 1¹. According to Eq. (6).e. 0¹ ¹ 1º . A  I ! Ci 2 Ci 3 Ci 4 2 Ci 3 Ci 4 0 0 1 0 1 ¸ ¹ 1¹. we can determine that the top-level sub-criterion is sub-criterion Ci4 in Table 3. 1 ¹ ¹ 1¹ º ¨1 © ©1 ©1 © ©0 ª 1 1 1 0 where the asterisk * indicates the derivative relation which does not emerge in the original relation matrix (i.. A+I).

3. 3. repeating the above steps. Ci 4 a) C that we divided into the final set (i.2.2 1. 3 The directed digraph of multi-level structures 3. The fuzzy comparison matrix is p % % A ! [ arr ' ].4 4 1.corresponding to sub-criterion Ci4 can be deleted from matrix T.e. we first define the fuzzy % comparison matrix A . Table 3 The reachability of criteria Cir 1 2 3 4 Rr r Rr ‰ 1.3 3 1.2. the second level can be determined.2..2 Fuzzy weights determination To calculate the fuzzy weights of each criterion.2. Ci 2 . Ci 4 a) can be plotted as Fig.4 1.. Ci 2 .3. the order result of the original set (i. Based on the reachability matrix and the multi-level diagraph. Then. Ci 3 .4 1. Ci 3 .e. S ! _Ci1 .2.3 1.2 3 4 Level 1 C 14 Level 2 C11 C12 Level 3 C 13 Fig. S ' ! _ i1 .4 1.2.

nvn p . % % where A is an n v n matrix of pairwise comparisons in which aiip is given the 14 5 4 4 r (7) .

. a set of n criteria compared pairwise for their relative importance (weights) can be denoted k k % % % % w k ! ( w1k .4. If this equation has a non-zero solution for w . Thus. When the criteria are independent of each other. According to Eq. I denotes the identity matrix.. wn ) .20.9. (7). (8) % % where w is an n v 1 vector. and w is an eigenvector % corresponding to Pmax . we use the geometric mean method to determine the tentative fuzzy weights of each criterion as follows [3. w2 ...10]: p % % % % ri p ! . the relative % weights of vector w can be determined by the fuzzy AHP method as % % % ( A  Pmax I ) w ! 0 . then % % % Pmax (which is a scalar) is an eigenvalue of A . which is a diagonal matrix with the main diagonal terms equal to 1 and zero elsewhere.value of the pairwise comparison between the criteria in the ith row and the i ' th column by evaluator Mp (obtained from questionnaires).

. The BNP value of the triangular fuzzy number ... 1 2 (9) (10) (11) % % % % % wip ! (li p .e. 2. i ! 1. n .. we employ the COA defuzzification method [28] to compute the BNP weights of the criteria (i. † wiP ]1/ P i i p % where ain is the fuzzy comparison value of criterion i1 to criterion in given by % evaluator Mp . Furthermore... mip . wi )... ‡ wiP ] / P or [ w1 † wi2 † . ui ) ! [ w1 ‡ wi2 ‡ . and ri p is the geometric mean of the fuzzy comparison values of criterion i to criterion n by evaluator Mp.. uip ) ! ri p † [ r1 p ‡ r2p ‡ . ‡ rnp ]1 % % % % % % % wi ! (li . mi .aip † aip † L † ain ...

li . ui can be found by (12) wi ! «. mi .

mi  li  .

­ ½ 15 .ui  li » 3  li .

For example. P -fuzzy measure) to find the grade of importance among sub-criteria can be employed. in this paper we use a fuzzy integral method to obtain the synthetic utilities with interactive effects between sub-criteria. Traditional synthetic utility Use uzzy measure and uzzy integral method Fuzzy integral synthetic utility A B C Fig. In the proposed model. 4. the uzzy measure assumes only monotonicity and boundaries are more general than the conventional Lebesgue measures (which assume additivity). It would be more appropriate to apply a uzzy measure to determine the grade o importance o multiple criteria in a human subjective evaluation process. are used to evaluate the appropriate vendor. their synthetic utilities could be expressed as Fig. several properties o the uzzy measure and the uzzy integral are presented as ollows: 6 6 A B C 16 . Here.3. non-additive methods. Unlike the traditional definition of a measure based on additive properties..3 The synthetic utility with interactive criteria calculation To conform to human behavior in a real world situation. given three sub-criteria A. 4 Comparison between the traditional additive method and the uzzy integral method or calculating synthetic utility In addition. the non-additive fuzzy measure and fuzzy integral are applied to evaluate problems with dependent multi-criteria. where the sum between the measure of a set and the measure of its complement is not equal to the measure of space.g. B and C of a criterion. in which the fuzzy measure (e.

x2 .«. is a finite set and let P(X) denote the power set of X or the set of all subsets of X. xt a.Let A and B be sub-criteria of criterion i. assume that X ! _x1 .1] such that: (1) g ir . A fuzzy measure g over a set X is a function g: P(X) [0. for r=1.L .t.

ˆ ! 0. g ir .

B Ž P . (2) If A.

X and A Ž B . then g ir .

A e g ir .

The P -fuzzy measure [31] is most widely used fuzzy measure. B Ž X with A ‰ B ! ˆ .B . we adopt a P -fuzzy measure to gauge the relationship of each sub-criterion. let A. which describes the degree of addition among sub-criteria. In the fuzzy measure. Thus. the P -fuzzy measure g P satisfying the following additional property: P P P P P g ir . and it is constrained by a parameter P .

A Š B ! g ir .

A  g ir .

B  P g ir .

A g ir .

the value of P can be found from g ir .B . (14). for 1 P P Based on Eq.

X ! 1 . and P generally the fuzzy density denoted as g rP ! g ir .

_xr a for sub-criteria can also be obtained by 1 ! P — .

1  P g  1 . based on the above properties one of the following three cases will be discussed.e.. g rP . Case 1: if P >0. P r r !1 t Therefore. i.

