Available online at www.sciencedirect.

com

European Journal of Operational Research 189 (2008) 971–986

www.elsevier.com/locate/ejor

A hybrid neuro-fuzzy analytical approach to mode choice

of global logistics management
Jiuh-Biing Sheu *

Institute of Traffic and Transportation, National Chiao Tung University, 4F 114 Chung Hsiao W. Road,
Sec. 1, Taipei 10012, Taiwan, ROC

Received 1 September 2005; accepted 1 June 2006

Available online 15 July 2007

Abstract

This paper presents a hybrid neuro-fuzzy methodology to identify appropriate global logistics (GL) operational modes
used for global supply chain management. The proposed methodological framework includes three main developmental
phases: (1) establishment of a GL strategic hierarchy, (2) formulation of GL-mode identification rules, and (3) develop-
ment of a GL-mode choice model. By integrating advanced multi-criteria decision-making (MCDM) techniques including
fuzzy analytical hierarchy process (Fuzzy-AHP), Fuzzy-MCDM, and the technique for order preference by similarity to an
ideal solution (TOPSIS), six types of global logistics and operational modes coupled with corresponding fuzzy-based multi-
criteria decision-making rules are specified in the second phase. Using the specified fuzzy decision-making rules as the input
database, an adaptive neuro-fuzzy inference system (ANFIS) is then developed in the third phase to identify proper GL
modes for the implementation of global supply chain management. A numerical study with a questionnaire survey data-
base aimed at the information technology (IT) industries of Taiwan is conducted to illustrate the applicability of the pro-
posed method.
 2007 Elsevier B.V. All rights reserved.

Keywords: Global logistics; Global supply chain management; Multi-criteria decision making; Fuzzy inference; Neural networks

1. Introduction

Selection and characterization of proper global logistics (GL) modes is vital to effective global supply chain
management (G-SCM). It is extensively noticed that although remarkable advances in information technolo-
gies (IT) have contributed to the achievement of effective information flow in global supply chains; issues in
coordinating physical flows in transnational distribution channels, including supplied raw materials, assem-
blies, and products, still remain in G-SCM. Furthermore, under conditions of global competition, forces ori-
ented either upstream from global supply resources or downstream from global markets have made numerous
international manufacturing enterprises perceive the urgent necessity of developing competitive GL modes.

*
Tel.: +886 2 2349 4963; fax: +886 2 2349 4953.
E-mail address: [email protected]

0377-2217/$ - see front matter  2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.ejor.2006.06.082
972 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986

Despite the importance of GL mode selection and characterization in G-SCM, developing corresponding
decision support systems for complex G-SCM issues is challenging and has received little discussion in previ-
ous literature. Most prior research has been limited to either simplified numerical studies using optimization
models (Cohen and Lee, 1988; Arntzen et al., 1995; Cachon and Fisher, 1997; Nagurney et al., 2003) or explor-
ing conceptual frameworks for system evaluation (Verter and Dincer, 1995; Motwani et al., 1998; Bowersox
et al., 1999; Mollenkopf and Dapiran, 1999; Carranza et al., 2002; Morash and Lynch, 2002; Closs and Mol-
lenkopf, 2004). Some typical models are illustrated in the following for reference.
Given that the number and locations of transnational facilities are fixed, a comprehensive framework which
involves respective analytical procedures is proposed in Cohen and Lee (1988) to evaluate the corresponding
performance of alternative manufacturing and material supply strategies in supply chains. Furthermore, a bi-
objective optimization model is introduced in Arntzen et al. (1995) to minimize a weighted combination of
total operational costs and activity days in a given global supply chain network. Specifically, Cachon and
Fisher (1997) proposed simple inventory management rules and tested them with actual demand data for
the operations of continuous replenishment (CR) strategies of a given international food corporations. Nagur-
ney et al., 2003) developed a framework to analyze the dynamics of a simplified 3-layer global supply chain,
which mainly considers the interactions among three distinct layers of decision makers, i.e., manufacturers,
retailers, and end-customers.
In addition, there are many researchers who are devoted to exploring the influential factors and strategies
for diverse international supply chain scenarios. For instance, factors of fluctuations in exchange rates and
government policies are discussed in Kogut (1985), followed by a more detailed analysis by Carter and Vickery
(1988). Comparisons between the decision-making process for locating facilities based on direct labor costs
and the core competencies of the company are conducted in Bartmess and Cerny (1993). In MacCormack
et al. (1994), a four-phase decision making process is further proposed for international location decisions,
where several key factors, e.g., adequate infrastructure and managerial issues, are investigated. Furthermore,
issues of global outsourcing strategies and corresponding channel relationship management in G-SCM
are also drawing growing attentions from many researchers (Ohmae, 1989; Fagan, 1991; Monczka and
Trent, 1991; Davis, 1992; Min et al., 1994; Tagaras and Lee, 1996; Talluri, 2002; Talluri and Narasimhan,
2003).
Clearly, developing sophisticated decision support systems in aid of selection and characterization of GL
modes for G-SCM needs more research effort. As can be found in our literature review, the optimization pro-
gramming methods remain used as a major solution technique to address related issues; however certain lim-
itations may still exist. For instance, The published models may not be suitable for addressing the issues of
uncertainties and complexities of GL strategic planning resulting from certain qualitative influence factors,
e.g., transnational resource availability and foreign regulations. Furthermore, most of the published decision
making approaches, e.g., analytic hierarchy process (AHP), appear incapable of dealing with the imprecise
and vague comparisons of qualitative criteria.
Accordingly, this study presents a methodological framework, which serves not only to characterize
GL modes but also to select appropriate GL operational modes for the use G-SCM of high-technology indus-
tries. The proposed methodology is established by integrating three major techniques, including (1) fuzzy
multi-criteria decision-making (Fuzzy-MCDM), (2) fuzzy inference theories, and (3) neural networks. Relative
to the existing GL strategic planning methods, such a process of embedding multiple techniques in a compre-
hensive framework can be quite sophisticated, but it is inevitably needed due to the complexities and uncer-
tainties of global logistics and operational environments, as mentioned above. Furthermore, to our best
knowledge, the utilization of such fuzzy-network techniques in GL strategic planning and management is
rarely found in related areas. These may help to clarify the incremental contribution of this paper to the early
literature.

2. Architecture of the proposed methodology

The corresponding architecture of the proposed methodology is composed mainly of three developmental
phases, including (1) establishment of a GL strategic hierarchy, (2) formulation of GL-mode identification
rules, and (3) development of a GL mode-choice model. They are detailed in the following.
 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986 973

2.1. Global logistics strategic hierarchy

In this phase, the GL hierarchical structure is specified, in which the number of layers embedded in the pro-
posed structure and corresponding criteria that characterize the distinctive features of layers are defined. Then,
the proposed GL strategic hierarchy is finalized through a questionnaire survey.
First, we analyzed existing GL strategies and operational models, then summarizing six typical types of GL
modes used as the GL-mode choice candidates. Based on the G-SCM philosophy in terms of integration and
sharing of transnational resources, six typical types of GL modes coded GL-modes A to F, illustrated in Fig. 1,

GL-mode A GL-mode D
1(a) 1(d)

GL-mode B GL-mode E
1(b) 1(e)

GL-mode C GL-mode F
1(c) 1(f)

Fig. 1. Illustration of specified six types of global logistics modes.

974 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986

are specified. Here each mode distinguishes itself from the others mainly in the degree of resource sharing and
integration with foreign enterprises. The distinctive operational features of these modes, as well as correspond-
ing enterprise representatives, are illustrated below.
GL-mode A, shown in Fig. 1a, represents a typical centralized GL mode which is dominated fully by the
internal manufacturing/assembling center for global logistics management. Enterprises featuring GL-mode A
tend to control the entire process of manufacturing, including primary assembling, and complete it internally.
However, they may look for any potential in terms of availability of raw materials overseas and benefits ori-
ented from global markets, and thus source raw materials and potential customers from abroad. In general,
mode A is applicable particularly for those international enterprises which manufacture sophisticated assem-
blies or high-value products such as integrated circuit (IC) and laptop computers.
GL-mode B, as depicted in Fig. 1b, has almost the same function as GL-mode A, except for its manufac-
turing/primary assembling, which is partly outsourced abroad to either reduce production costs or meet the
variety of overseas customer demands. Nevertheless, it is worth noting that given either GL-mode A or B, the
aggregate product plan remains controlled by the domestic firm. In addition, the overseas logistics-related
activities, e.g., product inventory and transportation, are dominated by local firms in GL-mode B.
In contrast, GL-mode C can be regarded as an extension of GL-mode A since it possesses the same features
as GL-mode A in terms of internal centralization of manufacturing; however differing in the overseas logistics
distribution function. In addition, GL-mode C permits external facilities, e.g., distribution centers and ware-
houses, in response to the variety of overseas customer demands. Overall, the firms with global brands may
prefer GL-mode C to GL-mode A for the convenience of distributing finished products to local channel mem-
bers overseas via these external logistics facilities.
GL-mode D represents a synthesized mode evolving from both GL-modes B and C, as shown in Fig. 1d. In
addition to the outsourcing to foreign manufacturers/primary assembling firms, as exhibited in GL-mode B,
GL-mode D also has the same property as GL-mode C in terms of utilizing the external distribution facilities
to coordinate the logistics operations with overseas distribution channels. Correspondingly, GL-mode D must
rely highly on the competence of foreign contracted logistics partners to implement GL strategies. Neverthe-
less, it is worth noting that GL-mode D may benefit those multinational enterprises by its mechanisms in
transnational resource integration and sharing for managing global supply chains.
GL-Mode E refers to a specific GL mode suitable particularly for the operations of postponement strate-
gies. As illustrated in Fig. 1e, almost-finished products can be manufactured internally, and then transported
overseas for final processing which is not undertaken until the corresponding local orders are placed. As such,
risks due to either the price appreciation of corresponding primary components or overproduction can be
alleviated.
In contrast with all the other modes, GL-mode F shown in Fig. 1f represents a relatively sophisticated GL
model because it covers almost all the value-added transnational logistics activities, including overseas assem-
bling, final processing, inventory and distribution. In reality, such a mode has been greatly advanced in global
logistical management, especially in terms of utilizing external sources for both final processing and
assembling.
After the aforementioned GL-mode specification procedure, a 3-layer GL strategic hierarchy is proposed,
where the top layer, termed the GL-mode alternative layer, includes the specified six GL-mode candidates; the
second layer, termed the criteria layer, defines six respective criteria domains, including: (1) management con-
trol, (2) core competitiveness, (3) trans-organizational coordination, (4) marketing and service, (5) resource
availability, and (6) environmental variability; and the third layer involves the corresponding factors associ-
ated with these criteria.
Then, the questionnaire survey aiming at international manufacturers of high-technology industries was
executed to finalize the proposed GL hierarchical structure, including determination of the corresponding
components embedded in layers 2 and 3. At this stage, we designed the contents of the questionnaire on
the basis of the proposed GL hierarchic framework, and randomly sampled international high-technology
manufacturers of Taiwan as well as researchers/professors in related areas. Here, our questionnaire aimed
to request each survey respondent to identify the primary criteria and the corresponding influence factors con-
sidered in determining GL strategies. Aggregated from the survey data, those items agreed by more than 80%
of the sampled respondents were then selected as the basis of the proposed 3-layer GL hierarchic framework.
 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986 975

Accordingly, a 3-layer GL hierarchic structure, as illustrated in Table 1, was finalized through the afore-
mentioned survey data collection and test procedures, where the corresponding components associated with
each layer and their relationships with the components of the corresponding upper layers are also summarized.

2.2. Global logistics identification rules

Our attempt in this phase is to develop fuzzy-based GL-mode identification rules used to characterize and
assess GL modes with respect to their distinctive operational attributes. Here the identification of GL modes is
formulated as a multi-criteria decision-making (MCDM) problem solved by the integration of Fuzzy-AHP,
Fuzzy-MCDM, and TOPSIS techniques, respective fuzzy-based GL-mode identification rules. To achieve
the aforementioned purpose, three developmental procedures are proposed in this phase: (1) estimation of
pair-wise comparison matrices, (2) specification of fuzzy-weighted criteria, and (3) development of fuzzy logic
rules.
First, two respective pair-wise comparison matrices associated with layers 2 and 3 of the specified 3-layer
GL strategic hierarchy are estimated using the Fuzzy-AHP technique (Laarhoven and Pedrycz, 1983; Buckley,
1984; Lootsma, 1997). Here each element of a given pair-wise comparison matrix indicates the relative impor-
tance between a given pair of components associated with a given GL layer. In the process of matrix gener-
ation, the components of a given GL layer (k) are ranked in linear order, and associated with specific ordinal
numbers based on survey respondents’ judgments concerning the relative importance of these components
measured on a 0-1 scale. Then, an element ekij of the pair-wise comparison matrix is given by ekij ¼ rki =rkj , where
rki and rkj represent the ordinal numbers associated with components i and j of layer k, respectively. Accord-
ingly, we have a 6 · 6 pair-wise comparison matrix associated with layer 2 (M2) and a 21 · 21 comparison
matrix associated with layer 3 (M3) of the proposed GL strategic.
Next, the fuzzy weights associated with the components in a given layer are approximated. Here, we employ
a simple method to facilitate the approximation of the fuzzy weights. Given an Nk · Nk pair-wise comparison