A Š B " g rP .

A  g rP .

then this implies A and B have a multiplicative effect. g rP . Case 2: if P =0.B .. i.e.

A Š B ! g rP .

A  g rP .

B . 17 8 r !1 r !1 r ! r 1 8 g rP . then this implies A and B have an additive effect.

_x1 . xt a ! § g rP P § 7 t ! 1.L . x 2 . (13) (14) g. (15) t 1 t § P g rP g rP  L  P t 1 g1P g 2 L g tP (16) .

g rP . i.Case 3: if P <0..e.

A Š B g rP .

A  g rP .

let h be a measurable set function defined on the fuzzy measurable space . Next.B . then this implies A and B have a substitutive effect.

‚ and suppose that hij . X .

x1 u hij .

x2 u L u hij .

then the fuzzy integral of fuzzy measure g rP .«.«. for i=1.xt . j=1.m.n.

™ with respect to hij .

™ can be defined by uij ! ´ hij ™ dgrP ! hij .

xt ™ grP .

Ht  « hij .

xt 1  hij .

xt » ™ grP .

Ht 1 L  « hij .

x1  h .

x2 » ™ grP .

H1 ij ­ ½ ­ ½ (17) ! hij .

xt ™ « grP .

Ht  grP .

Ht 1 »  hij .

xt 1 ™ « grP .

Ht 1  grP .

Ht  2 »  L  hij .

x1 ™ grP .

H1 ­ ½ ­ ½ where P g1P ! g 2 ! L ! g tP then hij .

x1 u hij .

x2 u L u hij .

In addition. The basic concept of Eq. can be determined (see Appendix 3). which in this study we call the unweighted score of the criterion for each vendor.xt is not necessary. Based on Eq. the synthetic utility between sub-criteria of each vendor is obtained. 5 The basic concept for calculating the synthetic utility using the fuzzy integral 18 9 1 ! _x1 a. H t ! _x1 . Fig. x2 a. xt a! X .L . if P =0 and h . (17) can be expressed as shown in Fig. H 2 ! _x1 . as discussed above. (17). A simple example will be used to show how the value of P can be obtained as well as how the relationships among criteria. x2 . 5.L .

x1 g .

H 1 h .

x2 g .

H 2 h .

x2  h .

x3 h .

x1  h .

x 2 h .

x3 h .

x4 g .

H n  1 h .

xn 1  h .

xn h .

xn g .

H n x1 x2 xn  1 xn h .

xn .

an empirical example of a vendor selection decision demonstrates that the integrated fuzzy MCDM technique is more appropriate than the traditional method.g. Motorola. IBM. 10. To achieve the strategic initiatives and strengthen the core competencies of a company. communications equipment. In line with the purchasing request (e. especially when sub-criteria interrelate. the consolidated revenues were US$13 billion dollars. a practical application will be used to demonstrate the real-world usefulness of the proposed fuzzy multi-criteria group decision-making technique. it is crucial to select appropriate vendors. 4. In 2004. and Sony. high technology industries in Taiwan have grown rapidly. An empirical study of vendor selection in Taiwan In this section. Its main customer groups include companies that are famous worldwide. In this section. and the company has over 100. In order to cope with business growth and sustain the company¶s competitive advantage. This section is divided into four subsections: (1) problem descriptions. which was based on scenario writing and brain-storming. such as Intel. (2) data collection via questionnaires. especially the electronic and information-technology (IT) industries. and (4) discussions.1 Problem descriptions In recent years. for heat sinks for notebook personal computers 19 .method 4. an effective vendor selection model for the company in a new purchasing project to select an appropriate vendor is needed. The case company is a well-known 3C component manufacturer. Dell.. Its products mainly include PC enclosures. HP. (3) results and analyses.000 pieces/week).000 employees around the world. and consumer electronic products.

E2. quality manager. Moreover. V4. ten experts gave their scores of their satisfaction levels for each candidate vendor in terms of each criterion in the hierarchical evaluation structure. M2. knowledge. Those vendors who successfully passed the screening processes were eligible for procurement. The major influencing criteria and sub-criteria involved in vendor selection are given in Table 4. M7) and ten experts (E1. Based on the information in Table 4. product manager and production manager). Table 4 Criteria and sub-criteria for the vendor selection Criterion Quality C1 Sub-criterion Quality performance (C11) Quality containment & VDCS feedback (C12 Price & Terms C2 Price C21 Terms C22 Responsiveness Lead time C24 VMI/VOI hub Set Up cost Supply chain support C3 Purchase order reactiveness Capacity support & flexibility C25 C31 C32 C23 20 . experience of managers. A purchasing committee with seventeen evaluators containing seven decision-makers (M1.«. V2. The criteria and sub-criteria involved in the vendor selection have been chosen in terms of the professional.e. purchasing manager. and V5 ).. choosing the possible evaluation criteria for the vendor selection involves a decision-making team which includes managers from different functional divisions of the case company (i.«. E10) was then constituted. V3. we will evaluate five candidate vendors (V1.in a consumer electronics business division. purchasing director.

see Appendix 1). the participants had to determine the level of the relative importance between the two elements. This questionnaire sought the satisfaction level of the experts concerning the candidate vendors using a Likert-type five-point scale. Furthermore. In this questionnaire. Through the expertise and knowledge of the decision-makers. the participants were also asked to respond to a questionnaire containing a series of pairwise comparisons using Saaty¶s nine-point scale (AHP Questionnaire. Meanwhile. see Appendix 1). a questionnaire (Appendix 2) for the evaluation by experts of the synthetic utilities of interactive sub-criteria was also administered. 4. For each pairwise comparison.2 Data collection via questionnaires In order to clarify the interrelations among the sub-criteria of a criterion in a hierarchical system and determine the weights of the criteria.3 Results and analyses Based on the fundamentals stated in Section 3. each question consisted of a pairwise comparison of two elements. the relationships among the sub-criteria of a criterion can be determined. Experts were asked to evaluate different cognitive levels given each grade of candidate vendor according to subjective perceptions.Delivery/VMI operation Technology C4 Technical support C41 Design involvement C42 C33 ECN/PCN process C43 4. the integrated techniques to solve 21 . the decision -makers were asked to complete a questionnaire (ISM Questionnaire.