Table 1
Proposed GL strategic hierarchy and components
Layer-1: GL-mode alternative Layer-2: Criteria Layer-3: Related influential factors
Six GL-mode candidates 1. Management control 1. Raw-material/assemblies purchasing process
(GL-modes A to F) 2. Inventory management
3. Outbound logistics distribution and transportation

2. Core competitiveness 4. Research and development (R&D)

5. Production scheduling
6. Production capacity
7. Ability to respond to changes of customer demands

3. Trans-organizational coordination 8. Transnational strategic alliance

9. Communications with IT technologies

4. Marketing and service 10. Marketing channels

11. Customer service
12. Market segmentation

5. Overseas resource availability 13. Global raw-material/assemblies supply networks

14. Grouping with overseas industries
15. Availability of overseas human resources
16. Availability of overseas financial resources
17. Availability of overseas technology support
6. Environmental variability 18. Foreign government policies
19. Foreign exchange
20. Modernization of infrastructure
21. Language and culture
976 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986

matrix associated with a given layer k (Mk), we then have the fuzzy weight associated with component i of
layer k (fik Þ:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Q 
Nk k
Nk j¼1 eij
fik ¼  rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Q ffi : ð1Þ
PN k Nk k
n¼1 N k j¼1 enj

Then, fuzzy-weighted criteria are generated. For this, we specify five linguistic terms, including ‘‘very high’’,
‘‘high’’, ‘‘medium’’, ‘‘low’’, and ‘‘very low’’, indicating five qualitative degrees to describe the subjective impor-
tance associated with each component of a GL hierarchic layer. These qualitative criteria were then mapped
into specific fuzzy membership functions to obtain raw fuzzy criteria via fuzzy-and-defuzzy transformation.
Using the fuzzy weights obtained previously, the fuzzy-weighted criteria associated with the components of
GL hierarchic layers were computed.
In addition, mapping the specified five qualitative criteria into specific fuzzy membership functions, refer-
ring to the process of fuzzy-and-defuzzy transformation, is a critical step. In this study, we associate the afore-
mentioned five linguistic criteria with five specific fuzzy membership functions, respectively, including two
trapezoidal and three triangular fuzzy membership functions, as shown in Fig. 2. Note that the parameters
of these fuzzy membership functions can be approximated on the basis of survey respondents’ viewpoints
through the proposed adaptive network-based fuzzy inference system (ANFIS).
After specifying fuzzy membership functions, the process of defuzzy transformation is then conducted.
Here, we employ the right-and-left scoring method, which is also suggested in Chen and Hwang (1992) for
its high efficiency in the process of quantifying linguistic variables. According to the fundamentals of the
right-and-left score approach, the defuzzy value (lT(Ur)) associated with a specific fuzzy membership Ur of
a given linguistic criterion r, termed the total score of the given fuzzy membership function, is computed by:
½lR ðU r Þ þ 1  lL ðU r Þ
lT ðU r Þ ¼ ; ð2Þ
2
where lR(Ur) and lL(Ur) represent the right and left score functions, given by:
lR ðU r Þ ¼ sup½U r ðxÞ ^ lmax ðxÞ; ð3Þ
x

lL ðU r Þ ¼ sup½lmin ðxÞ ^ U r ðxÞ: ð4Þ

x

Herein, lmax(x) and lmin(x) are defined as the sets of maximization and minimization with respect to x, and
their notations are given by:

x; 0 6 x 6 1;
lmax ðxÞ ¼ ð5Þ
0; otherwise;

1  x; 0 6 x 6 1;
lmin ðxÞ ¼ ð6Þ
0; otherwise:
Through the aforementioned fuzzy-and-defuzzy transformation process, the quantitative criterion associated
with each GL component can then be determined. For instance, given GL-mode h, the fuzzy-weighted crite-

very low low medium high very high

membership grades

0 a b a b a b a b 1.0
scale of subjective importance

Fig. 2. Fuzzy membership functions for qualitative criteria scale of subjective importance.
 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986 977

rion associated with a specific component i of layer k in the GL-strategy hierarchic framework (W ki ðhÞÞ is esti-
mated by
8 9
>
< k
>
=
w i ðhÞ
W ki ðhÞ ¼ fik  qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P ffi ; ð7Þ
>
: >
6
wki ðhÞ2 ;
½ h¼1

where  ki ðhÞ
w means the averaged raw fuzzy criterion associated with the component i of layer k of GL-mode h,
and its mathematical form is expressed as
P5
½l ðU r Þ  nU r ji;k;h
 ki ðhÞ ¼ r¼1 T
w ; ð8Þ
S
where nU r ji;k;h represents the number of samples that rate the component i of layer k of GL-mode h with the
qualitative criterion r, nU r ji;k;h is determined according to the data collected from a questionnaire survey to the
high-technology enterprises; and S is the valid sample size of respondents in the survey.
The next step is to develop fuzzy logic rules using both TOPSIS and estimated fuzzy-weighted criteria.
Given several alternatives considered in the multi-criteria decision-making process, the basic concept of TOP-
SIS is that the most preferred alternative should not only have the shortest distance from the ideal solution,
but also have the longest distance from the anti-ideal solution (Chen and Hwang, 1992; Opricovic and Tzeng,
2004). Here, the specified six GL modes are regarded as the alternatives considered in the proposed GL stra-
tegic hierarchy. Accordingly, employing TOPSIS and the fuzzy-weighted criteria (W ki ðhÞÞ estimated in the pre-
vious procedure, we develop respective fuzzy logic rules to both identify these specified GL modes with their
distinctive features and assess these modes by comparing the relative significance of their corresponding fuzzy-
weighted criteria. The corresponding developmental steps involved in this procedure are summarized below.
Step 1. The corresponding ideal and anti-ideal solution sets (i.e., Zk and Zk) associated with all the com-
ponents of each given layer k in the GL strategic hierarchy should be determined by
Zk ¼ fzki j i ¼ 1; 2; 3; . . . ; N k g; ð9Þ
k
Z ¼ fzki j i ¼ 1; 2; 3; . . . ; N k g; ð10Þ
where zki and zki represent the ideal and anti-ideal solutions associated with a given component i of layer k,
respectively, as given by
 
zki ¼ max W ki ðhÞji 2 I 0 _ min W ki ðhÞji 2 I 00 ; ð11Þ
h h
 
zki ¼ min W ki ðhÞji 2 I 0 _ max W ki ðhÞji 2 I 00 : ð12Þ
h h

Here I 0 and I00 represent the benefit-oriented and cost-oriented criteria groups, respectively.
Step 2. Calculate the Euclidean distance-based separations of each given GL-mode alternative (h) from
both the ideal and anti-ideal solutions in the proposed GL strategic hierarchy. For each GL mode alternative
(h), there are two groups of separation measures defined: (1) the layer-based aggregate separations, and (2) the
criterion-based disaggregate separations.
The layer-based aggregate separations (i.e., Y k ðhÞ and Yk(h)) refer to the aggregate separations of a given
GL mode (h) from both the ideal and anti-ideal solutions in a given layer k of the proposed GL strategic hier-
archy. Here Y k ðhÞ and Yk(h) are given by respectively by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X 2
Y k ðhÞ ¼ ½W ki ðhÞ  zki  ; for k ¼ 2; 3; ð13Þ
8i2I k
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
k
Y ðhÞ ¼ ½W ki ðhÞ  zki 2 ; for k ¼ 2; 3; ð14Þ
8i2I k

where Ik represents the set of components associated with a given layer k in the proposed GL strategic
hierarchy.
978 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986

In contrast, the criterion-based disaggregate separations (i.e., y ki ðhÞ and y ki ðhÞÞ represent the respective sep-
arations of a given GL mode (h) from both the ideal and anti-ideal solutions with respect to a given criterion i
of a given layer k in the proposed GL strategic hierarchy. Similarly, y ki ðhÞ and y ki ðhÞ are given by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
k kþ1 2
y i ðhÞ ¼ ½W kþ1j ðhÞ  zj  ; for k ¼ 2; ð15Þ
8j2I kþ1
i
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X
kþ1 2
y ki ðhÞ ¼ ½W kþ1j ðhÞ  zj  ; for k ¼ 2; ð16Þ
8j2I kþ1
i

where I kþ1
i represents the set of the corresponding components of criterion i, which are embedded in the lower
layer k + 1 of the proposed GL strategic hierarchy.
Step 3. Calculate both layer-based and criterion-based significance indexes (i.e., Pk(h) and pki ðhÞ, respec-
tively) associated with each given GL-mode alternative (h) by
Y k ðhÞ
P k ðhÞ ¼ ; for k ¼ 2; 3; ð17Þ
Y k ðhÞ þ Y k ðhÞ
y ki ðhÞ
pki ðhÞ ¼ k ; for k ¼ 2: ð18Þ
y i ðhÞ þ y ki ðhÞ
Step 4. Rank the GL-mode significance order for GL-mode characterization and assessment. Now, we can
compare these GL-mode alternatives using the corresponding estimated significance indexes. Using the prin-
ciples of TOPSIS (e.g., Opricovic and Tzeng, 2004), we further rank the corresponding GL-mode significance
order by the following proposed rules.
Given any pair of GL-mode alternatives, h* and h, for the comparison with respect to the relative signifi-
cance of a given layer k, GL-mode h* can be perceived relatively significant than GL-modeh if the following
two conditions hold.
Condition-1 : P k ðh Þ > P k ðhÞ; ð19Þ
Condition-2 : ðY k ðh Þ < Y k ðhÞÞ ^ ðY k ðh Þ > Y k ðhÞÞ; or ð20Þ
 k k 
Y ðhÞY ðh Þ
ðY k ðh Þ > Y k ðhÞÞ ^ ðY k ðh Þ > Y k ðhÞÞ ^ Y k ðh Þ < : ð21Þ
Y k ðhÞ
Otherwise, these two given GL-mode alternatives are ranked in the same order.
Similarly, given any pair of GL-mode alternatives h* and h for the comparison with respect to the relative
significance of a given criterion i, GL-mode h* can be perceived as relatively significant than GL-mode h if the
following two conditions hold:
Condition-1 : pki ðh Þ > pki ðhÞ; ð22Þ
Condition-2 : ðy ki ðh Þ < y ki ðhÞÞ ^ ðy ki ðh Þ > y ki ðhÞÞ; or ð23Þ

!
y ki ðhÞy ki ðh Þ
ðy ki ðh Þ > y ki ðhÞÞ ^ ðy ki ðh Þ > y ki ðhÞÞ ^ y ki ðh Þ < : ð24Þ
y k ðhÞ

Otherwise, these two given GL-mode alternatives are ranked with the same order in the given criterion-based
assessment.
Step 5. Formulate fuzzy logic rules for GL-mode identification. For each given GL-mode h, the respective
fuzzy-based GL-mode identification rule is given by
wki ð~
IF ½max U r ð wki ðhÞÞ; 8r > ½max U r ð
hÞÞ; 8r > ½max U r ð wki ðhÞÞ; 8r; 8i; k; ð25Þ


then the given GL-mode h can be identified readily by the respective linguistic criterion (r) associated with
each component of the given GL-mode h embedded in the proposed GL strategic hierarchy. Here, ~h and h

represent two neighboring GL-mode alternatives which are ranked with relatively higher and lower order, rel-
ative to the given GL-mode h, based on the previous criterion-based assessment.
 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986 979

2.3. Global logistics mode choice forecasting

This phase develops a GL-mode choice forecasting model using the technique of an adaptive neural-based
fuzzy inference system (ANFIS) which permits embedding the proposed fuzzy logic rules in an ANFIS frame-
work to efficiently find one respective GL mode most suitable for the G-SCM operations. Descriptions in
terms of the fundamentals of ANFIS techniques and related applications can be readily found elsewhere (Jang
and Gulley, 1995; Jang et al., 1997; Tsoukalas and Uhrig, 1997), and thus are omitted in this paper in consid-
eration of the space limit.
For simplicity, a 3-tier ANFIS architecture, shown in Fig. 3, is proposed, where each link in the ANFIS archi-
tecture represents the direction of information flow between a given pair of nodes. The square nodes represent
respective data-processing functions possessing adaptive parameters which need to be determined by network
training; and the circular nodes represent respective data-processing mechanisms without unknown parameters.
Considering the different layers existing in the original 3-layer GL strategic hierarchy, these nodes are further
classified into two groups, i.e., nodes of GL layers 2 and 3 of Fig. 3. The input data needed in this phase includes
(1) the subjective measures of any given targeted enterprise with respect to the components embedded in layers 2
and 3 of the original GL strategic hierarchy; and (2) the data outputted from the proposed ANFIS framework is
the forecasted GL mode that is identified as the most suitable GL mode for the targeted enterprise. In addition,
each tier of the proposed ANFIS architecture performs a certain task, as described below.