1 (relation)) for the direct relations with each of the other sub-criteria using the questionnaire to form an adjacency matrix A..the vendor selection problem were implemented according to the three following stages: Stage 1: Finding the interrelation among sub-criteria of a criterion To clarify the interrelationships among the sub-criteria of a criterion. which was calculated using the mode method. the seven evaluators were invited to determine the relationships among the sub-criteria based on the ISM method using the questionnaire. for each of the sub-criteria. 6. Subsequently. (4) and (5). the adjacency matrix A can be determined. the seven evaluators assigned values (i. In addition. Based on Eqs. the stable reachability T dcould be derived. In this method.e. (2) and (3). The levels among the sub-criteria for each criterion were plotted as Fig. Criterion 1 Quality Criterion 2 Price & Terms 22 . 0 (no relation). a reachability matrix T was also calculated using Eqs.

the interrelations among the sub-criteria of a criterion were found.2579) 0. Based on the fuzzy five-level scale (see Fig.3954. the four evaluation criteria in the vendor selection problem are quality (C1).0.0.5202.0.0. (10).5284.0.1269.0543.0210.Lead time .5076.0365) 0.0.0696.0.9010) (0.0.0516.0.5206 0.0057.0.0. price and terms (C2 ).0.0530) (0. The AHP Questionnaire was delivered to the same seven evaluators.0584 0.5441) (0.0. Their subjective judgments were integrated sequentially to obtain the fuzzy weights of criteria by Eq.0393 0.Responsiveness .0.5095 0.Quality containment (0.1349) (0. supply chain support (C3).4716.0298 23 .2997) (0.Terms .0838.0. The final fuzzy weights and nonfuzzy BNP values for each criterion were also determined using Eq.0.0440 0.3034 0.1179.5680 (0.0138.0187 0.0.Criterion 3 Supply Chain Support Criterion 4 Technology Fig.1420 0.0.1211) (0.0.1434 0. Table 5 Weights of criteria for evaluating the appropriate vendor by fuzzy AHP Normalized Normalized Overall weights (global) BNP values Criteria and sub-criteria Local weights criteria weights (local) sub-criteria weights (global) Quality .0.1802.4992 0.1454 0.0478.0930) (0.0476) (0.4794 0.3451.6940) (0.1594 0. as shown in Table 5.Quality performance .0.0.0.3001.0864 0.2828.0.2783.0563.4306) (0.0771 0.VMI/VOI hub set-up (0.2543 0.0.0089.0.0.0.1706.0188.0.1557 0.0276 0.1692) (0.Price .0891 0.0075.0360.0621.0751. A further goal for the ISM Questionnaire was to find the fuzzy weights of criteria using the fuzzy AHP method.0246 0.0.0.0284. 2).7627) (0.2788 0. and technology (C4). 6 The multi-level diagraph of four criteria derived by the ISM method Stage 2: Determining the evaluated criteria¶ weights According to the results of the ISM method.0.3382 0.1580 (0.1321.2727) 0. (12).0551.6564) & VDCS feedback Price & Terms .3281.

Let hij represent the fuzzy performance score of the jth candidate vendor under the ith criterion by the kth evaluator.0393.0.0. as shown in Table 6.Purchase order (0.0956.0413.2589) (0.6347) (0.4478.1239 0.2052) 0.0296 (0.4412 0.Design involvement .0.4421) flexibility .4447) (0.0.1498 0.0.0.0713 0.0868 0.0.0.2174.0659. That is 1 % % % % hij ! hij † hij2 † L † hijK .1120.0.Technical support .0511 (0.2497.2571 0.0.0.0.0240.0669. Then.5222) operation Technology .7585) reactiveness .0235.6588) (0.1128.1296 0.0272 0.0351 (0.0.0. based on their subjective judgments the evaluators defined their own individual range intervals using linguistic variables within a fuzzy k % scale to determine the performance scores of each candidate vendor.2576.0.1304) (0.0.3632 0.0. we select the fuzzy geometric mean method to aggregate the fuzzy performance score from K evaluators.1813.0.Capacity support & (0.2429) (0.0795 0.ECN/PCN process (0.0.1291 0.1495.0.1738) 0.2086) 0.1747) 0.3797 0.cost Supply chain support .3601.0183 (0.0750.2053.0259 0.0.0.0.0630.3032 0.0.2981) 0.0323.1158 Stage 3: Finding the performance score by linguistic variables Using the Questionnaire 2.1527.Delivery/VMI (0.0.0342.1029 0.3823.0.0900.2556 0.3001.0285.0.0.2520.0877 0.0644.

(12) is then used to compute the BNP values of the fuzzy % performance score hij . (18) Furthermore. 1K . Table 6 Fuzzy performance score of candidate vendors for sub-criteria 24 . using Eq. as shown in Table 7.

24.4.40.7.7.3.8.02) (5.79) (2.7.09.22) (2.83) (2.7.4.81) (2.98) C33 (3.69.63) C23 (3.71) (2.3.7.18.5.7.4.45.52.5.56) C22 (3.07.7.86.72.66) (3.89.23.04.81.5.7.5.19.91.7.61.33.5.99.4.88) (3.81.78.44) (6.93.767 4.6.5.6.74.57.5.94.6.4.16.24.46) (2.295 6.09.91.6.34) (2.19.49) (2.5.91) (4.6.85) (4.6.4.84.768 5.90.31.20.5.5.95.5.30.14) (2.7.06) (1.55) (5.14) (2.28.87.8.6.66) (4.18.71.25.67.46.7.3.25.36.07.39.36.71) (4.37.33.7.3.53) (5.5.5.43.6.065 5.42.85) (4.34.98) (5.2.93.4.376 7.6.6.6.3.33.3.93.43) (2.88.4.051 6.58) (2.8.02.089 5.71.36) Table 7 Defuzzied performance score of candidate vendors for sub-criteria Candidate vendor (V j) V1 C11 C12 C21 C22 C23 C24 C25 C31 C32 C33 C41 C42 C43 BNP values of sub-criteria 4.49.7.44.14.66.7.23) (2.54.5.5.5.94.24.64.12.88.5.55) (1.24) (4.5.4.50) (2.73.12.62.68) (2.8.5.5.5.43.42.289 6.4.08.02) (3.5.6.3.6.3.86.7.5.49) C24 V1 V2 V3 V4 V5 C25 C31 (2.4.52.69.81) (4.30) (3.58) (2.5.21.78) C32 (5.5.5.84.6.91) (3.69) (3.83) (4.68.18.51.97.4.11.5.09) (2.59.29.5.00) (1.2.5.18) (3.29) (5.02.84) (1.8.3.17) (2.22.6.42.94) (2.7.8.90.4.378 5.14.6.29.46.8.4.39.083 25 .66) (2.4.3.3.88.77.5.355 3.5.14.6.367 5.2.70.83.5.82) (1.25.18.17) (3.11.14.6.96.22.6.07.4.53) (2.81.86) (4.6.3.38.7.4.6.88) (1.4.72.50) (2.76.4.2.47) C41 V1 V2 V3 V4 V5 C42 C43 (3.5.5.43.33) (6.37.6.5.04.76.221 6.5.3.56.3.6.5.4.4.53.35.66.86.5.92) (4.25) (2.02.47.84.52.85.5.7.5.08) (4.71) (3.98.3.75.05) (4.77.82.7.6.75.Candidate vendor (Vj) V1 V2 V3 V4 V5 C11 C12 Sub-criteria C21 (4.88) (1.

006 3.819 5.138 4.326 4.499 4.692 3. 4.568 6.293 5.541 3.353 5.315 4.1376) and also among three sub-criteria of the technology criterion ( P = -0.167 5. the P values were still quite low.730 3. For the quality criterion.1170).674.377 4.419.196 4. the satisfaction values concerning the quality.853 4.609 7. whereas there were substitutive effects among three sub-criteria of the supply chain support criterion ( P = -0.272 4. The satisfaction values for the five candidate vendors could be gained to calculate the fuzzy integral values of each criterion (see Table 8 below).880 3.633 3.253 3. 7.635 4.917 respectively. A multiplicative effect was found among five sub-criteria of the criterion price and terms ( P = 0. and technology criteria for V1 were 5. the satisfaction 26 .459 4. by using the method proposed in this study. Consequently.534 5.215 5. non-additive multi-criteria evaluation techniques are employed to deal with situations involving interrelations among the sub-criteria of a criterion.540. Even though there were synthetic interactive effects.491 6. the evaluation decision process would be more practical and flexible using the different P values.897 7.969 3. Using Eq.863. 8.140. and 4. which could imply there is still a great deal of room for these candidate vendors to improve their performance.786 respectively.721 6. price and terms.863 2. Similarly.136 4. where the values of P ranged from ±1 to positive infinite value ( g ) (see Section 3. (16).1467).812 3.4 Results and discussions In the proposed method. and 2.224 4.692 6.705 6. the P value of each criterion can be obtained. the satisfaction values for the same criteria for V3 were 4.150 6.460 4.131 4.782 6. 1.016 2.854 3. We introduced the P value to represent the properties of substitutive or multiplicative effects between two criteria.026 5.V2 V3 V4 V5 5.868 5. Take V1 and V3.060 4.287 3. supply chain support.007.718 6.693 4.494 5.254 5. for example.022 5.288 3.3).395 2.

It can be seen that if the traditional simple additive weight model is utilized to aggregate the final overall scores. 7. In Fig. The overall scores can be obtained and the ranking is V2 f V1 f V3 f V5 f V4.e. These results are consistent for the same criterion among different candidate vendors. or overestimate when the criteria have substitutive effects (i. for 27 . a higher score was obtained by using our proposed method. that is to say. In contrast. where A f B represents that A is preferred to B. 7.values of V1 and V3 were the same no matter whether the simple additive weight method or the method proposed in this study were used. regardless of the substitutive or multiplicative effects among different sub-criteria. price and terms). the results of the ranking for the candidate vendors will be different. it can be seen that V2 has the highest score in most sub-criteria. thus emphasizing the gaps between the appropriate vendor (V2) and the ideal points. These aspired/desired/ideal points (10 scores) represent points at which all criteria of each vendor would be optimized. price (C21).e.. the scores were lower using our proposed method.. since the quality criterion is based on the assumption of independence. from Table 9. the ranking orders are different. This observation implies that criteria with synthetic interactive effects are more reasonable than those obtained by the traditional additive evaluation process. supply chain support and technology). the ranks of the overall scores of the five candidate vendors were found to be V2 f V3 f V1 f V5 f V4. for the other two criteria. As for the price and terms criterion. Furthermore. The results can also be compared with the traditional AHP method. we employed the concept of the ideal point to represent the results of the analysis in Fig. Moreover. supply chain support and technology. the appropriate vendor is V2. For instance. and capacity support & flexibility (C32). it will underestimate when the criteria have multiplicative effects (i. Although the same appropriate vendor is found using different methods. VMI/VOI hub set up (C25). Obviously. except in the sub-criteria quality performance (C11).