Tier-1: Given the subjective input measure (rli;k Þ associated with a given GL component (i) of GL-layer k,
obtained from a given targeted enterprise (l), the purpose of this tier is to generate the respective right
and left scores (lR ðU rl Þ and lL ðU rl ÞÞ associated with this given GL component using the pre-specified
i;k i;k
right and left score functions, i.e., lR(Ur) and lL(Ur) shown in Eqs. (3) and (4), respectively. Note that
because there are unknown adaptive parameters involved in this tier, e.g., the shape-related parameters of
those fuzzy membership functions, the corresponding nodes defined in this tier are represented by square
nodes.
Tier 2: This tier involves respective circular nodes, which calculate the corresponding total score lT ðU rl Þ
i;k
associated with each subjective measure rli;k , where the total score function shown in Eq. (2) is employed.
Tier 3: This is the output tier with only one circular node which forecasts the corresponding GL-mode
choice index (Ul) associated with the given targeted enterprise (l) by
XX
Ul ¼ fik  lT ðU rl Þ: ð26Þ
i;k
8k 8i

Then, using the estimated Ul coupled with the identification criteria shown in the proposed fuzzy-based GL-
mode identification rules, the most suitable GL mode (hl) for the given targeted enterprise (l) is determined
if the following condition holds.
( " # )
XX
k k
min Ul  fi  max U r ðwi ðhl ÞÞ; 8r ; 8h : ð27Þ
8k 8i

Tier--1 Tier-2 Tier-3

GL layer-2
input:
subjective
measures

output:
mode-choice
index

GL layer-3

Fig. 3. Proposed ANFIS architecture for GL-mode choice.

980 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986

3. Numerical examples

To demonstrate the applicability of the proposed method, particularly for GL mode-choice forecasting, a
numerical study using the aforementioned questionnaire survey which aimed at the high-technology manufac-
turing industries of Taiwan was conducted. The survey was distributed randomly to 376 high-technology
enterprises, and a total of 87 valid samples were obtained in this survey. The data gathered in the survey were
used as the input for the proposed method. According to the proposed three developmental phases, the results
output from the established fuzzy-based GL-mode identification rules and corresponding mode choice fore-
casting model are then discussed.
To generate the input data needed in the proposed method, survey respondents were asked to assess the
importance degree associated with these criteria and factors on a 0–1 measurement scale, followed by the cor-
responding Cronbach’s a tests to examine the reliability of the collected survey data. Furthermore, to approx-
imate the unknown shape-related parameters of specified fuzzy membership functions, these survey
respondents were also asked to quantify the upper and lower bounds associated with the specified five linguis-
tic criteria (e.g., very important, and not important at all, etc.) on the same measurement scale.
To ensure data reliability, the survey data collected in the 1st stage were examined utilizing the Cronbach’s
a statistic, which has been widely used to assess the internal consistency based on the correlation between
items, e.g., questions of the questionnaire. The results of the Cronbach’s a tests associated with the compo-
nents of GL layers 2 and 3 of the finalized GL strategic hierarchy are summarized in Table 2.
The numerical results shown in Table 2 indicate the acceptability of the specified criteria and corresponding
influential factors embedded in the proposed GL strategic hierarchy. As can be seen in Table 2, all the 21
Cronbach’s a measurements associated with GL-layer 3 are greater than 0.50, implying that the components
of layer 3 are acceptable; noticeably, 15 of them are highly acceptable, with Cronbach’s a measurements
greater than 0.75. Similarly, the Cronbach’s a statistics associated with the components of GL layer 2 also
exhibit acceptability in terms of the criteria specified in GL-layer 2. Note that various thresholds for Cron-
bach’s reliability tests have been utilized in previous literature (Cronbach and Meehl, 1955; Zaichkowsky,
1985); however, the thresholds 0.50 and 0.75 are generally suggested as loose and demanding thresholds,
respectively, for determining the reliability of data, and are thus used in this study.
The next step is to generate the fuzzy-based GL-mode identification rules through three developmental pro-
cedures executed in the 2nd phase of the proposed methodology. First, using the collected survey data, we
determined the ordinal numbers in terms of the comparative importance associated with the components in
either layer 2 or 3; and then estimated the pair-wise comparison matrices associated with GL layers 2 and
3, as well as corresponding fuzzy weights (i.e., fik shown in Eq. (1)) associated with the components of layers
2 and 3. Utilizing the aggregated data measured from collected survey data together with the estimated fuzzy
weights, the fuzzy-weight criteria (i.e., W ki ðhÞ shown in Eq. (7)) associated with the components of GL layers 2
and 3 were estimated. Then, using the estimates of fuzzy-weighted criteria and TOPSIS approaches, the esti-
mated linguistic criteria (r) and corresponding significance ranking order were determined, as summarized in
Tables 3 and 4, respectively. Given a GL-layer component, a lower ranking order number shown in Table 4
indicates higher significance of the given component exhibited in a given GL mode, relative to that exhibited in
the other GL modes. Note that, as mentioned above, these estimated linguistic criteria and corresponding
ranking order can be utilized to characterize the distinctive features of these GL modes. Corresponding gen-
eralizations obtained from Tables 3 and 4 are summarized.
Obtained from Tables 3 and 4, there are three major generalizations summarized below. First, despite the
differences of operational features exhibited among these GL modes, the 1st, 3rd, 4th 5th and 7th factors, i.e.,
inbound and outbound logistics management, research and development (R&D), production scheduling, and
ability to respond to changes of customer demands, remain vital to implementation of high-technology G-
SCM. Second, GL-mode A and GL-mode F can be regarded as two extremely different operational modes
for GL strategic planning and operations. According to their corresponding results shown in Tables 3 and
4, GL-mode A tends to execute the centralization-oriented GL strategies, and thus, the criterion of manage-
ment control, and corresponding influential factors, as well as the R&D factor, are highly appropriate for GL
strategic planning and management. In contrast, GL-mode F tends to implement G-SCM through decentral-
ization-oriented GL strategies. Accordingly, factors relating to transnational channel coordination, overseas
 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986 981