g.674 V4 3. there are gaps to the aspired points of 4. and so on) to achieve the aspired/desired values. price negotiation tactics.1198 respectively (see Table 5 below). Based on above results. it can also be seen from Table 8 that the satisfaction values of V2 in the quality and price and terms criteria are 5.3034 P value Candidate vendor (Vj) V1 5. Similarly. This information is useful for new vendors in this purchase project. the best vendor will also be obtained through R&D and innovation/creativity. the global weight values of price and quality performance are 0. It can be seen that the degree of importance of the two sub-criteria is higher than those of the other sub-criteria. the gaps between individual scores and the ideal point of the price sub-criterion are 4. continuous improvement. These results show that the evaluators are most concerned about the price and quality performance when selecting the appropriate vendor. Furthermore.733 respectively..140 28 V2 5.V2 and V3.737 - .267 respectively. the satisfaction values for the supply chain support and technology criteria are 3.466 respectively. Although the values are larger than one-half of the aspired point.815 and 8.279 and 3.815 V3 4. some suggestions can be made to stimulate the creativity and improve the performance of the appropriate vendor (V2 ) for some poor sub-criteria via different strategies (e.619 and 4. with respect to the ranking orders of the weights of sub-criteria for evaluators in the multi-criteria evaluation hierarchy. supply chain integration planning.553 V5 4.1227 and 0. Table 8 Fuzzy integral results of each dimension for candidate vendors Criteria (Ci) Quality AHP weights 0.185 and 1. which is consistent with the results found in a real purchase project. These values are less than the aspired point and the gaps to the aspired points for the two criteria are more than the gaps to the aspired points for the quality and price and terms criteria.190 respectively. Similarly.

it will be not necessary to use the fuzzy integral method.144) 5.925) 6.553) 5.842 (6.117 5.491) 2.152) (3.621 (3.1467 7. The 2nd~ 4th criteria represent criteria with synthetic effect.674) 8.1158 0.414) 2.533) (4. so the fuzzy integral method is needed.627) 7.437 (4.685) 7.366) -0.945) 0.946) 2.486) 5.917 (4.062 5.945 Ranking 2 1 3 5 4 Proposed method BNP 6.160 * 5.1170 4.283 (3.1376 4.140) Price & Terms Supply chain support Technology Overall scores 0.117 (5.766 (3.619 (6.692) 1.847 (3.815) 8.160) (4.342) 1.842* 7.733 (6.521 (4.863 (5.540 (4.419 (5.988 (5. ³( )´ represents the weighted synthetic scores of candidate vendors calculated using the traditional AHP method Table 9 Results obtained using the traditional AHP method and the proposed method Candidate vendor (Vj) V1 V2 V3 V4 V5 Traditional AHP method BNP 5.961 (3.961 Ranking 3 1 2 5 4 ³*´ represents the appropriate vendor 29 .502) 4.988 7.786 (4.190 (6. therefore.432 6.356) 2.062 (3.533 3.(5.152 3.007 (5.0713 - Note that the 1st criterion represents the criterion without synthetic effect.5095 0.214) 3.691) -0.737) 6.432) (5.

In addition.12. if a criterion were to contain additional sub-criteria.000 10.000 2. In this techni ue.466 6.000 4. However. vague. we employ the ISM method to clarify the interrelationships of intertwined sub-criteria in the complex structural hierarchy of vendor selection problems. Concl ions Vendor selection is a very complicated multiple criteria problem. y: ideal point Fi 7 The performance scores of the appropriate vendor and the ideal point in each sub-criterion 5. The final fuzzy weights of each criterion can also be obtained by applying 30 .000 4. j : V 3.000 8. the traditional MC M methods are based on the assumption of independenceamong sub-criteria. :V 5. which is more appropriate for evaluating the best vendor. and the information available for use in a multiple criteria decision process is usually uncertain. +: V4.000 0. and the analytical framework for vendor selection processes fails to consider the interdependence of the sub-criteria.279 3.000 1 2 3 Note: 4 : V1. In this paper. this would result in a stronger possibility of conflict and more intertwined complexity. 5 6 7 8 9 10 11 12 13 : V2. we demonstrate an integrated fuzzy MC M techni ue. In addition. we introduce fuzzy numbers to express linguistic variables that express the subjective judgment of evaluators. or imprecise and the criteria are not independent in real world situations.

Therefore. We further find that using this method. the difference of preference order will depend on the different P values with respect to the effect of criteria. The proposed method may well avoid overestimation when the criteria have substitutive effects or underestimation when the criteria have multiplicative effects. 31 . we effectively validate the non-additive fuzzy integral method to conduct the decision-making problem while the interactions of criteria are considered. we can provide the available information and strategies to stimulate the creativity of and improvement the appropriate vendor that they might achieve the aspired/desired values. The results show that the fuzzy integral method is better and more reasonable than the traditional method. This will provide useful information regarding substitutive or multiplicative effects among the criteria under consideration. we demonstrate that the non-additive multiple criteria evaluation techniques are more appropriate than the traditional method and provide practitioners with a valuable tool for use in a fuzzy MCDM environment to solve vendor selection problems.the fuzzy geometric mean method. By the concept of ideal point. Furthermore.

C. European Journal of Operational Research 130(3) (2001) 653-664. [3] J. P. European Journal of Operational Research 173(3) (2006) 1026 -1046. S. Choquet fuzzy integral-based hierarchical networks for decision analysis. [9] Y.H.P.H. Shyu.J. M. [4] M. Tzeng.Y. Tzeng. International Journal of Information Technology and Decision Making 4(2) (2005) 1-20. Chen. Tzeng. Fuzzy multiple-criteria decision-making approach for industrial green engineering. Chiu. [5] Y. Fuzzy Sets System 17(1) (1985) 233-247. Bellman. Zadeh. Cheng. IEEE Transactions on Fuzzy Systems 7(1) (1999) 63-71.K. Chiang. Decision making in a fuzzy environment. [10] T. Fuzzy MCDM for evaluating the e-commerce strategy. Tzeng. [2] J.A. [7] H.F. Fuzzy hierarchical analysis. Fuzzy MCDM approach for evaluation of expatriate assignments. G. Evaluating sustainable fishing development strategies using fuzzy MCDM approach. Chou. G. L.H.H. J. Management Science 31(6) (1985) 647-656. G. D. [6] J. Vincke.C. Tang. International Journal of Computer Applications in Technology 19(1) (2004) 12-22.E. Brans. Tzeng. 32 . Chen. G. Evaluating IT/IS investments: a fuzzy multi-criteria decision model approach.K. Tzeng. Management Science 17(4) (1970) 141-164. A preference ranking organization method.C. Omega 33(2) (2005) 223-234.H. Buckley. Chiu. Using fuzzy integral for evaluating subjectively perceived travel costs in a traffic assignment model. Environmental Management 30(6) (2002) 816-830.References [1] R. G.H. [8] H.H. G.Z. Chiu.W. Tzeng.

P. Ghoudsypour. Product involvement and industrial buying. Hsieh. [16] R. An analysis of vendor selection systems and decisions. O¶Brien. [14] J. A review of methods supporting supplier selection. Roodhooft.Y. European Journal of Purchasing & Supply Management 7(2) (2001) 75-89. Journal of Purchasing and Materials Management 17(2) (1981) 23-28. S. Evans. Dickson. International Journal of Project Management 22(7) (2004) 573-584. 33 . A model of human evaluation process using fuzzy measure.[11] L. [15] R.O.H. An evaluation of supplier selection methods from a total cost ownership perspective.H. G. M. Morlacchi. G. Florez-Lopez. Which contractor selection methodology? International Journal of Project Management 16 (3) (1998) 153-164. Information Sciences 173(1-3) (2005) 197±225. Lu. E.F. Information Sciences 177 (5) (2007) 1169-1179. Sugeno. Degraeve. Using fuzzy MCDM to select partners of strategic alliances for liner shipping. A decision support system for supplier selection using an integrated analytic hierarchy process and linear programming. Fuzzy integral in multicriteria decision-making.D. Tzeng. International Journal of Man-Machine Studies 22(1) (1985) 19-38. Journal of Purchasing 2(1) (1966) 5-17. [13] G. [17] S. E. Ishii. C. F.S. Ding. Strategic supplier selection in the added-value perspective: A CI approach. Labro. [18] M. Fuzzy MCDM approach for planning and design tenders selection in public office buildings. Grabisch. [19] G. W. [12] Z. International Journal of Production Economics 56-57(1-3) (1998) 199-212.H. De Boer. Fuzzy Sets and Systems 69(3) (1995) 279-298. Labro. [20] T. Holt.T. Liang. [21] K. European Journal of Operational Research 125(1) (2000) 34-59.

1974. [31] M. [24] L. [23] A. Morlacchi. Chiu. Supplier evaluation and rationalization via data envelopment analysis: An empirical example. Journal Supply Chain Management 37(3) (2001) 28-37. International Journal of Purchasing and Materials Management 28 (2) (1992) 31-36. Sugeno. Shankar. The Analytical Hierarchy Process. International Journal of Operations & Production Management 14(6) (1994) 52-59. [25] P. Mathematical and Computer Modelling 44(8) (2006) 749 -761. Mendez.J. Hill. S. Tzeng. Vrat. Y. A fuzzy goal programming approach for vendor selection problem in a supply chain. [30] H. Talluri. Dublin 1999. Mikhailov.G. Proceedings of 8th IPSERA Conference.[22] M. Hierarchical fuzzy integral stated preference method for Taiwan¶s broadband service market. [32] F. H. Vendor selection using interpretive structural modeling. Fuzziness and Knowledge-based Systems 11(5) (2003) 635-652. 34 . Tokyo.P. [26] R. Shyur. R. Vendor evaluation and selection: the design process and a fuzzy ±hierarchical model. [28] S. Tokyo Institute of Technology. Using the analytic hierarchy process to the supplier selection problem. Saaty. [27] R. Computers & Industrial Engineering 46 (1) (2004) 69-85. International Journal of Uncertainty. Theory of Fuzzy Integrals and its Applications. New York. McGraw-Hill. Opricovic. Defuzzification within a fuzzy multicriteria decision model. S. Shih. Fuzzy analytical approach of partnership selection in formation of virtual enterprises. [29] T. Tseng. Nydick. Omega 30(2) (2002) 393-401.S. Deshmukh. 1980.H. Omega 33(1) (2005) 55-64.L.L. Kumar.J. Narasimhan. R. G. P. Mandal. D.M. A hybrid MCDM model for strategic vendor selection.

Vendor selection criteria and methods. [40] C. A. Y.B. 1994. [39] C. Wiley Interscience. New York. Fuzzy sets. An optimization approach to determining the number of vendors to employ. Chen. Current. Yager. Warfield. Toward interpretation of complex structural modeling. IEEE Transactions Systems Man Cybernet 4(1) (1974b) 81-87. W. Verma.[33] E. European Journal of Operational Research 50(1) (1991) 2-18.P. Omega 26 (6) (1998) 739-750. An analysis of the supplier selection process. [34] G. Boeva. Essentials of Fuzzy Modeling and Control. 35 . V. J. J. Warfield. D. M.R. Societal Systems: Planning. Wiley. R. [36] J. Current. and Complexity. Policy. Filev. Ou Yang.R. Information Science 54(1) (1991) 103-130. Zhao.N.A. N. Weber. Multi-step ranking of alternatives in a multi-criteria and multi-expert decision making environment.C. Supply Chain Management: An International Journal 5(2) (2000) 90-98. Warfield.E. Benton. [35] R. Tzeng. [38] J. R. Lin. Govind. Desai.H. Information Sciences 176(18) (2006) 2673±2697. [41] R. Weber. Information and control 8(3) (1965) 338-353.A. Algebraic characteristics of extended fuzzy numbers. IEEE Transactions Systems Man Cybernet 4(5) (1974a) 405-417. C. [42] L. 1976. Information Sciences 169(3-4) (2005) 409-426. [37] J. Developing interconnection matrices in structural modeling. [43] R. C. Zadeh. Pullman. N. Hierarchical MADM with fuzzy integral for evaluating enterprise intranet web sites.P.T. A. Tsiporkova. New York.