Table 2
Summary of Cronbach’s a test results
Component Cronbach’s a value Result
GL layer 2
1. Management control 0.71 Acceptable
2. Core competitiveness 0.92 Highly acceptable
3. Trans-organizational coordination 0.72 Acceptable
4. Marketing and service 0.67 Acceptable
5. Overseas resource availability 0.88 Highly acceptable
6. Environmental variability 0.89 Highly acceptable
GL layer 3
1. Raw-material/assemblies purchasing process 0.65 Acceptable
2. Inventory management 0.58 Acceptable
3. Outbound logistics distribution and transportation 0.77 Highly acceptable
4. Research and development (R&D) 0.93 Highly acceptable
5. Production scheduling 0.86 Highly acceptable
6. Production capacity 0.87 Highly acceptable
7. Ability to respond to changes of customer demands 0.89 Highly acceptable
8. Transnational strategic alliance 0.82 Highly acceptable
9. Communications with IT technologies 0.64 Acceptable
10. Marketing channels 0.72 Acceptable
11. Customer service 0.45 Acceptable
12. Market segmentation 0.64 Acceptable
13. Global raw-material/assemblies supply networks 0.88 Highly acceptable
14. Grouping with overseas industries 0.81 Highly acceptable
15. Availability of overseas human resources 0.91 Highly acceptable
16. Availability of overseas financial resources 0.86 Highly acceptable
17. Availability of overseas technology support 0.85 Highly acceptable
18. Foreign government policies 0.90 Highly acceptable
19. Foreign exchange 0.82 Highly acceptable
20. Modernization of infrastructure 0.86 Highly acceptable
21. Language and culture 0.81 Highly acceptable

resource availability, and external environmental variability may highly influence the system performance of
GL-mode F. Third, according to the corresponding G-SCM network, GL-mode E may tend to execute post-
ponement strategies, and consequently the corresponding system performance may depend highly on the cri-
terion of core competitiveness, and corresponding factors, e.g., production scheduling, R&D, and ability of
responding to customers demands are distinctive features.
The following test scenario aims to demonstrate the capability of the proposed GL-mode choice forecasting
model. To accomplish this goal, two stages are involved: (1) network training and (2) forecast demonstration.
In the first stage, the developed fuzzy-based GL-mode identification rules are embedded in the proposed
ANFIS framework, and trained with collected survey data. Note that for convenience, the aforementioned
network training tasks can be readily conducted using existing commercial neural networks packages, e.g.,
NeuroShell 2 and MATLAB produced by Ward Systems Groups, Inc. and Math Works, Inc., respectively,
and thus are not detailed in this text. Herein, the aforementioned survey data collected from 87 samples were
used for training the proposed ANFIS framework through 5000, 6000, and 7000 runs of training, respectively.
The corresponding network training prerequisites, mainly including the rules for training termination and ini-
tial weight setting, as well as the resulting outputs are summarized in Table 5. Then, the respective thresholds
with respect to the mode-choice index associated with the six GL mode candidates are determined.
The second stage is to test the validity of the trained GL-mode choice model in GL-mode forecasting. Here,
a total of 20 Taiwanese high-technology manufacturing enterprises (T-1 to T-20) excluding the existing 87
valid samples were sampled. Here the corresponding survey respondents (e.g., managers of enterprises) were
asked to linguistically measure the importance of the pre-specified GL criteria and corresponding influential
factors with the pre-specified five linguistic terms (e.g., VERY LOW to VERY HIGH, etc.). In addition, infor-
mation regarding the existing GL operational modes implemented by these targeted enterprises was also
982 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986

Table 3
Estimates of linguistic criteria for characterization of GL strategic modes
Components (GL-mode identification criteria) Linguistic criteria
Mode-A Mode-B Mode-C Mode-D Mode-E Mode-F
Layer 2
1. Management control VH VH VH VH H H
2. Core Competitiveness VH VH VH H VH H
3. Trans-organizational coordination M H H H VH VH
4. Marketing and service H M VH VH H VH
5. Overseas resource availability M H H H VH VH
6. Environmental variability M M H H H VH
Components (influential factors)
Layer 3
1. Raw-material/assemblies purchasing process VH VH VH VH H H
2. Inventory management H H VH VH M M
3. Outbound logistics distribution and transportation VH VH VH VH VH H
4. Research and development (R&D) VH VH VH H VH H
5. Production scheduling H VH H VH VH VH
6. Production capacity VH H H H H H
7. Ability to respond to changes of customer demands H H VH VH VH VH
8. Transnational strategic alliance M H H H VH VH
9. Communications with IT technologies M H H VH H VH
10. Marketing channels H H VH VH VH VH
11. Customer service H M VH VH H H
12. Market segmentation H H H H VH VH
13. Global raw-material/assemblies supply networks H H H H H VH
14. Grouping with overseas industries M H H H VH VH
15. Availability of overseas human resources M H H H VH VH
16. Availability of overseas financial resources M H H H H H
17. Availability of overseas technology support H M M H H VH
18. Foreign government policies M M H H H H
19. Foreign exchange M M VH VH H H
20. Modernization of infrastructure M M H H H VH
21. Language and culture H VH M H VH VH
VH: very high (importance degree).
H: high (importance degree).
M: medium (importance degree).