if purchase order reactiveness does not exist. Seven decision makers were invited to evaluate the relations of criteria. the values given for the relationships between criteria to form the R matrix were as follows: Q P S T Q «0 1 0 1 » R ! P ¬ 0 0 1¼ . and there is no influence of the sub-criterion delivery/VMI operation. Take the four criteria evaluated by the first decision maker.1 According to defined evaluation criteria. price & terms (P). ³Supply Chain Support´ criteria Purchase reactiveness Purchase order reactiveness Capacity support & flexibility Delivery/VMI operation 0 order Capacity flexibility support 1 0 0 & Delivery/VMI operation 0 Appendix 1. the mean relation matrix ( R ) could be obtained. In addition. we utilize a simple relation matrix to verify the independence between criteria.5 ) was given by decision makers through discussions to decide whether the criteria were independent. for example. Thus. I can then establish an analysis model of vendor selection. capacity support & flexibility. we used an arithmetic mean method to calculate the mean values of the opinions of the seven decision makers concerning the relationships between criteria in the relation matrix (R). Row 1 means that if you think that purchase order reactiveness does not exist. Under this criterion. there are three sub-criteria ± purchase order reactiveness. support chain support (S). and delivery/VMI operation ± which are shown below. a threshold value (let L ! 0. then you should give it a score of ³0´. Similarly. your opinions will set up the relationship among criteria and sub-criteria. including quality (Q).Appendix 1 ISM Questionnaire This questionnaire is about vendor selection and the evaluation of relative sub-criteria and criteria. Furthermore. Instructions for this questionnaire: Example criterion 3: Supply Chain Support. and technology (T). Because of your expertise and knowledge. ¬ ¼ 0 0¼ S ¬ ¬ ¼ 0½ T ­ 36 . and the sub-criterion capacity support & flexibility does have an influence then you should give it a score of ³1´.

Appendix 1. the evaluations of the relations between the four criteria made by the seven decision-makers were summarized as the values of the relation matrix ( Rd Finally. l>1 stop. Based on the concept of the ISM method. and cat (C) and the relationships can be represented as the relation matrix in Fig. the four criteria should be regarded as independent.2 We also use a simple example to show the steps of the ISM. ¼ S ¬ 0 0.14¼ ¬ ¼ 0 ½ ­ and R ! P ¬ ¬ If the value of the mean relation matrix R is more than the threshold value ( rij u L ). mother (M). Last. son (S). whereas if not. the pairwise comparison criteria can be considered independent. the limiting reachability matrix obtained is as follows: 37 C 1 0 0 0» 0 1 1 0¼ ¼ 0 0 0 0¼ ¼ 0 0 0 1¼ 0 0 0 0¼ ½ . A1.Next. daughter (D).5. (3). by calculating the limiting power of the matrix T. the mean relation matrix R = [rij ]nvn could also be ).14 0. the relation matrix is added to the identity matrix to form the T matrix as follows: F F F T! M S «0 ¬0 ¬ ¬0 ¬ ¬0 ¬0 ­ M S C F M S F «1 ¬0 ¬ ¬0 ¬ ¬0 ¬0 ­ M 1 1 0 0 0 S M D S C C Next.29» 0 0. when T l ! T l 1 . Thus. using an arithmetic mean method. all values of the mean relation matrix R are smaller than 0. Clearly. obtained as follows: Q «0 3 1 2 » R' ! P ¬ 0 2 2¼ .29¼ .29 0.43 0. the tentative reachability matrix is obtained by powering the matrix T to satisfy Eq. and T ! A  I ! C 0 0 0» 1 1 0¼ ¼ 1 0 0¼ ¼ 0 1 1¼ 0 0 1¼ ½ C A B A B @ Q P S Q P S @ C . Assume the family members consist of father (F). they are dependent criteria. ¬ ¼ S ¬ 0 1¼ ¬ ¼ 0½ ­ Q «0 0.

2.4 1. ¼ 1¼ 1¼ ½ F M F M  S C Because matrix T 3 ! T 4 . Similarly. when l=3.3. the intersection of row 1 and column 3 represents the effects that father (F) has on mother (M) and has on son (S).4. (4) and (5). The first level can be derived according to Eq. the asterisk ** means the effects that father (F) has on mother (M) and has on daughter (D) and has on cat (C).5 3 4.5 E «1 1 1 1 ¬ ¬0 1 1 1 ¬0 0 1 0 ¬ ¬0 0 0 1 ¬0 0 0 0 ­ (When l ! 2)  and S C «1 1 1 1 1 » ¬ ¼ ¬0 1 1 1 1 ¼ ¬0 0 1 0 0 ¼ .2.2. Table A1 The reachability set and priority set Elements 1 2 3 4 5 Rr 1. Note that the reachability matrix generates the relationships between the father and the cat. (6) and is the father.2.3 1.3. the asterisk * indicates the derivative relation which does not emerge in the original relation matrix.4. the reachability set and the priority set are derived based on Eqs. A1 Hierarchical structure of the elements .2 1.F F M S M S  C 0» ¼ 1 ¼ 0¼ . For instance.5 5 Ar 1 1. In order to determine the levels of the elements in a hierarchical structure.4. matrix T 3 is stop.5 2. The other levels can also be determined with the same procedures (Table A1). ¬ ¼ ¬0 0 0 1 1 ¼ ¬0 0 0 0 1 ¼ ­ ½ (When l ! 3)  Rr ‰ Ar 1 2 l=2 3 4 5 38 E D C  D Level 1 l=3 Level 2 C l=2 D Level 3 l=2 S Level 4 M Level 5 F represents the inflects of l=2 represents the inflects of l=3 Note: Fig. In the limiting reachability matrix (when l=2).