obtained through the survey. The aforementioned collected survey data were then input to the trained GL-
mode choice forecasting model to forecast the respective GL-mode choice index (i.e., Ul shown in Eq.
(26)). The forecast output of the proposed model and the corresponding GL operational modes executed
by these targeted enterprises are summarized in Table 6. Findings and discussions are summarized in the
following.
The numerical results of Table 6 imply the applicability of the proposed model used as a decision-making
tool for GL-mode choice. As can be seen in Table 6, out of the sampled 20 Taiwanese high-technology enter-
prises, 14 (70%) are identified for using the GL modes consistent with those suggested by the proposed model.
The remaining six sampled enterprises (termed T-3, T-9, T-12, T-14, T-18, and T-20 shown in Table 6) pres-
ently execute GL modes differing from those predicted. Through our diagnosis of GL-mode choice, enterprises
including T-9, T-14, T-18, and T-20 tended to agree with the modes we suggested after our second-round inter-
view survey. For instance, among these, enterprises T-9, T-14, and T-18 highly accepted our suggestion that
GL-mode F should be more suitable than E for their G-SCM operational cases, particularly considering the
increasing difficulties and costs in gaining internal resources. In contrast, the other firms (T-3 and T-12)
declined our suggestion to replace GL-mode A with B. According to the resulting second-round interview sur-
vey, their two major concerns are: (1) the risks of product quality control in case of manufacturing outsourc-
ing overseas, and (2) the difficulty of communications and practice coordination with transnational
 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986 983

Table 4
Relative significance ranking order of GL modes
Components (GL-mode identification criteria) Significance ranking order
Mode-A Mode-B Mode-C Mode-D Mode-E Mode-F
Layer 2
1. Management control 1 2 4 3 5 6
2. Core competitiveness 2 3 5 4 1 6
3. Trans-organizational coordination 6 5 4 3 2 1
4. Marketing and service 5 6 1 2 4 3
5. Overseas resource availability 6 5 4 3 2 1
6. Environmental variability 6 5 4 3 2 1
Components (influential factors)
Layer 3
1. Raw-material/assemblies purchasing process 1 2 3 4 5 6
2. Inventory management 3 4 2 1 5 6
3. Outbound logistics distribution and transportation 2 1 3 4 5 6
4. Research and development (R&D) 1 3 4 5 2 6
5. Production scheduling 2 4 6 3 1 5
6. Production capacity 1 2 3 5 4 6
7. Ability to respond to changes of customer demands 6 5 4 3 2 1
8. Transnational strategic alliance 6 5 4 3 2 1
9. Communications with IT technologies 6 5 4 2 3 1
10. Marketing channels 5 6 1 2 4 3
11. Customer service 5 6 2 1 4 3
12. Market segmentation 5 6 3 4 2 1
13. Global raw-material/assemblies supply networks 6 5 4 3 2 1
14. Grouping with overseas industries 6 5 4 3 2 1
15. Availability of overseas human resources 6 5 4 3 1 2
16. Availability of overseas financial resources 6 5 3 4 2 1
17. Availability of overseas technology support 4 6 5 2 3 1
18. Foreign government policies 6 5 4 3 2 1
19. Foreign exchange 6 5 1 2 4 3
20. Modernization of infrastructure 5 6 3 2 4 1
21. Language and culture 6 3 4 5 2 1

Table 5
Network training perquisites and outputs
Training prerequisites
Sample size 87
Termination rules Average training error < 0.00005
Maximum training error < 0.0005
Training iterations: 5000 runs
Initial weight 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Output (total error) 0.75 0.667 0.625 0.625 0.583 0.5 0.375 0.333 0.333
Training iterations: 6000 runs
Initial weight 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Output (total error) 0.333 0.458 0.5 0.417 0.417 0.042 0.375 0.333 0.333
Training iterations: 7000 runs
Initial weight 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Output (total error) 0.417 0.208 0.458 0.375 0.208 0.042 0.375 0.333 0.333

organizations which did not share common language and organizational culture. Nevertheless, 18 out of our
20 suggestions were accepted in this test scenario, implying the applicability of the proposed method for prac-
tical uses.
984 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986

Table 6
GL-mode choice forecast results
Target GL mode (presently GL mode (suggested by the proposed Accepted/rejected (by the GL mode
sample used) model) enterprise) (finalized)
T-1 A A Accepted A
T-2 A A Accepted A
T-3 A B Rejected A
T-4 C C Accepted C
T-5 C C Accepted C
T-6 C C Accepted C
T-7 E E Accepted E
T-8 E E Accepted E
T-9 E F Accepted F
T-10 F F Accepted F
T-11 F F Accepted F
T-12 A B Rejected A
T-13 B B Accepted B
T-14 E F Accepted F
T-15 A A Accepted A
T-16 C C Accepted C
T-17 E E Accepted E
T-18 E F Accepted F
T-19 A A Accepted A
T-20 C D Accepted D

4. Concluding remarks

This paper has presented a hybrid neuro-fuzzy approach, which integrates Fuzzy-MCDM, TOPSIS, and
ANFIS techniques to develop a decision support system used for analyzing and determining GL operational
modes in the global supply chain environment. The architecture of the proposed methodology involves three
main developmental phases, including (1) establishment of GL strategic hierarchy, (2) formulation of GL-
mode identification rules, and (3) development of GL mode-choice model. After finalizing the proposed
GL strategic hierarchy, which considers six typical types of GL operational modes as GL-mode choice can-
didates, the fuzzy-based GL-mode identification rules are specified using Fuzzy-AHP, Fuzzy-MCDM, and
TOPSIS techniques. Then, a GL-mode choice forecasting model is developed employing the proposed ANFIS
framework. Moreover, data collected from questionnaire surveys are used in the study case to demonstrate
feasibility of the proposed method for real applications.
The corresponding analytical results have revealed the distinctive features of these GL operational modes
and corresponding ranking order to explore the relative significance of these operational features among these
GL modes. Particularly, factors including inbound and outbound logistics management, research and devel-
opment (R&D), production scheduling, and ability to respond to changes of customer demands, appear to be
generally important, no matter which GL mode is chosen for high-technology G-SCM. In addition, according
to the corresponding analytical results presented in Tables 3 and 4, GL-modes A, C, E and F appear to have
their own critical criteria significantly influencing system operations and performance. For instance, GL-mode
A can be more suitable for those centralization-oriented high-technology enterprises, and thus, it may highly
depend on the criterion of management control and corresponding influential factors. In contrast, GL-mode F
may be more appropriate for the implementation of decentralization-oriented GL strategies. Therefore, fac-
tors such as transnational channel coordination, overseas resource availability, and external environmental
variability appear to have relatively significant effects on the system performance of GL-mode F. In addition,
the applicability of the proposed GL-mode choice forecasting model has been successfully demonstrated in the
other test scenario using a different survey database.
In addition, the numerical results of Table 6 have implied the potential applicability of the proposed model
used as a decision-making tool for GL-mode choice. Among 20 sampled enterprises, 18 (90%) agree with the
GL-modes we suggested using the proposed method.
 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986 985

Despite the advantages of the proposed decision support system in characterizing and assessing such com-
plicated GL strategic hierarchy for GL mode choice, it should be clarified that there still are various unknown
factors, either supply-oriented or demand-oriented, that may significantly influence practical operations of
global logistics and global supply chain management. Particularly, from a practical point of view, such sophis-
ticated decision making maneuvers may also rely highly on marketing and financial objectives of the enter-
prises, as well as their organizational culture and decision-making mechanisms. This also explains our
reasoning for limiting the proposed method to the extent of decision support tools.
Nevertheless, the methodology proposed in this study is expected to stimulate more research in the related
fields of global supply chain management and corresponding strategic planning. In addition, we hope that this
study may help developing logic rules and analytical skills for practical use in addressing issues regarding the
uncertainty and complexity of global supply chain management of international high-technology industries.
Extension and modification of the proposed model for other industries and operational cases may also war-
rant more research. Further effort in training the proposed neuro-fuzzy based model with more valid data is
also needed for practical applications.

Acknowledgements

This research was supported by the grant NSC 94-2416-H-009-019 from the National Science Council of
Taiwan. We would like to thank the referees for their constructive comments.

References

Arntzen, B.C., Brown, G.G., Harrison, T.P., Trafton, L.A., 1995. Global supply chain management at Digital Equipment Corporation.
Interfaces 25 (1), 69–93.
Bartmess, A., Cerny, K., 1993. Building competitive advantage through a global network of capabilities. California Management Review,
78–103.
Bowersox, D.J., Closs, D.J., Stank, T.P., 1999. 21st Century Logistics: Making Supply Chain Integration a Reality. Council of Logistics
Management, Oak Brook, IL.
Buckley, J.J., 1984. The multiple judge, multiple criteria ranking problem: A fuzzy set approach. Fuzzy Sets and Systems 13 (1), 25–
37.
Cachon, H., Fisher, M., 1997. Campbell Soup’s continuous replenishment program: Evaluation and enhanced inventory decision rules.
Production and Operations Management 6 (3), 266–276.
Carranza, O., Maltz, A., Antun, J.P., 2002. Linking logistics to strategy in Argentina. International Journal of Physical Distribution and
Logistics Management 32 (6), 480–496.
Carter, J.R., Vickery, S.K., 1988. Managing volatile exchange rates in international purchasing. Journal of Purchasing and Materials
Management 24 (4), 13–20.
Chen, S.J., Hwang, C.L., 1992. Fuzzy Multiple Attribute Decision Making Methods and Applications. Springer-Verlag, New York.
Closs, D.J., Mollenkopf, D.A., 2004. A global supply chain framework. Industrial Marketing Management 33, 37–44.
Cohen, M.A., Lee, H.L., 1988. Strategic analysis of integrated production–distribution systems: Models and methods. Operations
Research 36 (2), 216–228.
Cronbach, L.J., Meehl, P.E., 1955. Construct validity in psychological tests. Psychological Bulletin 52, 281–302.
Davis, E.W., 1992. Global outsourcing: Have US managers thrown the baby out with the bath water?. Business Horizons (July–August)
58–65.
Fagan, M.L., 1991. A guide to global sourcing. The Journal of Business Strategy (March–April), 21–25.
Jang, J.-S., Gulley, N., 1995. Fuzzy Logic Toolbox for Use with MATLAB. The Math-Works, Inc., Natick, MA.
Jang, J.-S., Sun, C.-T., Mizutani, E., 1997. Neuro-Fuzzy and Soft Computing. Prentice-Hall International, Inc., Upper Saddle River, NJ.
Kogut, B., 1985. Designing global logistics: Profiting from operational flexibility. Sloan Management Review, Fall, 27–38.
Laarhoven, P.J.M., Pedrycz, W., 1983. A fuzzy extension of Saaty’s priority theory. Fuzzy Sets and Systems 11, 229–241.
Lootsma, F., 1997. Fuzzy Logic for Planning and Decision-Making. Kluwer, Dordrecht.
MacCormack, A.D., Newman III, L.J., Rosenfield, D.B., 1994. The new dynamics of global manufacturing site location. Sloan
Management Review, Summer, 69–80.
Min, H., LaTour, M., Williams, A., 1994. Positioning against foreign supply sources in an international purchasing environment.
Industrial Marketing Management 23, 371–382.
Mollenkopf, D.A., Dapiran, G.P., 1999. World Class Logistics: How Well Do Australian/New Zealand Firms Perform? Council of
Logistics Management Annual Conference, Toronto, Canada.
Monczka, R.M., Trent, R.J., 1991. Global sourcing: A development approach. International Journal of Purchasing and Materials
Management (Spring), 2–8.
986 J.-B. Sheu / European Journal of Operational Research 189 (2008) 971–986

Morash, E.A., Lynch, D.F., 2002. Public policy and global supply chain capacities and performance: A resource-based view. Journal of
International Marketing 10 (1), 25–51.
Motwani, J., Larson, L., Ahuja, S., 1998. Managing a global supply chain partnership. Logistics Information Management 11 (6), 349–
353.
Nagurney, A., Cruz, J., Matsypura, D., 2003. Dynamics of global supply chain supernetworks. Mathematical and Computer Modelling
37, 963–983.
Ohmae, K., 1989. The global logic of strategic alliances. Harvard Business Review (March–April), 143–154.
Opricovic, S., Tzeng, G.-H., 2004. Compromise solution by MCDM methods: A comparative analysis of VIKOR and TOPSIS. European
Journal of Operational Research 156, 445–455.
Tagaras, G., Lee, H.L., 1996. Economic models for vendor valuation with quality cost analysis. Management Science 42, 1531–1543.
Talluri, S., 2002. A buyer–seller game model for selection and negotiation of purchasing bids. European Journal of Operational Research
143, 171–180.
Talluri, S., Narasimhan, N., 2003. Vendor evaluation with performance variability: A max–min approach. European Journal of
Operational Research 146, 543–552.
Tsoukalas, L.H., Uhrig, R.E., 1997. Fuzzy and Neural Approaches in Engineering. John Wiley and Sons Inc., New York, NY.
Verter, V., Dincer, M.C., 1995. Global manufacturing strategy. In: Drezner, Z. (Ed.), Facility Location: A Survey of Applications and
Methods. Springer-Verlag, New York, pp. 263–282.
Zaichkowsky, J.L., 1985. Measuring the involvement construct. Journal of Consumer Research 12 (3), 341–352.