³very strongly important´. . For instance. ) . . and ³equally important´ on a fuzzy five-level scale. ) . . ) . Pairwise of criterion Quality Quality Quality Price Terms Price Terms Supply Chain Support & & Relative importance (³9´ is maximum ´1´ is minimum) 9:1 8:1 7:1 6:1 5:1 4:1 3:1 2:1 1:1 1:2 1:3 1:4 1:5 1:6 1:7 1:8 1:9 Pairwise of criterion Price & Terms Supply Chain Support Technology Supply Chain Support Technology Technology 39 . 4. we can build a vendor selection evaluation model. PART II: Questionnaire According to the suggestions of experts and scholars.8 ! ( . . if you think the subjective % perception value might be 3 ! ( 2 . With your opinions. 7. namely quality.9 ! ( . . Used to represent compromises between the priorities listed above. ) . The scale of ³1~9´ is defined (see in Fig. please indicate relative importance levels in terms of pairs of the criteria listed below.7 ! ( .3.6. Please indicate your values on a TFN scale % % % % % as follows: 1 ! ( . . According to your experience of vendor selection. a vendor is defined and the level of relative importance of a criterion is chosen from five different levels.8. According to your subjective perceptions. The purpose of this questionnaire is to ask your opinions concerning vendor selection. .5. 9 respectively. 3. price & terms. ³weakly important´. % % % % 6 ! ( . ³essentially important´. ).4. The descriptions of each level are shown below: Intensity of fuzzy scale (example) 1 3 5 7 9 2. 4 ) . ). 5 ! ( .8 Definition of linguistic variables for relative weights of criteria Equally important Weakly important Essentially important Very strongly important Absolutely important Intermediate values between two adjacent judgments.6. .2.4. ) . supply chain support. 2. 3 .9 represents 1.7.AHP Questionnaire This research is about appropriate vendor evaluation. ) . 6. namely ³absolutely important´. through which its range is defined (range lies between 1~9). 3 ! ( . please answer this questionnaire based on your own opinions. 2 ! ( .8. . PART I: Description of and instructions for this questionnaire In this research. and technology. 2). 4 ! ( . ). 5. %%%%% % % %% Note that 1. this research is investigating a draft measurement for four vendor selection criteria.

high). Please answer this questionnaire according to your subjective perceptions of vendor selection. It has five different levels ± ³very bad´. The getting lowers among them and can overlap high mark. For example. This questionnaire adopts a Likert-type five-point scale.Appendix 2 Vendor selection satisfaction questionnaire This questionnaire is about your satisfaction level toward vendor selection. and that the three different scales for the satisfaction level ³good´ are (70. through which its range is defined (range lies between 0~100).100) respectively.. and ³very good´ ± on a fuzzy five-level scale. Please answer this questionnaire according to your perceptions. Low Medium High Q A .80. Give a score between 1~100 to indicate three different scales (i. ³fair´.90. you might think the three different scales for the satisfaction level ³very good´ (or correspond) are (80. low.e. Please give different cognitive levels for each of the following grades. ³good´. medium.90) respectively. ³bad´.

very good ( ) ( ) ( ) 0. 0 ) ) ) 0. there are two sub-criteria: quality performance. 0 101525 3540 50 60 65 75 85 90100 x Example dimension 1: Quality. very bad bad fair good ( ( ( ( ) ) ) ) ( ( ( ( ) ) ) ) ( ( ( ( ) 1. Under this criterion. and quality containment & VDCS feedback. 4. 2. 5 1 2 3 4 5 5.x % 1. 3. Performance Vendor 1(V1) Vendor 2(V2) Vendor 3(V3) Vendor 4(V4) Vendor 5(V5) C11 Quality very good good performance fair bad very bad C12 Quality very good containment & good fair VDCS feedback bad very bad Criterion 40 .

management skill (C1 ). the committee sets the weights of each criterion as g P . Here we give an example to compare the results with those obtained using the traditional assumption of independence among the criteria under consideration. professional knowledge (C2 ) and experience (C3). the evaluation committee sets three criteria. and the scores from the evaluators are summed up as shown in the following table: Employer Management Professional Experience Skill (C1) Knowledge (C2) (C3) P1 90 80 50 P2 50 60 90 P3 70 75 70 In addition. Three persons. are interviewed. we utilize non-additive Choquet integrals to aggregate fuzzy performance scores with weights. P1. P2 and P3. Considering the case of an employer who would like to promote a new manager for a company.Appendix 3 In this article.

_C1a ! g P .

45 ._C2 a ! 0. g P .

g P ._C3 a ! 0.3 .

C2 a ! 0. g P ._ 1 .5 .

C3 a ! g P ._C2 .

C where g(*) indicates the values of the fuzzy measure for the criteria.50a 76.._C1 .88 interactive case among criteria: g P . the evaluation results are as shown in the following table: Employer Synthetic score Fuzzy integral method Simple weight method P1 69. C3 a ! 0. simple weight method.25b P2 P3 a 68. Using the fuzzy integral with the above fuzzy measure and the traditional method (e.25 63.00 72.75 71.g. SWM).9 .

45._C1 a ! 0. g P .

g P .5. C2 a ! 0._C1 .

_C1 . Synthetic score= ´ hdg ! . C3 a ! 1 . C2 .

90  80 v 0.45  .

50. b additive case among criteria: Find the weights of each criterion through normalization. That is g P .5  50v 1 ! 69.80  50 v 0.

_ 1a ! g P .

375._ 2 a ! 0. g P .

25 C C C Synthetic score= 90 v 0.25 ! 76.375  80v 0.375 50v 0._ 3 a ! 0.25 41 .

From the above results. we find that if g P .

_x1a  g P .

the fuzzy integral method is more suitable than a traditional method (e. Thus.9 1 .g.. then P " 0 . 42 . SAW) when there are non-independent effects (either substitutive or multiplicative) among the considered criteria. we can see the difference between the two methods to identify the best alternative by ranking using the synthetic scores. this can increase the overall score of an alternative if the criteria are enhanced simultaneously. In other words. This implies that the relations among criteria have a multiplicative effect. Based on the above results._x2 a ! 0.